<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Gordon Silvera on Medium]]></title>
        <description><![CDATA[Stories by Gordon Silvera on Medium]]></description>
        <link>https://medium.com/@thedatastrategist?source=rss-182bce5c9ee8------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*VxIMOEYO9DYxDxov5IjL-A.jpeg</url>
            <title>Stories by Gordon Silvera on Medium</title>
            <link>https://medium.com/@thedatastrategist?source=rss-182bce5c9ee8------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Fri, 20 Nov 2026 20:42:50 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@thedatastrategist/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[Bridging the Gap: Techniques for Data Analysts to Influence Decision-Making]]></title>
            <link>https://thedatastrategist.medium.com/bridging-the-gap-techniques-for-data-analysts-to-influence-decision-making-c4661fbd8230?source=rss-182bce5c9ee8------2</link>
            <guid isPermaLink="false">https://medium.com/p/c4661fbd8230</guid>
            <category><![CDATA[data-analysis]]></category>
            <category><![CDATA[communication]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[stakeholder-management]]></category>
            <dc:creator><![CDATA[Gordon Silvera]]></dc:creator>
            <pubDate>Fri, 15 Dec 2023 14:16:31 GMT</pubDate>
            <atom:updated>2023-12-15T14:16:31.719Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GrmmI2dwle-zP6Pxo9f8Mg.png" /><figcaption>DALL-E: surrealistic vector illustration of data scientists building a bridge</figcaption></figure><blockquote>“Nearly half (45%) of <strong>data experts say they struggle to interpret domain experts’ data questions</strong> or needs, and more than a third (36%) surveyed believe <strong>domain experts do not understand what can and can’t be done with data</strong>.”</blockquote><blockquote>— Sigma Computing</blockquote><p>One way analytics professionals can differentiate themselves is by <em>effectively</em> communicating to executives. Many data professionals have little interest in this — applying far greater focus on the journey taken to produce their work than the presentation of the result.</p><p>Regardless of your level of interest, data professionals need to become effective communicators. Effective communication and presentation is the best way for your work to have an <strong>impact</strong> on your organization.</p><p>Sigma Computing found that a third of data professionals feel unable to convey the value of their work.</p><blockquote><em>“</em>After all the time and resources that go into preparing these reports, data experts feel their work is under-appreciated by company leadership, with <strong>34% sharing they’ve felt unable to convey the value of an analysis</strong> to an executive or key decision maker<em>.”</em></blockquote><p>For this reason, we’re going to discuss ways to approach analytical conversations when you’re in presentations or boardrooms. Specifically, we’ll discuss how to communicate:</p><ol><li>Certainty</li><li>Consistency</li><li>Insights &amp; Recommendations</li><li>Severity</li></ol><p><strong>TL;dr.</strong> Effectively communicating analytical topics requires a balance of (a) technical depth and thoroughness and (b) clarity and simplicity in your message. You are not in the room to prove your intelligence, you’re there to translate the “thesis” of your work and its relevance to your organization.</p><h3>1. Communicating Certainty</h3><p>Statistics allow us to assess the magnitude of an impact or relationship and our certainty that it is real. We measure this certainty with two key measures: <strong>p-values and confidence intervals</strong>. When discussing these in a meeting, it is important to provide a simple explanation of how to interpret them.</p><ul><li><strong><em>P-values</em></strong> indicate the likelihood that our findings (i.e. that reject the null hypothesis) are not due to randomness in user behaviors. For instance, a p-value of 0.05 in an A/B test implies that we are 95% sure that the test variant outperforms the control group.</li><li><strong><em>Confidence intervals</em></strong> represent the range in which we are X% certain that the actual value falls. As our tests consider only a sample of users, there is a level of uncertainty in our predictions. By using confidence intervals, we can be precise about this level of uncertainty. For example, if a confidence interval is 95%, with a lower bound of 2 and an upper bound of 4, we can be 95% certain that the true value lies between 2 and 4.</li></ul><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F9jW9G8MO4PQ%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D9jW9G8MO4PQ&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F9jW9G8MO4PQ%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="640" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/cd0ab38fe4c4e0d6a6c087933ced832b/href">https://medium.com/media/cd0ab38fe4c4e0d6a6c087933ced832b/href</a></iframe><p>We can also describe our <strong>confidence in the data itself</strong>. Identifying data quality is an important step in enterprise analytics. The following dimensions should be considered when classifying the quality of underlying data.</p><ul><li><strong><em>Data availability</em></strong>. We have the data required to complete the analysis (ie no missing columns or datasets).</li><li><strong><em>Data completeness</em></strong>. There’s no missing data (ie missing rows) in the dataset. If there are, we have discussed how to address the missing values with the business stakeholders.</li><li><strong><em>Data triangulation</em></strong>. When our datasets are aggregated (for instance, by calculating total users and revenue), high-level metrics align with our team’s key business reports.</li><li><strong><em>Data semantics</em></strong>. Metrics and dimensions are defined in the data in the same way that business domain experts discuss them.</li><li><strong><em>Data bias</em></strong>. We are aware of outliers and potential biases in our data. Here are two excellent articles on data bias (<a href="https://towardsdatascience.com/survey-d4f168791e57">one</a> and <a href="https://towardsdatascience.com/types-of-biases-in-data-cafc4f2634fb">two</a>).</li></ul><h3>2. Communicating Consistency</h3><p>We evolve our understanding of a business through iterative evidence generation, a process that bears resemblance to the Scientific Method. Though the following terms aren’t traditionally used within the scientific or data community, we can them to precisely communicate the consistency of the findings we’ve observed.</p><p><strong><em>Early indication</em></strong>. Use this caveat when discussing a finding for the first time with your audience.</p><ul><li>“The survey found that highly engaged business users have lower CSAT than the average. Some feedback points <strong>indicate</strong> that this is due to challenges with integrations. We will conduct deeper research to understand the drivers of this group’s lower CSAT.”</li></ul><p><strong><em>Trend</em></strong>. Link multiple findings across time or categories (e.g. marketing channel, product line).</p><ul><li>“The redesign of the new product pages has performed well. As you’ll see in the chart below the redesign <strong>outperforms across all</strong> product categories .”</li><li>“Our ‘buy-one-get-one’ offers <strong>have continued to</strong> outperform our ‘10% discount’ offers for the past 3 months. If we continue to see this outperformance, we’ll move a larger share of our promotional offers to BOGOs.”</li></ul><p><strong><em>Contrarian finding</em></strong>. Call out findings that counter-trends we’ve seen previously, or where a subgroup’s behavior differs from the overarching customer base.</p><ul><li>“We see high levels of price elasticity in all product categories <strong>except</strong> for furniture and other large ticket items.”</li><li>“This is <strong>the first year</strong> our holiday promotion didn’t drive incremental profits. We believe that the current economic climate has impacted consumer spending habits.”</li></ul><p><strong><em>Hypothesis</em></strong>. These are beliefs that you or other colleagues have about a business topic. The findings that you collect during your research will either support or refute this hypothesis.</p><ul><li>“Same-store sales are down 4% year-over-year. We believe that the current economic climate has impacted consumer spending habits. However, we need to conduct further analysis to understand the impact of market pressures on the business.”</li></ul><p><strong><em>Theory</em></strong>. These are collectively held beliefs about a particular topic in the organization. These beliefs are supported by insights and trends over time. That does not mean, however, that theories cannot change based on shifts in the competitive and economic climate.</p><ul><li>“After conversations with the Revenue and Internationalization teams, we collectively believe that we should add Spanish natively into the product and focus our expansion efforts in Latin America. This is for several reasons. (1) Our existing LATAM user base is among the most engaged regions, (2) adding Spanish will have the greatest potential impact on our User base than any other language, (3) regulatory and operational hurdles are less significant in LATAM than other target markets.”</li></ul><h3>3. Communicating Insights &amp; Recommendations</h3><p>Analysts should be careful when giving recommendations, especially as a consultant.</p><ul><li>Firstly, the analyst may not have the complete business context and, therefore, may provide a misinformed insight.</li><li>Secondly, the audience may (even subconsciously) disregard your recommendation if it counters their preconceived hypotheses — even if it’s a great insight.</li></ul><p>For this reason, analysts (whether internal or external) should incrementally build to a recommendation using an iterative, collaborative process between the researcher and their stakeholders.</p><ol><li>Analysts <strong>present “early findings” to stakeholders</strong> to determine (a) whether the data that underlies the potential insight makes sense and (b) what business context can be added to produce value-added, relevant insights.</li><li>Analysts <strong>receive feedback from stakeholders</strong> and adjust or deepen the existing insight as needed.</li><li>Analysts <strong>present the finalized insight</strong>, which combines a visual of the underlying data point(s) <em>and</em> the business context.</li><li>Analysts and stakeholders <strong>jointly review related insights to determine what actions</strong> <strong>should be taken</strong> based on the new findings. This recommendation can take the format — “because {<em>insights</em>} indicate that {<em>reason for business trend</em>}, we recommend {<em>business action</em>}.</li></ol><h3>4. Communicating Severity</h3><p>It’s challenging being the bearer of bad news. We can use the following phrases to discuss findings that may have a significant impact on the business. The conversations have been tailored to the startup, ClassPass.**</p><p><strong><em>Monitor</em>.</strong> Watching a particular metric or trend to see whether it continues or becomes stronger. This term suggests that the stakeholder should not take action on a finding, but may consider doing so in the future. Words like monitor communicate, “We’re aware of what’s going on and we’ll keep you informed.” Very reassuring for executives.</p><ul><li>“We will continue to <strong>monitor</strong> the decline in the share of luxury classes, which has led to a 6% YoY decline in Average Monthly Revenue per Active. The sustained decline may require a review of our pricing relative to our competitor set.”</li></ul><p><strong><em>Consider.</em></strong> This is a “soft” recommendation to take action, based on a finding or set of business knowledge. There are several reasons we may want to use soft language. For instance, (a) it may influence the strategic, operational, or political state of the business, or (b) because the insights that support the recommendation have not been supported in a greater trend.</p><ul><li>“Our average price for running classes within the Chicago and Los Angeles markets exceeds our competitor set by 20%. We have seen preliminary indications that sales within this product segment have stagnated MoM due to this price disparity. <strong>Consider</strong> reducing the price to 10% above the competitive set.”</li></ul><p><strong><em>Suggest</em>.</strong> “Suggest” has a moderate tone when used within board rooms. Executives will most likely perk up and listen to what follows, but you’re not risking your job based on a suggestion gone awry. That said, “suggestions” should be supported by multiple data points and collaboratively developed with strategic stakeholders. IMHO, data scientists are often not equipped with the necessary information to give business recommendations to an executive. However, insights combined with wisdom gleaned from executives results in a powerful perspective.</p><ul><li>We’ve seen a substantial increase in boxing interest — with a 220% year-over-year growth in boxing class signups. We spoke to the Partnerships team and <strong>suggested</strong> that expanding boxing studio partnerships should be a focus this year.</li></ul><p><strong>Recommend.</strong> The term <em>recommendation</em> has been a commonly used term throughout the history of business analytics and data science. Recommendations can be described as “suggestions for strategic change in an organization, based on information generated through analytics.” Due to the historical weight of the term — “we recommend” — building to a recommendation takes more collaboration and iteration than suggestions (as discussed above).</p><ul><li>“The new variant of the user flow increased conversions by 12% versus our current design. We saw statistically significant gains across all of our primary user segments. Therefore, we <strong>recommend</strong> moving forward with a full release of the new design.</li></ul><p><strong>Declare.</strong> Simply stating “we are doing {action}” is the most unequivocal delivery of a recommendation. This should only be used when</p><ul><li>The finding or recommendation is an obvious and non-controversial decision.</li><li>All relevant decision-makers are in favor of the declared action.</li></ul><p>For example, “The Pink creative consistently has 10% — 20% higher click-through rates than the Purple creative. We <strong>will</strong> continue with the Pink creative until we see signs of creative fatigue.”</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2Far1oQPk__AI%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3Dar1oQPk__AI&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2Far1oQPk__AI%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/d361a9bbe33c5fb66e691982183867e0/href">https://medium.com/media/d361a9bbe33c5fb66e691982183867e0/href</a></iframe><h3>Conclusion</h3><p>In conclusion, communicating insights can be a challenging endeavor, even for an experienced data scientist. The words we use to communicate have a greater weight than we often realize. By <strong>remaining aware of the words we use during meetings and using the “communication conventions” above</strong>, analysts can have effective conversations with executives (or any strategic stakeholders) when presenting their work.</p><p><em>** References to specific companies are for hypothetical purposes only. We do not have access to these companies’ data nor do we have insight into their business, aside from the publicly available information.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c4661fbd8230" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Maximizing Business Impact: A Strategic Approach to Investing in Analytics]]></title>
            <link>https://medium.com/swlh/maximizing-business-impact-a-strategic-approach-to-investing-in-analytics-814f914198f1?source=rss-182bce5c9ee8------2</link>
            <guid isPermaLink="false">https://medium.com/p/814f914198f1</guid>
            <category><![CDATA[strategic-investments]]></category>
            <category><![CDATA[modern-data-stack]]></category>
            <category><![CDATA[startup]]></category>
            <category><![CDATA[analytics]]></category>
            <dc:creator><![CDATA[Gordon Silvera]]></dc:creator>
            <pubDate>Wed, 27 Sep 2023 09:31:08 GMT</pubDate>
            <atom:updated>2023-09-27T09:31:08.473Z</atom:updated>
            <content:encoded><![CDATA[<h3>Maximizing Business Impact: A Strategic Approach to Data Analytics</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aKTmtMbveqspqMO2-Q2ozw.png" /><figcaption>DALLE-2: Surrealistic vector illustration of a strategic approach to finding value in data.</figcaption></figure><p>In today’s data-driven world, using data analytics strategically can be a game-changer for startups. While growth or revenue benefits may not be immediately evident, investing in data can provide invaluable insights and pave the way for future success.</p><h3>Why Invest In Data?</h3><h4>Challenges in Data Investment</h4><p>Investing in “data” can be challenging for founders because there is no direct return on growth or revenue (at least in the short term). It’s difficult to get reliable data, metrics, and dashboards. So teams tend to quickly implement a solution and worry about analytics once they or their investors start getting curious about business performance.</p><p>Finding data talent can be difficult and costly. Great data professionals are in demand, costly, and hard to find. And most data professionals either have technical depth or strong business acumen — rarely both. Given that startups need T-shaped employees, it would be ideal to have data people with technical and strategic competency. This leads to <a href="https://www.sigmacomputing.com/blog/breaking-down-the-data-language-barrier">shortfalls in understanding between business stakeholders and data professionals</a>.</p><blockquote>“85% of data experts struggle to interpret business stakeholders’ data needs” — Sigma Computing</blockquote><h4>The Consequences of Delaying Data Strategy</h4><p>The problem with delaying a data strategy is that — until you create a data strategy — information will always be misaligned in your business. When your team communicates metrics and insights, they will have a different underlying understanding of that information, which, in turn, leads to an inconsistent view of the business.</p><h4>A Strategic Approach to Data Investment</h4><p>Startups can invest in data in incremental, low-cost ways. And this approach will provide compounding gains as your business and data scale. To understand this, we should break “data investment” into three categories:</p><ul><li><strong>Strategic Investment</strong>. Time spent thinking about your business’ data conceptually, and building frameworks to organize and align your data. Examples of this include: creating OKRs; creating event, data, and metric catalogs; creating a semantic data layer; and discussing data security and governance.</li><li><strong>Technical Investment.</strong> Investment in analytical infrastructure. Technical investment is comprised of these groups. <em>Cloud Services</em>: AWS, GCP, Azure. <em>Web Analytics</em>: Google Analytics, Google Tag Manager, Amplitude, Mixpanel. <em>Modern Data Stack SaaS</em>: Fivetran, dbt, Tableau, Looker, Mode, Hex. <em>Data Sources</em>: Nielsen, Acxiom, Bloomberg, Facebook Graph, Feedly.</li><li><strong>Human Capital Investment.</strong> Hiring data professionals on a part or full-time basis. Common roles include <em>Engineers</em> (Backend, Data, Analytical, or ML), <em>Analysts</em> (e.g. Business Analysts, Data Scientists), and <em>Leadership</em> (e.g. Analytics Manager, VP of Data Science). This can also include <em>Consultants</em> — part-time hires or agencies with subject matter expertise.</li></ul><p>These categories are ordered from least to most expensive. The more time we spend on strategic investment, the less costly and more effective technical and human capital investments will be.</p><p>In the sections below, we will discuss <strong>three low-cost ways startups can invest in data strategy</strong>.</p><ol><li>Think early, grow incrementally</li><li>Get event tracking right</li><li>Create a semantic layer</li></ol><h3>1. Think Early, Grow Incrementally</h3><h4>Think Early</h4><p>There is little cost in thinking about your data. However, thinking about — and, more precisely, strategically defining and organizing — your data will have massive gains down the line. And the earlier you do this, the greater the benefit. When your data is conceptually aligned at its core, it will also aligned in how it’s stored, how it’s queried, and how it’s communicated across the organization. This is <a href="https://thedecisionlab.com/reference-guide/philosophy/system-1-and-system-2-thinking">System 2</a> thinking about your business data.</p><h4>Grow Incrementally</h4><p>Data can get messy and complex fast. To prevent this from happening, we should take small, incremental steps when building analytical capabilities and assets in our organization. What are examples of analytical capabilities and assets?</p><ul><li><strong>Data Assets</strong>: datasets, data and ML pipelines, metrics, dashboards, self-service tools, and presentations</li><li><strong>Data Capabilities</strong>: building a customer segment catalog, building a machine-learning model to predict customer churn, and sending customers personalized promotions.</li></ul><p>These capabilities would be difficult to build without first making <strong>strategic</strong> and <strong>foundational technical investments</strong>.</p><p><strong>Strategic investments</strong>, such as</p><ul><li>Defining customer segments and their underlying metrics</li><li>Data trust in the Marketing team</li></ul><p><strong>Foundational technical investments</strong>, such as</p><ul><li>Semantically structured data with high quality and consistency</li><li>Robust but flexible data pipelines</li></ul><p>Notice these “capability stepping stones” are <strong>strategic investments or foundational data assets</strong>. We’ve already discussed strategic investments. However, foundational data assets are the company’s most frequently and broadly used data assets. They are either downstream data sources that have many dependencies or important datasets, data pipelines, or dashboards that are used broadly across the organization.</p><h3>2. Get Event Tracking Right</h3><blockquote>Garbage in, garbage out. And when your initial source is garbage, then you’re in deep sh*t.</blockquote><p><a href="https://amplitude.com/blog/measurement-metrics?utm_source=google-ads&amp;utm_medium=cpc&amp;utm_campaign=Dynamic_Search_EMEA_Features-Capabilities_Growth&amp;utm_content=146903321472&amp;utm_term=&amp;gad=1&amp;gclid=Cj0KCQjw06-oBhC6ARIsAGuzdw1bBG16qCXSZ6Ezz-FvIH07MyyOU15Gm5OVHG_lMYR-XH4wdrBmB9EaAqa4EALw_wcB"><strong>Instrumentation</strong></a> is the process of tracking events in your application. This is our company data’s point of inception and all the data our company creates and uses will be based on our instrumentation. Therefore it is critical we track our events accurately.</p><p>There are several tools you can use to track events, but the <a href="https://blog.logrocket.com/product-management/product-instrumentation-guide/">processes you use to track events</a> are more important than the tools. Specifically, you should use these</p><ol><li>Have an event-tracking workshop</li><li>Create an event catalog</li></ol><h3>Event Tracking Workshop</h3><p>Some cringe when they see the word “workshop.” You can call it a meeting or run it digitally. Regardless, we should gather key stakeholders from Finance, Product, Marketing, Engineering, and Analytics to identify the events we want to track. During this process, we should consider all the ways we’ll use our data — these may include:</p><ul><li>Setting OKRs</li><li>Creating quarterly reports</li><li>Creating KPIs for each department</li><li>Tracking product performance</li><li>Providing content to customer support</li></ul><h3>Creating an Event Catalog</h3><p>At the end of our Event Tracking Workshop, we should have a comprehensive list of the events we want to track in our application. We will document this in an <strong>Event Catalog</strong>, which will provide a single view of our data across the organization. Our Event Catalog should contain the following information for each event:</p><ul><li>Event Name</li><li>Event Description</li><li>Event Trigger</li><li>Event Pages or Screens</li><li>Entity (e.g. users, products)</li><li>Metrics</li><li>Data Source</li></ul><p>Below is a <a href="https://docs.google.com/spreadsheets/d/1gYhr5mbuC8eVdisYN590msnk5iPd-J_Q7Me88npdFdE/edit?usp=sharing">sample events catalog</a> for an e-commerce business. Completing these two steps will set your team up for instrumentation success, regardless of the tool you use.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*hDji8lQYh4KWVMRXPavzdg.png" /><figcaption>A sample events catalog in Google Sheets.</figcaption></figure><p>Below is a <a href="https://docs.google.com/spreadsheets/d/1gYhr5mbuC8eVdisYN590msnk5iPd-J_Q7Me88npdFdE/edit?usp=sharing">sample events catalog</a> for an e-commerce business. Completing these two steps will set your team up for instrumentation success, regardless of the tool you use.</p><p>A sample events catalog</p><h3>3. Create a Semantic Layer</h3><p>A <a href="https://airbyte.com/blog/the-rise-of-the-semantic-layer-metrics-on-the-fly">semantic data layer</a> is a structured and standardized way of organizing and representing data within an organization to ensure that data is consistently interpreted and understood by both humans and computer systems.</p><p>Creating a complete semantic data layer requires considerable investment on both the technical and human capital sides. But there as several strategic investments we can make that will have lasting impacts on how we understand and discuss our business. And meeting and thinking time are the only costs.</p><p>Here are three strategic steps we can take to build a semantic layer.</p><ol><li>Identify objectives and use cases</li><li>Create a logical data model</li><li>Define business terminology</li></ol><h4>Identifying Objectives &amp; Use Cases</h4><p><strong>Identifying Objectives. </strong>What business goals can be supported by data, and how can data support them? By clearly articulating this, we’ll have a unified view of <em>why</em> we’re doing <em>what</em> we’re doing with our data.</p><p><strong>Identifying User Cases. </strong>We should identify how and why each employee will use data. That way we can build data products to best serve them. Think of this as a Product Manager would customer journeys.</p><h4>Create a Logical Data Model</h4><p>A <a href="https://www.gooddata.com/blog/how-build-logical-data-models-scale-analytical-applications">logical data model</a> is a high-level representation of an organization’s data, focusing on its structure, organization, relationships, and business rules. It provides a conceptual view of data that is independent of any specific technology, database management system, or physical data storage. Building a logical data model is a crucial step for startups to ensure that their data is well-structured and aligned with business objectives.</p><p>Some of the key steps to creating a logical data model include:</p><ul><li>Identifying Data Entities</li><li>Determining Entity Attributes</li><li>Identifying Entity Relationships</li></ul><p>Creating a conceptual map of your data will help technical stakeholders as they transform and analyze your data.</p><h4>Define Business Terminology</h4><p>Giving specific business-oriented names and definitions to your data assets is a critical step in connecting your data team to the business. In particular, we should define:</p><ul><li>Events</li><li>Entities</li><li>Metrics</li><li>Segments</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eyYVIBbX_6e6zSGQ2NhE4w.png" /></figure><h3>Conclusion</h3><p>We started by asking, “Why invest in analytics?” There are clear risks and challenges. However — by making early, strategic investments in analytics — we can reap short and long-term business gains with almost no cost (only meeting and thinking time).</p><p>Three ways we can make low-cost, strategic investments in analytics are:</p><ol><li>Thinking early, growing incrementally. Define and organize conceptual data maps early in the business’ existence. Then build foundational data assets and incrementally produce value-added outputs from that foundation.</li><li>Getting event tracking right. Event tracking is our data’s point of inception. If it’s inaccurate, all other business information will be incorrect. Track and document your events carefully.</li><li>Creating a semantic layer. The semantic layer translates your data assets into a business comprehensible format.</li></ol><p>These steps will provide compounding gains as your data team scales. In our next article, we’ll discuss when and how startups should invest in analytics. Stay tuned!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=814f914198f1" width="1" height="1" alt=""><hr><p><a href="https://medium.com/swlh/maximizing-business-impact-a-strategic-approach-to-investing-in-analytics-814f914198f1">Maximizing Business Impact: A Strategic Approach to Investing in Analytics</a> was originally published in <a href="https://medium.com/swlh">The Startup</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Jupyter Notebook to Start (And Finish) Data Science Projects Faster]]></title>
            <link>https://thedatastrategist.medium.com/a-jupyter-notebook-to-start-and-finish-data-science-projects-faster-e88730aca853?source=rss-182bce5c9ee8------2</link>
            <guid isPermaLink="false">https://medium.com/p/e88730aca853</guid>
            <category><![CDATA[lean-analytics]]></category>
            <category><![CDATA[google-colab]]></category>
            <category><![CDATA[jupyter-notebook]]></category>
            <category><![CDATA[google-big-query]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Gordon Silvera]]></dc:creator>
            <pubDate>Wed, 13 Sep 2023 15:13:03 GMT</pubDate>
            <atom:updated>2023-09-20T10:26:03.632Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>“Time-to-delivery”</strong> is one of the most important yet unsung metrics in data science. It’s tricky to track, but thinking about it conceptually can help us identify ways to (a) help data scientists work faster and (b) allow business stakeholders to get the information they need faster.</p><p>There are many ways you can reduce your analytics team’s time to insight. Today, we’ll focus on <strong>reducing the time to complete simple data science tasks</strong>.</p><p>The way we’ve approached it is by creating a <strong>Google Colab Quickstart Notebook</strong>. It provides a simple notebook setup and code snippets for common data science tasks. Specifically, we focus on:</p><ul><li>Notebook Setup</li><li>Importing Data</li><li>Profiling Data</li><li>Visualizing Data</li><li>Exporting Data</li><li>Resources</li></ul><p>We’re writing a post because <strong>this notebook has a set of modular code snippets that will help data scientists work faster</strong>.</p><p><a href="https://colab.research.google.com/drive/1qEPki-Nb_Er_9ecJwPj36v7R-5Z4VTtb?usp=sharing">Google Colaboratory</a></p><h3>Setting Up The Notebook</h3><p>Below are code snippets often used in the header of a Google Colab notebook. You’ll want to import libraries and declare global variables here as well.</p><h4>Authenticating Google Colab</h4><p>This snippet connects the Colab instance to your Google profile, giving the instance the same authentication as your personal profile.</p><pre># Provide your credentials to the runtime<br>from google.colab import auth<br>auth.authenticate_user()</pre><h4>Loading Collab Settings</h4><p>You can customize your <a href="https://colab.research.google.com/notebooks/data_table.ipynb">dataframe displays</a>, use <a href="https://colab.research.google.com/github/jdwittenauer/ipython-notebooks/blob/master/notebooks/language/IPythonMagic.ipynb">cell magic</a>, make <a href="https://colab.research.google.com/notebooks/basic_features_overview.ipynb">bash calls</a>, and more with a simple set of API calls.</p><pre># Colab includes the google.colab.data_table package that can be used <br># to display large pandas dataframes as an interactive data table.<br>%load_ext google.colab.data_table<br><br># If you want to return to classic dataframe view<br>%unload_ext google.colab.data_table</pre><h4>Installing nbextensions</h4><p>Notebook extensions, or <strong>nbextensions</strong>, are a helpful addition to Jupyter Notebooks. We can add much of its functionality to Colab notebooks as well. This is particularly useful when trying to visualize HTML or interactive elements.</p><pre>!jupyter contrib nbextension install --user</pre><h3>Loading Data</h3><p>Inputting and outputting data is one of the most common tasks for a data scientist. However, setting up the connections to do so can be time-consuming. In our Quickstart Notebook we’ve included exports to the following destinations:</p><ul><li>From BigQuery to a dataframe</li><li>From a dataframe to BigQuery</li><li>From a dataframe to Google Sheets</li><li>From Colab to a local directory</li><li>From Colab to Google Storage</li></ul><p>See the notebook, <a href="https://colab.research.google.com/notebooks/io.ipynb">External data: Local Files, Drive, Sheets, and Cloud Storage</a>, for more examples of how to ingest and send data from Google Colab.</p><h4>From BigQuery to a dataframe</h4><p>In March 2023, the BigQuery team updated the BigQuery Client API to allow users to output a query directly as a dataframe. This removes the need to use pandas-gbq (a personal favorite for many years!).</p><pre>from google.cloud import bigquery<br>import pandas as pd<br><br>client = bigquery.Client(project=GOOGLE_PROJECT_ID)<br><br>sql = &quot;&quot;&quot;<br>    SELECT name, SUM(number) as count<br>    FROM `bigquery-public-data.usa_names.usa_1910_current`<br>    GROUP BY name<br>    ORDER BY count DESC<br>    LIMIT 10<br>&quot;&quot;&quot;<br><br># Call the API<br>df_names = client.query(sql).to_dataframe()<br><br># View the first 5 rows of the dataframe<br>df_names.head()</pre><h4>From a dataframe to BigQuery</h4><p>It’s often helpful to load a dataframe back to BigQuery. For instance, we may want to record the outputs of a model back to our database. We can use the Google Cloud API to do so.</p><p>With the `bigquery.LoadJobConfig` parameter, <a href="https://gcloud.readthedocs.io/en/latest/bigquery-schema.html">schema</a>, we can give specifics about our schema, including its name, field type (e.g. STRING, FLOAT), mode (e.g. NULLABLE, REQUIRED), and description.</p><pre># Get (or create) a dataframe<br># This dataset is available as a sample dataset in Google Colab<br>df = pd.read_csv(&#39;/content/sample_data/california_housing_test.csv&#39;)<br><br># Specify table_id to the ID of the table to create.<br>table_id = &quot;my-project-id.test.california_housing_test&quot;<br><br>job_config = bigquery.LoadJobConfig(<br>    # Specify a (partial) schema. All columns are always written to the<br>    # table. The schema is used to assist in data type definitions.<br>    schema=[<br>        # Specify the type of columns whose type cannot be auto-detected.<br>        # Field definitions will be added to the BigQuery console.<br>        bigquery.SchemaField(<br>            &quot;households&quot;, <br>            bigquery.enums.SqlTypeNames.INTEGER,<br>            description=&quot;The number of households in the geographic area&quot;<br>            ),<br>        bigquery.SchemaField(<br>            &quot;median_income&quot;,<br>            bigquery.enums.SqlTypeNames.FLOAT,<br>            description=&quot;The log of the median income&quot;<br>            ),<br>    ],<br>    # Optionally, set the write disposition. BigQuery appends loaded rows<br>    # to an existing table by default, but with WRITE_TRUNCATE write<br>    # disposition it replaces the table with the loaded data.<br>    write_disposition=&quot;WRITE_TRUNCATE&quot;,<br>)<br><br># Load dataframe to BigQuery<br>job = client.load_table_from_dataframe(df, table_id, job_config=job_config)<br>job.result()  # Wait for the job to complete.<br><br># Get BigQuery table object<br>table = client.get_table(table_id)<br>print(&quot;Loaded {} rows and {} columns to {}&quot;.format(<br>    table.num_rows, len(table.schema), table_id<br>    )<br>)</pre><p>As you’ll see below, the field descriptions we’ve provided are now in the BigQuery console.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*zqN-UcJOvDkue3PfWS3ZrQ.png" /><figcaption>Field descriptions are stored in the BigQuery console via the google.cloud.bigquery API.</figcaption></figure><h4>From a dataframe to Google Sheets</h4><p>Loading data to Google Sheets can be helpful, especially within a business context. Many business stakeholders are comfortable doing analysis in Google Sheets (or Excel) but aren’t comfortable with Python.</p><p>We can easily bridge this technical gap by completing a portion of an analysis in Python and loading the relevant results in Google Sheets for the business stakeholders.</p><pre>import gspread<br>from google.auth import default<br>creds, _ = default()<br><br># Authorize gspread<br>gc = gspread.authorize(creds)</pre><p>First, we get the spreadsheet object using the gspread API. We can do this via the spreadsheet ID, URL, or name. See each example below.</p><pre># Open a Google Sheets spreadsheet by ID<br>spreadsheet_id = &#39;1VlIdq73MUBPSJdVcpZMGD1Dj-VuksY_CNH3r5DIe0cs&#39;<br>spreadsheet = gc.open_by_key(spreadsheet_id)<br><br># Open a Google Sheets spreadsheet by URL<br>spreadsheet_url = &quot;https://docs.google.com/spreadsheets/d/{}&quot;.format(spreadsheet_id)<br>spreadsheet = gc.open_by_url(https://medium.com/spreadsheet_url)<br><br># # Get the Google Sheets spreadsheet by name<br>spreadsheet = gc.open(&#39;Liquidator Lookout: product map features&#39;)</pre><p>Next, we select a specific <em>worksheet</em> (or tab) in the <em>spreadsheet</em>. We can do this via its index or name. From there, we use get_all_values() to get the data out of the worksheet and load it to a dataframe.</p><pre># Get the Worksheet by Spreadsheet tab index<br>worksheet = spreadsheet.get_worksheet(3)<br><br># Get the Worksheet by name<br># You can combine gc.open() with the line below to load the Worksheet directly.<br>worksheet = spreadsheet.worksheet(&#39;product_information&#39;)<br><br># get_all_values gives a list of rows<br>rows = worksheet.get_all_values()<br><br># Load rows (a list of lists) to a dataframe<br>df_products = pd.DataFrame(rows[1:], columns=rows[0])</pre><h4>From Colab to a local directory</h4><p>Loading data from your local dictionary is simple using google.colab.files. This generates a pop-up that allows you to browse and select the files you’d like to upload.</p><pre>from google.colab import files<br><br># Plugin to load local files<br>uploaded = files.upload()</pre><p>Once you load a file or dataset, it will be available in the content/ folder in your notebook instance (see below).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*euS4kyh7nWipzSdwuxyq6Q.png" /><figcaption>Find your loaded data in the Files tab in the /content folder.</figcaption></figure><h4>From Colab to a local directory</h4><p>We can then do the inverse — loading a dataset from Google Colab back to our local computer.</p><pre>from google.colab import files<br><br># Download the file to your local computer<br>files.download(&#39;/content/sample_data/california_housing_test.csv&#39;)</pre><h4>From Colab to Google Storage</h4><p>Google Storage is a useful tool for data scientists, as it allows us to save unstructured data (e.g. text files, images, audio files) on a cloud server. This saves space on your computer, and eaises collaboration through a centralized data source.</p><p>There are two main ways we can load data to Google Storage — with <strong>gsutil</strong> or the <strong>Google Python API</strong>. Regardless of the method, we need to connect to Google Storage and create a bucket.</p><pre># Authenticate the user.<br>from google.colab import auth<br>auth.authenticate_user()<br><br># Specify tyour Google Project ID.<br>project_id = &#39;my-google-project-id&#39;<br><br># Create a unique bucket name.<br># The bucket name must be unique across all GCS sources globally.<br># Use uuid to create a universally unique ID.<br>import uuid<br>bucket_name = &#39;colab-sample-bucket-&#39; + str(uuid.uuid1())<br>print(bucket_name)</pre><p>From there, we can run the following set of commands to upload or download files using <strong>gsutil</strong>.</p><pre># Configure gsutil to use the project we specified above by using gcloud.<br>!gcloud config set project {project_id}<br><br># Make a bucket to which we&#39;ll upload the file<br>!gsutil mb gs://{bucket_name}<br><br># Copy the file to our new bucket<br>!gsutil cp /content/sample_data/mnist_test.csv gs://{bucket_name}/<br><br># Dump the contents of our newly copied file to make sure everything worked<br>!gsutil cat gs://{bucket_name}/mnist_test.csv</pre><p>And we can download files from GCS with the following command.</p><pre># Copy file from GCS to local directory<br>!gsutil cp gs://{bucket_name}/mnist_test.csv /content/mnist_test.csv</pre><p>You can see how to load data to or from GCS in the <a href="https://colab.research.google.com/drive/1qEPki-Nb_Er_9ecJwPj36v7R-5Z4VTtb#scrollTo=vrm3nEZCV_KJ">Quickstart Notebook</a>.</p><h3>Profiling Data</h3><p>Once we get our data into our notebook, we need to get a high-level understanding of the data and account for any missing, inaccurate, or outlying values. This is where data profiling comes in.</p><p>Generally, I’m against a ‘one size fits all’ approach to EDA. However, these libraries are very useful when getting a base understanding of your data.</p><ul><li>pandas_profiling</li><li>sweetviz</li><li>AutoViz</li></ul><p>In the Quickstart Notebook, we provide simple examples of data profiling for each of these libraries, along with several resources on data profiling.</p><h4>pandas_profiling</h4><pre># Install pandas-profiling library<br>!pip install -U pandas-profiling<br><br># Install Jupyter&#39;s nbextension<br># Allows other extensions to be added to this notebook<br>!jupyter nbextension enable --py widgetsnbextension<br><br># Import libraries<br>import pandas as pd<br>from pandas_profiling import ProfileReport<br><br># Run profiling report<br>profile = ProfileReport(<br>  df_ohlc,<br>  title=&quot;SPY Data&quot;, <br>  html={&#39;style&#39; : {&#39;full_width&#39;:True}}<br>)<br><br># Load to colab file<br>profile.to_file(output_file=&quot;pandas_profiling.html&quot;)<br><br># Display the pandas_profiling output<br># You may not be able to view pandas_profiling about inline<br># Therefore, use this code to view it (or any HTML file).<br>import IPython<br>IPython.display.HTML(filename=&#39;pandas_profiling.html&#39;)</pre><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2FkRACuas9zG0&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3DkRACuas9zG0&amp;image=http%3A%2F%2Fi.ytimg.com%2Fvi%2FkRACuas9zG0%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/8e57a74e298d9ab35427d328580d9ca8/href">https://medium.com/media/8e57a74e298d9ab35427d328580d9ca8/href</a></iframe><h4>sweetviz</h4><p>Sweetviz enables data scientists and analysts to effortlessly generate comprehensive reports that provide insights into the structure, distributions, and relationships within their datasets.</p><pre># Install sweetviz library<br>!pip install sweetviz<br><br># Import sweetviz librart<br>import sweetviz as sv<br><br>my_report = sv.analyze(df_ohlc)<br>my_report.show_html(&#39;sweetviz_report.html&#39;)<br><br># Display the SweetViz output<br>import IPython<br>IPython.display.HTML(filename=&#39;sweetviz_report.html&#39;)</pre><p>You can also use Sweetviz to compare two datasets. Here, we’re comparing two stock tickers. However, this is particularly useful if you want to compare training and test datasets.</p><pre># Get stock OHLC data<br>df_aapl = web.DataReader(&#39;AAPL&#39;, &#39;stooq&#39;, start=&#39;01/01/2023&#39;, end=&#39;06/01/2023&#39;)<br>df_amzn = web.DataReader(&#39;AMZN&#39;, &#39;stooq&#39;, start=&#39;01/01/2023&#39;, end=&#39;06/01/2023&#39;)<br><br>import pandas as pd<br>import numpy as np<br>import sweetviz as sv<br><br># Specify the two datasets and the reporting name, <br># along with the target variable (optional).<br>my_report = sv.compare([df_aapl, &#39;Apple&#39;],[df_amzn, &#39;Amazon&#39;], &#39;Close&#39;)<br>my_report.show_html(filepath=&#39;sweetviz_comparison_report.html&#39;)<br><br># Test/Train Example:<br># my_report = sv.compare([df_train, &#39;Train&#39;],[df_test, &#39;Test&#39;], &#39;target_var&#39;)<br><br>import IPython<br>IPython.display.HTML(filename=&#39;/content/sweetviz_comparison_report.html&#39;)</pre><h3>Visualizing Data</h3><p>One of the main challenges with plotting in Jupyter is that customizing charts takes a LOT of code and time. Creating interactive charts is even more difficult.</p><p>We’ve focused on Pareto efficiencies here — we have included Colab’s chart suggestion feature as our go-to plotting solution. However, Colab has several good data viz notebooks from other libraries.</p><ul><li><a href="https://drive.google.com/file/d/1jky2SKKPp2AWvldoKPggLT0DLSVfG3HQ/view?usp=sharing">Data Visualization.ipynb</a> by <a href="https://www.quantuniversity.com/">QuantUniversity</a></li><li><a href="https://colab.research.google.com/github/adelnehme/intro-to-data-visualization-Python-live-training/blob/master/Introduction_to_data_visualization_in_Python_live_session_full.ipynb">Introduction to data visualization in Python live session_full.ipynb</a> by DataCamp</li></ul><h4>Colab Chart Suggestions</h4><p>As mentioned, Colab Chart Suggestions is a quick way to build charts. It uses matplotlib and Seaborn (sns) under-the-hood but abstracts many of the more complex parts of using these libraries.</p><p>The user flow for Colab Chart Suggestions is:</p><ol><li>Create a Pandas dataframe</li><li>Click the ‘Chart Suggestions’ icon</li><li>Click the chart you’d like to use</li><li>Click ‘Accept’ to create a new code block that contains a function to output the plot</li></ol><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fwww.youtube.com%2Fembed%2F5zDfN0wcyFI%3Ffeature%3Doembed&amp;display_name=YouTube&amp;url=https%3A%2F%2Fwww.youtube.com%2Fwatch%3Fv%3D5zDfN0wcyFI&amp;image=https%3A%2F%2Fi.ytimg.com%2Fvi%2F5zDfN0wcyFI%2Fhqdefault.jpg&amp;key=a19fcc184b9711e1b4764040d3dc5c07&amp;type=text%2Fhtml&amp;schema=youtube" width="854" height="480" frameborder="0" scrolling="no"><a href="https://medium.com/media/80e760a4f200fad74e83133f455e0508/href">https://medium.com/media/80e760a4f200fad74e83133f455e0508/href</a></iframe><h3>Resources (and Conclusion)</h3><p>We use the modules above to quickly start data science projects. However, we use several other templates depending on the methods we use in the analysis. Generally, using templated code will allow your analytics team to start and finish their work faster.</p><h4>On the shoulders of giants …</h4><p>In both our Quickstart Notebook and this article, we’ve referenced several other notebooks and articles. We encourage you to review them as well. We’ve included some of our favorites below.</p><p><strong>Importing &amp; Exporting Data</strong></p><ul><li><a href="https://colab.research.google.com/notebooks/io.ipynb">Google Colab | io.ipynb | External data: Local Files, Drive, Sheets, and Cloud Storage</a></li></ul><p><strong>Profiling Data</strong></p><ul><li><a href="https://colab.research.google.com/github/Tanu-N-Prabhu/Python/blob/master/Exploratory_data_Analysis.ipynb">Google Colab | Exploratory Data Analysis</a></li><li><a href="https://towardsdatascience.com/3-tools-for-fast-data-profiling-5bd4e962e482">Medium: 3 Tools for Fast Data Profiling</a></li><li><a href="https://stackoverflow.com/questions/52292456/installing-jupyter-contrib-nbextension-on-google-colab">StackOverflow | Installing jupyter-contrib-nbextension on Google Colab</a></li><li><a href="https://stackoverflow.com/questions/61283382/pandas-profiling-error-within-google-colab">StackOverflow | Pandas Profiling error within google colab</a></li><li><a href="https://www.kaggle.com/code/rishabh6377/eda-using-sweetviz-and-autoviz">Kaggle | EDA Using sweetviz and autoviz</a></li></ul><p><strong>Visualizing Data</strong></p><ul><li><a href="https://github.com/AutoViML/AutoViz">Github | AutoViz</a></li><li><a href="https://pypi.org/project/autoviz/">PyPI Docs | AutoViz</a></li><li><a href="https://altair-viz.github.io/gallery/">Altair Chart Gallery</a></li></ul><p><strong>Useful Notebooks</strong></p><ul><li><a href="https://colab.google/notebooks/">Google Colab Notebook Library</a></li><li><a href="https://colab.research.google.com/">Welcome To Colaboratory.ipynb</a></li><li><a href="https://colab.research.google.com/github/cs231n/cs231n.github.io/blob/master/python-colab.ipynb">colab-tutorial.ipynb</a>. From Stanford’s CS231 class on Python.</li><li><a href="https://colab.research.google.com/drive/1gCqFEquqNvEoTDX3SNhR2PZkXWPHKXnc?usp=sharing">Advanced Python Tutorial.ipynb</a></li><li><a href="https://colab.research.google.com/github/GoogleCloudPlatform/vertex-ai-samples/blob/main/notebooks/official/bigquery_ml/bqml-online-prediction.ipynb">bqml-online-prediction.ipynb</a>. BigQueryML + Vertex AI are used for churn prediction from Google Analytics event-level data.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e88730aca853" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Bringing Kaizen to Your Business: continuous improvement through iterative experiments]]></title>
            <description><![CDATA[<div class="medium-feed-item"><p class="medium-feed-image"><a href="https://thedatastrategist.medium.com/bringing-kaizen-to-your-business-continuous-improvement-through-iterative-experiments-e5cad85a4413?source=rss-182bce5c9ee8------2"><img src="https://cdn-images-1.medium.com/max/2432/1*St-Ubtc-xZ17Ii5MyLn1mg.png" width="2432"></a></p><p class="medium-feed-snippet">Kaizen, a Japanese term meaning &#x201C;continuous improvement,&#x201D; has become a guiding principle for businesses seeking to drive incremental&#x2026;</p><p class="medium-feed-link"><a href="https://thedatastrategist.medium.com/bringing-kaizen-to-your-business-continuous-improvement-through-iterative-experiments-e5cad85a4413?source=rss-182bce5c9ee8------2">Continue reading on Medium »</a></p></div>]]></description>
            <link>https://thedatastrategist.medium.com/bringing-kaizen-to-your-business-continuous-improvement-through-iterative-experiments-e5cad85a4413?source=rss-182bce5c9ee8------2</link>
            <guid isPermaLink="false">https://medium.com/p/e5cad85a4413</guid>
            <category><![CDATA[lean-startup]]></category>
            <category><![CDATA[business-strategy]]></category>
            <category><![CDATA[experimentation]]></category>
            <category><![CDATA[kaizen]]></category>
            <category><![CDATA[a-b-testing]]></category>
            <dc:creator><![CDATA[Gordon Silvera]]></dc:creator>
            <pubDate>Mon, 17 Jul 2023 14:52:20 GMT</pubDate>
            <atom:updated>2023-07-17T14:52:20.394Z</atom:updated>
        </item>
        <item>
            <title><![CDATA[3 Types of Information that Lead to Business-Changing Insights]]></title>
            <link>https://thedatastrategist.medium.com/3-types-of-information-that-lead-to-business-changing-insights-8642e278e2c8?source=rss-182bce5c9ee8------2</link>
            <guid isPermaLink="false">https://medium.com/p/8642e278e2c8</guid>
            <category><![CDATA[business-strategy]]></category>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[insights]]></category>
            <category><![CDATA[ecommerce]]></category>
            <dc:creator><![CDATA[Gordon Silvera]]></dc:creator>
            <pubDate>Fri, 07 Jul 2023 07:06:01 GMT</pubDate>
            <atom:updated>2023-10-16T14:32:06.670Z</atom:updated>
            <content:encoded><![CDATA[<h3>Decoding Data: The Art of Asking Questions That Transform Business</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xGuQNW-5Cq4DLfH9AP-8Vg.png" /></figure><h3>Introduction</h3><p>In the world of data analysis, the ultimate goal is to uncover insights that have the power to guide, drive, or even change the direction of a business. However, without a clear focus, an analysis can easily result in useless findings. By categorizing business questions into three distinct groups, Analysts can guide their work toward generating valuable insights.</p><p>In this article, we will explore the <strong>three types of questions that lead to business-changing insights</strong>:</p><ul><li>Known-Knowns</li><li>Known-Unknowns</li><li>Unknown-Unknowns</li></ul><h3>Known-Knowns: confirming existing knowledge</h3><p>Known-Knowns refer to business questions or information that we already understand or have a strong hypothesis about. These questions focus on confirming what we already know or believe to be true. When working with Known-Knowns, it is essential to ask:</p><ul><li>What is the key decision at stake during this analysis?</li><li>What is the most important stakeholder trying to achieve?</li></ul><p>By addressing these questions, Analysts can establish a solid foundation and ensure that the analysis aligns with the key objectives of the business.</p><p>Examples of known-knowns in an e-commerce business include:</p><ol><li><strong>Conversion Rate</strong>: What is the current conversion rate of the website, and how does it vary across different traffic sources or user segments?</li><li><strong>Average Order Value</strong>: What is the average value of orders placed by customers, and how does it differ based on factors like product category, customer location, or marketing campaigns?</li><li><strong>Customer Lifetime Value</strong>: How much revenue does the average customer generate over their entire relationship with the company, and how does it vary based on their purchase history or engagement with loyalty programs?</li><li><strong>Churn Rate</strong>: What is the rate at which customers stop making purchases or become inactive, and what are the common factors associated with customer churn?</li><li><strong>Product Performance</strong>: Which products are the best sellers, and how does their performance change over time or during specific promotions or seasons?</li><li><strong>Customer Satisfaction</strong>: What is the overall satisfaction level of customers, as measured through surveys or feedback ratings, and what are the key factors influencing customer satisfaction?</li><li><strong>Return Rate</strong>: What percentage of products purchased are returned by customers, and what are the primary reasons for returns?</li><li><strong>Traffic Sources</strong>: Which channels or marketing campaigns drive the highest volume of website traffic, and what is the quality and conversion rate of traffic from different sources?</li></ol><p>These questions represent areas where the company already has some understanding or hypotheses, and the analysis can focus on confirming, refining, or gaining deeper insights into these aspects of the business.</p><h3>Known-Unknowns: answering guiding questions</h3><p>Known-Unknowns encompass the unanswered questions within the business or related to the specific project. Unanswered business questions should be presented as “guiding questions” to the Analyst, along with enough business context to help them properly answer the question.</p><p>Guiding questions are focused on discovering, learning, or understanding something new. To uncover these insights, Analysts should ask:</p><ul><li>What are the unanswered questions within the business or related to this project?</li><li>What business hypotheses do we have that we’re uncertain about?</li></ul><p>Domain Experts should have a good understanding of their business area’s Known-Knowns and Known-Unknowns. This is even worth recording in a central document, such as Notion or Google Sheets. Addressing the Known-Unknowns provides Analysts with a roadmap for exploration and allows them to delve into areas where valuable insights may be waiting to be discovered.</p><p>Here are some potential examples of Known-Unknowns in an e-commerce business:</p><ol><li><strong>Customer Behavior</strong>: What are the key factors that drive customers to make a purchase or abandon their shopping carts? How can we optimize the user experience to increase conversion rates?</li><li><strong>Competitive Analysis</strong>: What strategies are our competitors implementing to attract customers and gain market share? How can we differentiate ourselves and stay ahead in a competitive landscape?</li><li><strong>Pricing Optimization</strong>: What is the optimal pricing strategy for our products? Are there specific pricing tiers or discount strategies that resonate better with our target audience?</li><li><strong>Customer Segmentation</strong>: How can we effectively segment our customer base to tailor marketing campaigns and promotions to specific groups? What are the characteristics and preferences of different customer segments?</li><li><strong>Product Recommendations</strong>: How can we improve our product recommendation algorithms to provide personalized and relevant recommendations to customers, thereby increasing cross-selling and upselling opportunities?</li><li><strong>Customer Lifetime Value (CLV) Prediction</strong>: Can we develop models or predictive analytics to estimate the future CLV of customers and identify high-value customers who are likely to have long-term relationships with our brand?</li><li><strong>Mobile Optimization</strong>: How can we enhance the mobile shopping experience and address any specific challenges or barriers that mobile users may face?</li><li>Social Media Influence: What is the impact of social media on our brand perception and customer engagement? How can we leverage social media platforms to drive brand awareness and customer loyalty?</li></ol><p>These examples represent areas where the company has recognized gaps in knowledge or areas where they want to explore further to gain valuable insights. By addressing these Known-Unknowns through data analysis and research, the company can uncover new opportunities for growth, optimization, and customer satisfaction.</p><h3>Unknown-Unknowns: shifting perspectives with novel insights</h3><p>Unknown-Unknowns represent the realm of novel information that the analysis brings to light. These insights have the potential to challenge existing perspectives and reshape the way business domain experts understand and make decisions.</p><p>To uncover Unknown-Unknowns, Analysts should ask:</p><ul><li>What new information can we provide that can impact the understanding or decision-making within the business?</li><li>What unexpected information do I see my analysis? Does this counter our previously established Known-Knowns?</li></ul><p>By addressing the Unknown-Unknowns, Analysts can generate groundbreaking insights that have the power to transform the business landscape.</p><p>Unknown-Unknowns represent novel information or insights that the analysis can bring to light, challenging existing perspectives or assumptions. Potential examples in an e-commerce business include:</p><ol><li><strong>Emerging Customer Segments</strong>: Through data analysis, the company may discover previously unidentified customer segments that exhibit unique behavior or preferences, providing opportunities to tailor marketing strategies or product offerings accordingly.</li><li><strong>Unforeseen Market Trends</strong>: The analysis may uncover unexpected market trends or shifts in consumer behavior that were not previously recognized, allowing the company to adapt its strategies and offerings to capitalize on emerging opportunities.</li><li><strong>Hidden Customer Needs</strong>: The analysis might reveal previously unrecognized customer needs or pain points that the company can address through new products, services, or improvements to the customer experience.</li><li><strong>Cross-Category Insights</strong>: By analyzing customer purchase patterns, the company may discover untapped opportunities for cross-selling or bundling products across different categories, leading to increased sales and customer satisfaction.</li><li><strong>Disruptive Technologies</strong>: The analysis may uncover emerging technologies or innovative solutions that could disrupt the e-commerce industry, prompting the company to explore partnerships, investments, or adoption of new technologies to stay ahead of the curve.</li><li><strong>Untapped Marketing Channels</strong>: The analysis might reveal alternative marketing channels or platforms that have the potential to reach untapped audiences or drive higher engagement and conversion rates.</li><li><strong>Unexplored International Markets</strong>: Through data analysis, the company may identify international markets with high growth potential that were previously overlooked, prompting expansion strategies into new regions or countries.</li><li><strong>Synergistic Partnerships</strong>: The analysis may reveal unexpected synergies or partnership opportunities with complementary businesses or industries, opening up new avenues for collaboration and revenue generation.</li></ol><p>These examples demonstrate how strategic analysis can uncover hidden opportunities, insights, and possibilities that were previously unknown to the company. Identifying Unknown-Unknowns often requires a combination of internal and external data (provided by the Analyst) and a deep understanding of the market (provided by the Domain Expert).</p><h3>Conclusion</h3><p>By using the framework of Known-Knowns, Known-Unknowns, and Unknown-Unknowns, Analysts can distinguish the types of insights they are generating, determine how to present them effectively and identify the appropriate stakeholders to share them with.</p><p>Understanding the different categories of questions and their corresponding insights allows Analysts to provide focused and impactful analysis that can truly drive business success. By aligning their work with the business&#39;s key objectives and exploring the unknown, Analysts become valuable partners in shaping the organization&#39;s future.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8642e278e2c8" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[4 Reasons Why You’re Still Waiting for Insights from Your Data Scientist]]></title>
            <link>https://thedatastrategist.medium.com/4-reasons-why-youre-still-waiting-for-insights-fcc66461cffc?source=rss-182bce5c9ee8------2</link>
            <guid isPermaLink="false">https://medium.com/p/fcc66461cffc</guid>
            <category><![CDATA[chief-data-officer]]></category>
            <category><![CDATA[actionable-insights]]></category>
            <category><![CDATA[business-strategy]]></category>
            <category><![CDATA[data-strategy]]></category>
            <category><![CDATA[business-analysis]]></category>
            <dc:creator><![CDATA[Gordon Silvera]]></dc:creator>
            <pubDate>Tue, 01 Nov 2022 19:27:49 GMT</pubDate>
            <atom:updated>2023-09-13T15:46:07.701Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Simple things we can do to get actionable insights faster.</strong></p><p>Have you heard conversations similar to these?</p><blockquote>“Do we have a Q2 customer acquisition projection? No, we don’t? We’ve been waiting on this for weeks. Isn’t that just something we should just … have?”</blockquote><blockquote>— A Chief Marketing Officer</blockquote><blockquote>“We still don’t have reliable results from our last A/B test. We need to confirm whether the events have been collected accurately.”</blockquote><blockquote>— A Product Manager</blockquote><blockquote>“We don’t pull the SaaS data into the database right now. We’d have to add the request to the data engineering team’s backlog. After that, we can start analyzing it.”</blockquote><blockquote>— A Business Analyst</blockquote><p>If you’ve heard comments such as these, you more not alone. A study from <a href="https://www.sigmacomputing.com/resources/data-language-barrier/">Sigma Computing</a> found that <strong>63% of employees say they can’t get insights from their data solutions</strong> in the right timeframe.</p><h3>Overview</h3><p>In this article, we identify 4 reasons why insights are delayed in organizations, as well as ways we can mitigate these issues.</p><ol><li>The project scope isn’t well defined</li><li>There are problems with the data</li><li>Insights aren’t delivered iteratively</li><li>A lack of analytical templates</li></ol><p>We can use the metric <em>time to insight</em> to track the speed of insight delivery. Though teams rarely track this metric, it’s useful to think of this metric (even conceptually) as we talk through the following issues. 55% of business domain experts find that the average time to delivery is 1–4 weeks, and 71% of domain experts have unmet expectations from their analytics team.</p><h3>1. The project scope isn’t well defined</h3><p>It’s difficult for data experts and domain experts to understand each other. The prior has spent much of their career acquiring a set of skills the latter isn’t expected to. In fact, one-third of business domain experts are not confident in their ability to articulate data questions or needs to their data teams (<a href="https://www.sigmacomputing.com/resources/data-language-barrier/">Sigma Computing</a>).</p><p>By creating a research brief, we can set expectations before any work begins and we create a lingua franca between data and domain experts.</p><p>Project briefs are one to two-page documents that provide an overview of a project. To many, this may seem like a superfluous step, especially when working on projects internally. However, I would argue that project briefs are critical for the following reasons.</p><ul><li><strong>It’s a contract</strong> between technical and non-technical stakeholders that sets expectations of what will be delivered. The analyst will always have the destination in sight, even while they dig through the data forest.</li><li>Provides a <strong>centralized, consistent reference</strong> throughout the duration of the project. Any newcomers to the project can use this document to quickly onboard. We make our project briefs living documents, providing updates in the document throughout the duration of the analysis.</li><li>It’s a <strong>“lean” mode of communication</strong>. Research briefs contain the most important information and the most poignant insights related to a particular business topic. It’s a good way to disseminate actionable insights across the business — a single page provides everything an employee needs to know about the given project.</li><li>It’s a <strong>persistent artifact</strong> that lives in the business’ collective knowledge. When you have a consistent way of documenting analysis, the learnings from that analysis can continue to impact the organization long after its creators leave.</li></ul><p>You can access <a href="https://www.thedatastrategist.com/build">sample research briefs</a> from theDataStrategist <a href="https://www.thedatastrategist.com/build">here</a>.</p><figure><img alt="A sample research brief from theDataStrategist.com." src="https://cdn-images-1.medium.com/max/1024/1*eHG1XTmSwdUCknwVnt85jA.png" /><figcaption>A sample research brief from <a href="https://docs.google.com/document/d/1TbGNT90rJx5ADNxmQ8BwwmKpbj54RBoz4RE1FWsSSLs/edit#heading=h.hfb4ydts6jf">theDataStrategist.com</a>.</figcaption></figure><h3>2. Issues with data</h3><p>The not-so-old adage says that “data scientists spend 80% of their time cleaning data.” Even if it’s not 80%, there are always issues with data.</p><p>Given the advent of the modern data stack, there are two technical approaches to addressing data issues:</p><ol><li><strong>Solve the issues upstream</strong>. There are many ELT (extract, load, transform) services that simplify pulling data from your product, database, or SaaS partners. Examples of these platforms include Fivetran, Stitch Data, and Airbyte. A major positive to using these platforms is that they produce consistent data outputs. If you’re pulling information from <a href="https://github.com/fivetran/dbt_amplitude">Amplitude</a>, <a href="https://github.com/fivetran/dbt_stripe">Stripe</a>, <a href="https://github.com/fivetran/dbt_klaviyo_source">Klaviyo</a>, Google Analytics, or other common SaaS platforms, there are likely open-source code repositories that provide all the necessary transformations your data requires.</li><li><strong>Identify the issues downstream</strong>. Even with high-quality data coming from the source, analysts should always sense-check their data before proceeding with the analysis. If there are fundamental issues with the data (e.g. events are getting tracked incorrectly), we may not be able to expedite the analysis. But the faster we identify <em>bad</em> data, the faster to can make <em>good</em> decisions in the organization.</li></ol><p>Analysts can use the following checklist to quickly QA their data before using it in an analysis:</p><ul><li><strong>Data availability</strong>. We have the data required to complete the analysis (ie no missing columns or datasets).</li><li><strong>Data completeness</strong>. There’s no missing data (ie missing rows) in the dataset. If there are, we have discussed how to address the missing values with the business stakeholders.</li><li><strong>Data triangulation</strong>. When our datasets are aggregated (for instance, by calculating total users and revenue), high-level metrics align with our team’s key business reports.</li><li><strong>Data semantics</strong>. Metrics and dimensions are defined in the data in the same way that domain experts discuss them.</li><li><strong>Data bias</strong>. We are aware of outliers and potential biases in our data. Here are two excellent articles on data bias (<a href="https://towardsdatascience.com/survey-d4f168791e57">one</a> and <a href="https://towardsdatascience.com/types-of-biases-in-data-cafc4f2634fb">two</a>).</li></ul><h3>3. Insights aren’t delivered iteratively</h3><p>Some data science teams wait until the end of their analysis to present findings. This may be weeks after the start of the analysis. And any misalignments that arise between the analyst and business stakeholders during the duration of the project may impact the project’s efficacy.</p><p>A better approach is to iteratively release results. Drip-releasing insights throughout the analysis not only engage our audience but also allows us to pivot work based on the feedback we receive.</p><p>During your next analysis, take the following approach to release insights. See whether this expedites <em>time to insight</em> and reduces unmet expectations.</p><ol><li><strong>Identify the goals of the analysis</strong> and add them to your project brief. Treat the project brief as your first deliverable in the analysis. It sets the tone and approach for the rest of your work.</li><li><strong>Create a presentation outline as soon as possible</strong>. This is similar to an outline of a story, with the headline to each slide taking us one step towards our final conclusions. Of course, we do not have any conclusions at this point, but we can populate the presentation as we receive new information and insights.</li><li><strong>Continue to update the presentation</strong>. As you move further into the analysis, your story becomes increasingly clear. Providing your clients with an unpolished view of the analysis will help them understand where you’re going. And they will be able to provide feedback that will help the analyst answer the most important business questions.</li><li><strong>Get early feedback on preliminary insights</strong>. As mentioned, drip-releasing insights on an ongoing basis allow domain experts to give continuous feedback. You can also caveat that any early insights may change (however they shouldn’t change so significantly that they change the interpretation of the results).</li><li><strong>Give updates on a daily basis</strong>. Even if you don’t have results yet, it’s good practice to provide daily updates on the project. This can be a short Slack message or email containing <em>Updates</em>, <em>Next Steps</em>, and <em>Blockers</em>.</li><li><strong>Finalize insights with domain experts</strong>. Before your final presentation, all insights and conclusions should be reviewed by our primary business stakeholders. Doing so will make them champions of your work as it’s distributed throughout the organization.</li></ol><p>Taking this approach creates an ongoing feedback loop between data and business domain experts. When you have periodic touch bases, you’ll be able to course correct more quickly whenever there are misalignments between stakeholders.</p><h3>4. Lacking templates</h3><p>Talented analysts can generally answer any business question given the right data and enough time. Yet there are things that we can do to help analysts work faster, regardless of their talent level.</p><p>Templates are one of the easiest ways to expedite analysts’ work, especially when using a method that has been used elsewhere in the organization. As a former manager once told me: “The first time we analyze something, just do it. The second time, streamline it. The third, automate it.”</p><p>Even if we don’t go as far as automating analyses, templates go a far way to streamline analysts’ work. And analysts always have the option to augment or deviate from the templates as they see fit. And as our analytics team scales, templates are a good way for individuals to contribute to the company’s data practice.</p><p>Our company can use both <em>strategic</em> and <em>technical</em> frameworks to reduce our <em>time to insight</em>.</p><p><strong>Strategic frameworks</strong> help expedite processes by creating a consistent approach to operations. Rather than reinventing an approach to doing something, the analyst can focus on optimizing the end results. Common types of strategic frameworks include:</p><ul><li><em>Metric selection frameworks</em>. These are consistent approaches to identifying metrics, such as the Google HEART metrics framework.</li><li><em>Presentation templates</em>. Consulting teams are known for using template presentation decks. Internal teams can leverage presentation templates as well — including emails, notebooks, presentation decks, and reports.</li></ul><p><strong>Technical frameworks</strong> are code repositories that give analysts or data scientists 80% of the code needed to complete a given task. Since analysts don’t have to recreate that 80% of the code (or effort), they can focus on deeper, more valuable business questions from the outset. Examples of these frameworks include:</p><ul><li><em>A/B testing</em>. Especially when a team runs several tests in a single domain (e.g. product improvements or marketing optimization), we can templatize A/B tests and common outputs related to the test (e.g. metric box plots for the test vs. control groups). In this case, we merely need to change a set of experiment parameters — such as, the success metrics, time periods, and test flags.</li><li><em>Forecasts</em>. Forecasts are generally run on a monthly basis and they use the same approach each time. This makes is an excellent candidate for templatization.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/885/1*t1UJuVpcQDMvRh7ZfzR77A.png" /><figcaption>Examples of Jupyter Notebook templates.</figcaption></figure><h3>In Conclusion</h3><p>Most companies experience bottlenecks in the insights production process. By closing the gap between data and business domain experts, we can produce more valuable insights faster. We’ve discussed the following ways to close that gap:</p><ol><li><strong>Using a project scope</strong> to define the scope of the project.</li><li><strong>Manage data issues</strong> upstream (using ELT tools) or downstream (with data sense checks).</li><li><strong>Delivered insights iteratively</strong>, getting continuous feedback from business stakeholders.</li><li><strong>Leverage frameworks</strong> that expedite often complete analytical operations.</li></ol><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fcc66461cffc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[6 Easy Steps to a Data-Driven Startup on Day 1]]></title>
            <link>https://medium.com/swlh/6-easy-steps-to-a-data-driven-startup-on-day-1-4e4f900c2667?source=rss-182bce5c9ee8------2</link>
            <guid isPermaLink="false">https://medium.com/p/4e4f900c2667</guid>
            <category><![CDATA[lean-startup]]></category>
            <category><![CDATA[business-strategy]]></category>
            <category><![CDATA[data]]></category>
            <category><![CDATA[metrics]]></category>
            <category><![CDATA[data-driven]]></category>
            <dc:creator><![CDATA[Gordon Silvera]]></dc:creator>
            <pubDate>Tue, 06 Sep 2022 13:51:07 GMT</pubDate>
            <atom:updated>2023-10-17T01:53:39.033Z</atom:updated>
            <content:encoded><![CDATA[<h3>6 Simple Steps to a Data-Driven Startup</h3><p>Entrepreneurs have asked me, “What are the fewest things I need to make my business data-driven?” After spending the past year on the ground floor of a couple of startups, I’ve gained a better sense of the quickest steps to build a data-driven business.</p><p>My greatest takeaway is that, in the early stages of a business, <strong>planning how your company will use data is more important than implementing data solutions</strong>. Only two items in our list require technical implementation — instrumentation and reporting.</p><p>In this article, we’ll identify six actions (plus a bonus) you can take to pave a path towards a data-driven business.</p><ol><li>OKRs</li><li>User Flows</li><li>Profit Models</li><li>Instrumentation &amp; Events Catalog</li><li>Metrics Catalog</li><li>Reporting</li><li>A/B Testing (Bonus!)</li></ol><h3>1. Objectives &amp; Key Results</h3><p>“What do OKRs have to do with data?” one may ask. However, John Doerr, the person attributed for bringing OKRs to Google, would answer “OKRs are data.” Specifically, they are data about our organization’s ability to achieve its expected goals. See Doerr discuss OKRs in more detail <a href="https://www.youtube.com/watch?v=HiQ3Ofcmo50">here</a> and how Google uses OKRs <a href="https://www.youtube.com/watch?v=mJB83EZtAjc&amp;t=3042s">here</a>.</p><p><strong>Objectives</strong> — a qualitative description of a team or company’s goals — articulate our business strategy. <strong>Key Results</strong> (or KRs) contain specific measures that track a team or company’s progress toward its Objectives. We call these measures <strong>Key Performance Indicators</strong>. KPIs are data that track our business’ progress towards its goals. Companies often overlook their internal data. Using OKRs in your organization is one of the easiest and most effective ways to be data-driven.</p><p><strong>How to Make It Data-Driven</strong></p><p>Data-driven companies consistently set OKRs, track them with KPIs, and embed them into daily business operations. These companies also use OKRs at the company, business, and team levels. Regardless of the level in the org chart, we should:</p><ul><li><strong><em>Make them accessible</em></strong>. Make team, department, and company OKRs easily and widely available. Maintain dashboards that show KPIs next to their associated OKRs.</li><li><strong><em>Use them as a learning tool</em></strong>. Conduct retrospectives at the end of each quarter and year. Take retrospective learnings and incorporate them into future OKR planning and business operations.</li><li><strong><em>Store them as data</em></strong>. Track team and company OKRs in a way that we can access forever, ideally in a central repository. Store KPIs in a central database.</li></ul><p><strong>What To Do Day 1</strong></p><p>Start by drafting a <strong>Company Mission Statement</strong>. This is a statement that articulates the high-level purpose and aspirations of the business. These will act as guardrails we brainstorm our company’s OKRs since each OKR should move the company towards its Mission.</p><h3>2. User Flows</h3><p>User flows are diagrams of how users move through our product. By (1) understanding how users move through our app, we can (2) quantify how users move through our app. And, as a corollary, we can (3) track and improve how users move through our app.</p><p>User flows also build a bridge between strategic and technical teams. For every step in the user flow, we should be able to specify the events (i.e instrumented actions), metrics, web pages, mobile screens, datasets, services, or any other technical entity that relates to that step in the user flow.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*P_ClopJagooQxraDdCc-2Q.png" /><figcaption>A sample user flow, with mappings for associated teams, services, and metrics.</figcaption></figure><p><strong>How to Make It Data-Driven</strong></p><p>User flows capture users’ meaningful movements through our business — such as purchases on Amazon, a page creation on Notion, or an article read on Medium. Once we’ve identified meaningful metrics, we can optimize those metrics by improving the user experience.</p><p>Given this, there are a few simple ways to make the user flows data-driven.</p><ul><li><strong><em>Identify the meaningful actions in the business</em></strong> — such as transactions, upgrades, or other indicators of increased loyalty to the product. Use Amplitude, Google Analytics, or the like to track users’ movement towards these meaningful actions.</li><li><strong><em>Create metrics that quantify transition points towards meaningful actions</em></strong>. For instance, if your meaningful action is “transactions,” a supporting metric could be “Transactions per MAU.” If “upgrades” is a meaningful metric, “% of Registered Users with Upgrade” can track this.</li><li><strong><em>Align A/B testing strategy with the user flow</em>.</strong> We want to be as strategic as possible when choosing experiments to operationalize. Most tests take effort to properly implement and analyze; therefore, we should use our “meaningful action metrics” and historical tests</li></ul><p><strong>What To Do Day 1</strong></p><p><strong>Build a user flow for your core UX and/or features</strong>. If you have an engineer, refer to the user flow when identifying the events to the instrument. Refer to these resources about building user flows.</p><ul><li><a href="https://medium.com/the-uxplorer/ux-101-how-to-create-user-flows-182059ad4562">UX 101: How to create user flows</a> by Leow Hou Teng</li><li><a href="https://medium.com/7bits/user-flow-101-what-it-is-how-to-do-it-84052141fe86">User Flow 101: What it is &amp; How to do it</a> by Gabriel Sa e Farias</li></ul><h3>3. Profit Model</h3><p>Financial modeling seems like a challenging concept that requires a Series 7 or MBA. But don’t worry — we’re not talking discount cash flows here. By “profit model,” we mean a simple but analytically thoughtful approach to understanding how our business makes money.</p><p>When building profit models, we want to account for 4 things:</p><ul><li><strong>Revenue</strong>. How we make money. We can use two simple frameworks — Revenue Decomposition and Customer Lifetime Value — to project revenue margins as our business scales (see more below).</li><li><strong>Cost of Goods Sold</strong> (COGS). Direct costs related to our product or services. This excludes external costs, such as marketing and human capital costs. We should know our production costs at the product level, as well as how those costs change when we sell 100, 1K, or 10K monthly units.</li><li><strong>Marketing Costs</strong>. How much does it cost to acquire a customer as we scale? We can simply use the average CPA, but we should also track how that changes as our acquisition spend scales.</li><li><strong>Contingencies and business dependencies</strong> that materially impact any of the points above. For instance, if costs fall significantly as we purchase more units from our suppliers, we should account for different supply levels.</li></ul><p><strong>How to Make It Data-Driven</strong></p><p>We can use the following revenue frameworks to get a comprehensive view of revenues, costs, and favors.</p><ul><li><strong>Revenue Decomposition</strong>. This is a process where we deconstruct revenue into its subcomponents, e.g. revenue per user, transactions per user, and revenue per transaction. This article by GooglePlay’s Alyssa Perez on <a href="https://medium.com/googleplaydev/a-kpis-guide-for-google-play-apps-and-games-intro-to-metrics-214da979f5a2">building a metrics tree</a> provides a good summary of this approach.</li><li><strong>Customer LTV</strong>. We can model the revenue we receive from customers over time to understand their lifetime value to the business. It’s important to have awareness of this number and the factors that impact it as we invest in customer acquisition and retention. This <a href="https://towardsdatascience.com/best-practices-in-ml-observability-for-customer-lifetime-value-ltv-models-c5a2fc063f4c">article</a> is a (not overly) technical overview of calculating LTV.</li></ul><p><strong>What To Do Day 1</strong></p><p><strong>Build a Revenue Tree. </strong>A Revenue Tree is simply a breakdown of revenue into its substituent metrics. Your business’ revenue tree will depend on its monetization. That said, we’ve provided a sample revenue tree below. See more here about <a href="https://medium.com/googleplaydev/a-kpis-guide-for-google-play-apps-and-games-intro-to-metrics-214da979f5a2">building a metrics tree</a>.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/960/1*Y10qEpPeiwlziwWxQISHaQ.jpeg" /><figcaption>A sample revenue tree for an e-Commerce company.</figcaption></figure><p><strong>Build a Customer Lifetime Value (LTV) Model. </strong>Even before our business generates revenue, we can build an LTV model to serve as a projection and benchmark. As time passes, you will gain a more clear view of your customers’ LTV. Similar to the revenue tree, accurately modeling LTV depends on how the business makes money. However, a simple formula for LTV is:</p><p><strong>LTV = ((Average Sales ✕ Purchase frequency) / Churn)) ✕ Profit Margin</strong></p><p><em>where:</em></p><ul><li>Average Sales = (Total Sales) / (Total # of Orders)</li><li>Purchase frequency = (Total # of Orders) / ( Total # of Unique Customers)</li><li>Churn = 1 — Retention Rate</li></ul><h3>4. Instrumentation &amp; Events Catalog</h3><p>Before we produce data, metrics, or insights — we must collect the data. Events are actions that we can capture digitally. They are data’s moments of inception.</p><p><strong>Instrumentation</strong> is the process of identifying and capturing events. Needless to say, the instrumentation process is very important. Garbage in, garbage out. And we should set instrumentation early as possible because, as time goes on, it’s increasingly difficult to change your data’s foundation.</p><p><strong>How To Make It Data-Driven</strong></p><p>There are two steps to instrumentation: planning and execution. Execution means adding tracking code to our product. Analytical platforms such as Google Analytics, Firebase, and Amplitude make implementation easier. However, the logic we establish during the planning is just as important as how we implement it.</p><p>This planning process should:</p><ul><li><strong><em>Include PMs, Engineers, Designers, and Analysts</em></strong>. Building a product is a collective effort. Therefore, collecting information about the product should be as well. Different roles will provide different perspectives on how we should track user behavior.</li><li><strong><em>Take KPIs and other metrics into consideration</em></strong>. As we instrument our product, we should ensure that we can track all the metrics we use. If we’re using Google Analytics or Tag Manager, we should create custom events to track these metrics. Other analytics platforms (e.g. Amplitude) have similar functionality as well.</li><li><strong><em>Record event names, definitions, and other relevant metadata in an Event Catalog</em></strong> that is broadly available across the organization. This will give all employees visibility into how our data are produced. This, in turn, makes our entire organization more data-informed.</li><li><strong><em>Consider all data sources</em></strong>. Product instrumentation alone may not provide all of the information we need. We may need to collect data from 3rd party systems, audit logs, or employee/customer surveys. Dare to think unconventionally about data collection.</li></ul><h3>5. Metrics Catalog</h3><p>”What metrics do I need for my business?” is another common question I hear from pre-launch founders. We’ve identified 3 frameworks to identify business metrics. However, a more important question to ask is “how should I use metrics in my business?”</p><p>We can think about how metrics are used in a business using a lifecycle. We can break this “metrics lifecycle” into 3 steps:</p><ol><li>Metrics Selection</li><li>Metrics Production</li><li>Metrics Usage</li></ol><h4>Metrics Selection</h4><p>This is the process of identifying metrics to use within a company or team. Below are 3 metrics frameworks that will provide a comprehensive set of metrics for an early-stage startup.</p><ul><li><strong><em>Revenue Trees</em></strong>. As discussed in the “Profit Model” section, this framework deconstructs our business’ revenue over a given time period.</li><li><strong><em>HEART Metrics</em></strong>. A framework developed by Google to help teams select a thorough set of product metrics. This article discusses how to <a href="https://medium.com/agileinsider/how-to-set-user-centered-metrics-the-google-heart-framework-ab41bf217a34">set metrics using the HEART framework</a>.</li><li><strong><em>VC Metrics</em></strong>. If you’re looking for outside capital, there is a set of metrics that VCs and other investors will commonly request. Examples include &lt;&lt; give examples &gt;&gt;. This article outlines some of the <a href="http://tapes.scalevp.com/vc-reveal-startup-metrics/">common metrics VC firms use</a> to evaluate startups.</li></ul><p><strong>How to Make It Data-Driven.</strong> In data-driven organizations, metrics are tailored to the business or teams they support. Company and team leaders ensure their team can improve the Key Performance Indicators they select</p><h4>Metrics Production</h4><p>This is the process of converting data into metrics that can be stored, updated, and visualized. We don’t need complex software or processes to produce metrics. Google Analytics, Amplitude, and other analytical SaaS platforms will provide most metrics without a single line of code.</p><p><strong>How to Make it Data-Driven.</strong> Use a Single Source of Truth when sourcing and calculating metrics. Systems such as <a href="https://docs.getdbt.com/">dbt</a> help companies consistently query metrics across the organization.</p><h4>Metrics Operations</h4><p>Once we have metrics, we must leverage them in day-to-day business operations. There is a myriad of ways data-driven companies use metrics. The key is for teams to consistently show and discuss metrics across the organization, and hold themselves accountable to the KPIs they set for themselves.</p><p><strong>How to Make it Data-Driven.</strong> Weave data into business operations, conversations, and stories.</p><ul><li><strong><em>Business Operations</em></strong>. Anyone who completes revenue-generating tasks in your business will be able to complete them better with contextual information. The key is not to overwhelm non-technical employees with data visualizations — keep Ops reporting short and simple.</li><li><strong><em>Conversations</em></strong>. Agile teams and companies source perspectives of the entire team. These perspectives are shared through conversations, debates, or (at times) arguments. By adding facts (i.e. references to metrics and insights), we can make more informed communal decisions.</li><li><strong><em>Stories</em></strong>. It’s often said that business stories are the best way to communicate insights. By adding the occasional metric to your business stories, they will become more memorable and shareable for the audience.</li></ul><blockquote><strong><em>Don’t Say This:</em></strong><em> The new release to our mobile app has been great! People like it and use it more than the previous version.</em></blockquote><blockquote><strong><em>Instead, Say This:</em></strong><em> The new release to our mobile app has been great! Existing users spend 20% more time on the app per week than the previous version, and our mobile NPS increased 12 points!</em></blockquote><h3>6. Reporting</h3><p>Once you have metrics, you must visualize them. Almost all businesses use reporting, but few use reporting to inform business decisions in a meaningful way. To do so requires the right culture and processes, as well as good reporting. For now, we’ll focus on the reporting itself, and discuss ways we can build reports that set our teams up for success.</p><p>There are 3 foundational types of reporting we should build for our startup. As the business scales, we can add more reports to our portfolio, but these will allow us to be data-driven from the outset and will become important assets for our business as it scales.</p><ul><li>Executive Reporting</li><li>Product Reporting</li><li>Operational Reporting</li></ul><h4>Executive Reporting</h4><p>Though it’s in the name, these reports aren’t only for executives. These reports contain the most important information and metrics about the business, such as:</p><ul><li>Key Performance Indicators (KPIs) from company Key Results</li><li>High-level metrics, such as MAU or MRR</li><li>Custom metrics that are most relevant to the business</li></ul><p>Similar to OKRs, executive reporting aligns all employees by providing a single view of the business. A manager once told me, “treat the business like you own it.” With executive reporting, you can give all employees an owner’s perspective.</p><h4>Product Reporting</h4><p>This report shows the health of your business’s core product. Generally, these are websites, apps, or other digital products; however, they can also be physical products, courses, or experiences. Regardless of what you sell, we must build a portfolio of metrics that track the performance of what we sell. We can use the metrics frameworks we discussed previously to build a robust set of product metrics.</p><h4>Operational Reporting</h4><p>Many startups make tremendous efforts to optimize their product, tracking and analyzing every interaction that users make. Yet they don’t consider <em>customer-to-employee</em> or <em>employee-to-employee</em> interactions. Operational reporting provides information to employees that help them do their daily jobs better. This may range from real-time information sent to assembly line employees to employee satisfaction metrics sent to managers.</p><h4>How to make it Data-Driven</h4><p>Again, reporting is everywhere. “Data-driven reporting” depends on the principles and processes used to build the report. Here is a short list of reporting principles.</p><ol><li><strong><em>Be precise</em></strong>. A small change in a SQL query can substantially change in the definition of the metrics it produces. A small change in the definition of a metric can substantially change its relevance to the business. Therefore we should be precise about how we calculate, define, and communicate metrics found in reports.</li><li><strong><em>Be polished</em></strong>. Many people believe that aesthetics doesn’t influence analytics. They do. Use clean, simple visualizations that can be understood without an explanation (aside from what’s already in the report).</li><li><strong><em>Be contextual</em></strong>. People need information about a number to understand why the number matters. Adding metric definitions, hyperlinks to references, or metadata about the data source (e.g. date-time since the last update) helps viewers comprehend what they’re viewing and why it matters.</li></ol><h3>7. A/B Testing (bonus!)</h3><p>The purpose of analytics in any organization can be broken into 4 areas:</p><ol><li>Understanding the business</li><li>Identifying ways to improve the business</li><li>Implementing methods to improve the business</li><li>Calculating the impact of said improvements</li></ol><p>Analytics generally focuses on #1, #2, and #4. #3 is more often completed by Engineers (by making changes to the product), Strategists (e.g. by changing campaigns and marketing spending), or Operators (by executing revenue-generating tasks).</p><p>Of the steps we’ve discussed in this article, Steps 1–6 are focused on understanding the business. A/B testing is one of the ways we can identify ways to quantify the impact of changes on the business.</p><p>We should implement A/B testing as early as possible. Doing so will create a strong test-and-learn culture that will persist as the business scales. And implementing A/B tests doesn’t require complex software or processes (provided that we have accurate instrumentation). <a href="https://marketingplatform.google.com/about/optimize-360/">Google Optimize</a> is a free, easy solution for A/B testing.</p><p><strong>How To Make It Data-Driven</strong></p><p>When conducting A/B testing, we should:</p><ol><li><strong><em>Think strategically about what to test</em></strong>. Prioritize tests that can have the greatest potential impact on the business. Maintain a backlog of tests the team would like to implement and refer back to it when considering new tests.</li><li><strong><em>Treat experimentation as a continuous learning process</em></strong>. Collect results and takeaways from every experiment. With a centralized experiments repository, your team will eventually gain “meta insights” about the levers that best impact your business.</li></ol><h3>Conclusion</h3><p>Analytics is often an afterthought for founders of early-stage startups. There are so many responsibilities to “keep the lights on” and data, frankly, isn’t on that list.</p><p><strong>The problem is that good data practices become imperative as the business scales</strong> (hello, auditors), and data-informed decisions will as well (hello, competitors). And, if we don’t think strategically about data from the start, building an effective analytics practice will be far more challenging in the future.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4e4f900c2667" width="1" height="1" alt=""><hr><p><a href="https://medium.com/swlh/6-easy-steps-to-a-data-driven-startup-on-day-1-4e4f900c2667">6 Easy Steps to a Data-Driven Startup on Day 1</a> was originally published in <a href="https://medium.com/swlh">The Startup</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[What are Python’s most used functions?]]></title>
            <link>https://thedatastrategist.medium.com/what-are-pythons-most-used-functions-d760dc28fd96?source=rss-182bce5c9ee8------2</link>
            <guid isPermaLink="false">https://medium.com/p/d760dc28fd96</guid>
            <category><![CDATA[python-programming]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[bigquery]]></category>
            <category><![CDATA[functions-in-python]]></category>
            <category><![CDATA[jupyter-notebook]]></category>
            <dc:creator><![CDATA[Gordon Silvera]]></dc:creator>
            <pubDate>Wed, 24 Aug 2022 16:07:34 GMT</pubDate>
            <atom:updated>2023-09-13T15:40:01.573Z</atom:updated>
            <content:encoded><![CDATA[<h3>The Top 10 Python Functions used by Data Scientists</h3><h4>We analyzed 200K+ IPython notebooks to find out.</h4><p><a href="https://datastudio.google.com/reporting/a5096f7e-26c8-48f7-a496-da1fdef6b008"><strong>“The Most Python” report</strong></a> attempts to provide a simple view of how Python is used for data science. We did this by analyzing 200K+ IPython (aka Jupyter) notebooks and 2M+ StackOverflow questions, which resulted in:</p><ol><li>A list of <strong>the most-used Python functions</strong> in IPython notebooks</li><li>For each function, the <strong>most common questions</strong> from StackOverflow</li><li>For each function, <strong>code samples</strong> from the most popular notebooks on Github</li></ol><p>See the <a href="https://datastudio.google.com/reporting/a5096f7e-26c8-48f7-a496-da1fdef6b008">full report</a> here or check out the top 10 functions below.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/0*c1UFMeh9IXXgWTCL.jpg" /><figcaption>See the full, interactive <a href="https://datastudio.google.com/reporting/a5096f7e-26c8-48f7-a496-da1fdef6b008">“The Most Python” report</a>.</figcaption></figure><h3>Methodology</h3><p>Github makes code from open-source licensed available in BigQuery’s <a href="https://console.cloud.google.com/bigquery?ws=!1m5!1m4!4m3!1sbigquery-public-data!2sgoogle_trends!3sinternational_top_terms!1m4!1m3!3m2!1sbigquery-public-data!2sgithub_repos">public data</a>. We filtered this data to “.ipynb” files and, after much regex and data manipulation, we parsed the data into notebook cells, functions, and libraries. From this, we have a (mostly) clean connection from Github repositories → file paths → files (i.e. notebooks) → notebook cells → code, functions, or libraries. Seen <a href="https://github.com/the-datastrategist/the_most_python/tree/main/sql/">the code</a> here.</p><h3>Top 10 Most-Used Python Functions</h3><p>We’ve collected a list of <strong>the top 10 most-used Python functions in Jupyter notebooks.</strong> This is based on 200K+ open-source Jupyter notebooks on Github. Then we cross-referenced this data with 2M+ StackOverflow questions to identify the most common <em>questions</em> about the most common <em>Python functions</em>.</p><h3>10) np.zeros</h3><p>Produces a Numpy array of zeros. This is particularly useful when creating vectors in TensorFlow or for other machine learning applications. See more in the <a href="https://numpy.org/doc/stable/reference/generated/numpy.zeros.html">Numpy docs</a> or see how <strong>np.zeros</strong> and <strong>TensorFlow</strong> are used in <a href="https://github.com/xlnwel/book-mooc/blob/master/Udacity-deep-learning/intro-to-tflearn/TFLearn_Sentiment_Analysis.ipynb">sentiment analysis</a>, with code from a Udacity Deep Learning course.</p><p><strong>StackOverflow’s Most Common Questions</strong></p><ul><li><a href="https://stackoverflow.com/questions/50973041/typeerror-data-type-not-understood-numpy-zeros">My data type is not understood when using np.zeros</a>.</li><li><a href="https://stackoverflow.com/questions/18397805/how-do-i-delete-a-row-in-a-numpy-array-which-contains-a-zero">How do I delete a row in a numpy array which contains a zero?</a></li></ul><h3>9) float</h3><p>Converts a string or integer to a floating-point number. Float is used in 9.5% of notebooks and is used 19 times per notebook on average. See more about float() at <a href="https://www.w3schools.com/python/ref_func_float.asp">w3 schools</a>.</p><p><strong>StackOverflow’s Most Common Questions</strong></p><ul><li>How to limit floats to two decimal points.</li><li>How to check if a number is a float.</li></ul><h3>8) __init__</h3><p>When we create a Python object by running a class, <strong>__init__</strong> initializes the data stored in the object. From there, we can apply methods (aka functions) to transform the data in our object. See more about <a href="https://www.geeksforgeeks.org/__init__-in-python/">__init__</a> in the docs from GeeksforGeeks.</p><p>The example below shows how we use __init__ in a simple Python class.</p><pre># Sample class with init method<br>class Person:<br> <br>    # init method or constructor<br>    def <strong>__init__</strong>(self, name):<br>        self.name = name<br> <br>    # user-defined method<br>    def say_hello(self):<br>        print(&#39;Hello, my name is&#39;, self.name)<br> <br>p = Person(&#39;Roger&#39;)<br>p.say_hello()</pre><p><strong>StackOverflow’s Most Common Questions</strong></p><ul><li><a href="https://stackoverflow.com/questions/11536764/how-to-fix-attempted-relative-import-in-non-package-even-with-init-py">How to fix “Attempted relative import in non-package” even with __init__.py</a>.</li><li><a href="https://stackoverflow.com/questions/2491819/how-to-return-a-value-from-init-in-python">How to return a value from __init__ in Python?</a></li></ul><h3>7) np.array</h3><p>Numpy’s array function outputs an n-dimensional array, based on the inputs specified. Numpy arrays are often used in machine learning, as they allow for smaller data storage and faster processing <a href="https://stackoverflow.com/questions/993984/what-are-the-advantages-of-numpy-over-regular-python-lists">[i]</a> than Python data objects.</p><p><strong>StackOverflow’s Most Common Questions</strong></p><ul><li><a href="https://stackoverflow.com/questions/384759/how-do-i-convert-a-pil-image-into-a-numpy-array">How do I convert a PIL Image into a NumPy array?</a></li><li><a href="https://stackoverflow.com/questions/6910641/how-do-i-get-indices-of-n-maximum-values-in-a-numpy-array">How do I get indices of N maximum values in a NumPy array?</a></li><li><a href="https://stackoverflow.com/questions/9236926/concatenating-two-one-dimensional-numpy-arrays">Concatenating two one-dimensional NumPy arrays</a>.</li></ul><h3>6) format</h3><p>The format function provides a simple way to print strings dynamically. See examples below or in <a href="https://datastudio.google.com/u/0/reporting/a5096f7e-26c8-48f7-a496-da1fdef6b008/page/XGi0C?params=%7B%22df13%22:%22include%25EE%2580%25800%25EE%2580%2580IN%25EE%2580%2580format%22%7D">“The Most Python” report</a>. Or see more about the format function from <a href="https://www.w3schools.com/python/ref_string_format.asp">w3 schools</a>.</p><pre>import numpy as np <br>x, y = np.full(4, 1.0), np.full(4, 2.0) <br>print(&quot;{} + {} = {}&quot;).<strong>format</strong>(x, y, x + y)</pre><p>Here the data scientist is using <strong>.format()</strong> to build filepath references and display the first image within each subfolder.</p><pre>for str in [&#39;A&#39;, &#39;B&#39;, &#39;C&#39;, &#39;D&#39;, &#39;E&#39;, &#39;F&#39;]: <br>    root = &#39;notMNIST_small&#39; <br>    path = os.listdir(&#39;{}/{}&#39;.<strong>format</strong>(root, str))[0]     <br>    display(Image(&#39;{}/{}/{}&#39;.<strong>format</strong>(root, str, path)))</pre><p><strong>StackOverflow’s Most Common Questions</strong></p><ul><li><a href="https://stackoverflow.com/questions/5466451/how-do-i-print-curly-brace-characters-in-a-string-while-using-format">How do I print curly-brace characters in a string while using .format?</a></li><li><a href="https://stackoverflow.com/questions/8450472/how-to-print-a-string-at-a-fixed-width">How to print a string at a fixed width?</a></li></ul><h3>5) int</h3><p>The function, <strong>int()</strong>, converts an input into an integer. This input can be a string, number, or bytes object. It’s used in 15% of notebooks. See examples below or read more from <a href="https://www.programiz.com/python-programming/methods/built-in/int">Programiz.com</a>.</p><p><strong>StackOverflow’s Most Common Questions</strong></p><ul><li><a href="https://stackoverflow.com/questions/7368789/convert-all-strings-in-a-list-to-int">Convert all strings in a list to int</a>.</li><li><a href="https://stackoverflow.com/questions/20449427/how-can-i-read-inputs-as-numbers">How can I read inputs as numbers?</a></li><li><a href="https://stackoverflow.com/questions/2269827/how-to-convert-an-int-to-a-hex-string">How to convert an int to a hex string?</a></li></ul><h3>4) str</h3><p>The function, <strong>str()</strong>, converts inputs into a string. It is found in 14% of notebooks and is used an average 25 times per notebook. The example below converts a set of dates into strings to be included filenames.</p><pre>for row, item in publications.iterrows():<br>    md_filename = <strong>str</strong>(item.pub_date) + &quot;-&quot; + item.url_slug + &quot;.md&quot; <br>    html_filename = <strong>str</strong>(item.pub_date) + &quot;-&quot; + item.url_slug <br>    year = item.pub_date[:4]</pre><p><strong>StackOverflow’s Most Common Questions</strong></p><ul><li><a href="https://stackoverflow.com/questions/3437059/does-python-have-a-string-contains-substring-method">Does Python have a string ‘contains’ substring method?</a></li><li><a href="https://stackoverflow.com/questions/3277503/how-to-read-a-file-line-by-line-into-a-list">How to read a file line-by-line into a list?</a></li></ul><h3>3) range</h3><p>Range returns a sequential list of numbers. Range takes the inputs: <em>start</em>, <em>stop</em>, and <em>step</em>, where:</p><ul><li><strong>Start</strong> is the starting number of the list, 0 by default</li><li><strong>Stop</strong> is the number at which the range should end</li><li><strong>Step</strong> is the value to increment in each step</li></ul><p>Range() is commonly used in for() loops, partly why it’s found in <strong>36% of notebooks</strong>. Here’s an application of range in a for loop:</p><pre># Return a list of batch size pairs<br>def get_batches(x, y, batch_size=100):<br>    n_batches = len(x)//batch_size <br>    x, y = x[:n_batches*batch_size], y[:n_batches*batch_size] <br>  <br>    for ii in <strong>range</strong>(0, len(x), batch_size): <br>        yield x[ii:ii+batch_size], y[ii:ii+batch_size]</pre><p><strong>StackOverflow’s Most Common Questions</strong></p><ul><li><a href="https://stackoverflow.com/questions/477486/how-do-i-use-a-decimal-step-value-for-range">How do I use a decimal step value for range()?</a></li><li><a href="https://stackoverflow.com/questions/7286365/print-a-list-in-reverse-order-with-range">Print a list in reverse order with range()</a><a href="https://stackoverflow.com/questions/477486/how-do-i-use-a-decimal-step-value-for-range">?</a></li></ul><h3>2) len</h3><p><strong>len()</strong> returns the length of a string, list, dataframe, or any othe type of Python object. Needless to say, this is commonly-used in IPython notebooks, as data scientists regularly check the size of the data they’re transforming. <strong>38% of notebooks</strong> use len().</p><p>See an example below, where we use len() to get for non-empty values from a list, and use it again to print summary information about the output list.</p><pre># Get all index values with non-empty reviews<br>non_zero_idx = [<br>    ii for ii, review in enumerate(reviews_ints) if <strong>len</strong>(review) != 0]</pre><pre># Print the number of non-empty reviews<br>print(<strong>len</strong>(non_zero_idx))</pre><h3>1) print</h3><p>No other function is used as often as the <strong>print()</strong> in IPython notebooks. This makes sense, as data scientists output information at the end of nearly every cell. One-third of notebooks use print() and it’s used <strong>31 times per notebook</strong> on average.</p><pre>import numpy as np <br>x, y = np.full(4, 1.0), np.full(4, 2.0)<br><strong>print</strong>(&quot;{} + {} = {}&quot;.format(x, y, x + y))</pre><pre># Print using list comprehension<br><strong>print</strong>([x + y for x, y in zip([1.0] * 4, [2.0] * 4)])</pre><p>In the following example, the data scientist uses print to provide status updates during a data load.</p><pre>import tensorflow as tf<br>from tensorflow.examples.tutorials.mnist <br>import input_data  </pre><pre><strong>print</strong>(&#39;Getting MNIST Dataset...&#39;)<br>mnist = input_data.read_data_sets(\&quot;MNIST_data/\&quot;, one_hot=True) <strong>print</strong>(&#39;Data Extracted.&#39;)</pre><p><strong>StackOverflow’s Most Common Questions</strong></p><ul><li><a href="https://stackoverflow.com/questions/17153779/how-can-i-print-variable-and-string-on-same-line-in-python">How can I print variable and string on same line in Python?</a></li><li><a href="https://stackoverflow.com/questions/15286401/print-multiple-arguments-in-python">Print multiple arguments in Python</a></li></ul><h3>View “The Most Python” report on Python Functions</h3><p>Check out “<a href="https://datastudio.google.com/reporting/a5096f7e-26c8-48f7-a496-da1fdef6b008"><strong>The Most Python</strong></a>” report to see 100+ of the most-used Python functions, along with common questions and code samples for each.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/750/0*cgsSZ-mdV6GWoMzW.jpg" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d760dc28fd96" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[So you hired a data scientist … now what? Key players in a data-driven business.]]></title>
            <link>https://thedatastrategist.medium.com/so-you-hired-a-data-scientist-now-what-key-players-in-a-data-driven-business-8eebd88d7b90?source=rss-182bce5c9ee8------2</link>
            <guid isPermaLink="false">https://medium.com/p/8eebd88d7b90</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[management]]></category>
            <category><![CDATA[data-strategy]]></category>
            <dc:creator><![CDATA[Gordon Silvera]]></dc:creator>
            <pubDate>Thu, 22 Apr 2021 16:13:06 GMT</pubDate>
            <atom:updated>2021-04-22T16:13:06.034Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*c6xMjCSiWDWxc5rvf8QSng.jpeg" /></figure><h3>So you hired a data scientist … now what? Key players in a data-driven business.</h3><p>When organizations first acquire a data scientist or an analytics team, they may be unsure how to properly maximize their new-found talent. Analytics teams cannot create business value in isolation. This requires engagement from stakeholders across verticals and pay-grades. Below are the “prototypes” of stakeholders needed to make the most of a company’s analytics team.</p><h3>Key players in a data-driven company</h3><h4><strong><em>The Advocate</em></strong></h4><p>A business leader who champions analyses from the data team and shields them from inconsequential work. Ideally, this is someone in a strategic position that is well-regarded within the organization. Migrating from a <em>managerial-driven </em>to a <em>data-driven</em> strategy can be difficult in companies with traditional management structures. Therefore, it is important to have <strong>someone with the mandate to push analytical recommendations into strategic operation</strong>.</p><h4><strong><em>The Translator</em></strong></h4><p>Analytics can be complex and should, therefore, be simplified for the relevant audience and given an authoritative voice. In cases when analysts cannot effectively communicate findings, you will need to find <strong>someone who can clearly communicate insights</strong> to all levels of an organization. Even if this person cannot code, they should have the technical acumen to understand all aspects of a given analysis, as well as the ability to communicate the analysis effectively.</p><h4><strong><em>The Connector</em></strong></h4><p>A critical person within any business, but particularly within analytics. They (a) bring business issues discussed from the “water cooler” to the analytics team, and (b) bring insights from the analytics team to the company at large. Generally speaking, <strong>they connect people and ideas across the business</strong>. Malcolm Gladwell delves into this prototype in <a href="http://www.amazon.com/gp/product/B000OT8GD0/ref=dp-kindle-redirect?ie=UTF8&amp;btkr=1">Tipping Point</a>.</p><h4><strong><em>The Machinist</em></strong></h4><p>The person who cranks code and ensures it’s done properly. These are often more junior analysts. While this role may be a more mundane data role, it is critical and often requires intense attention to detail. When hiring for these roles, use case and situational questions to gauge candidates&#39; attention to detail. Also, encourage these analysts to deeply explore and understand the company’s pre-existing code base — this will provide a learning opportunity and will enable them to more easily detect shortcomings or QA issues in any existing reports or analysis.</p><h4><strong><em>The Rogue</em> (optional)</strong></h4><p>The whip-smart but slightly cynical analyst or programmer who has a constant stream of ideas, and who build things that are “cool” but not necessarily business-relevant. They’re the <a href="https://www.youtube.com/watch?v=69V__a49xtw">Richard Hendricks</a> of your organization (while he was still at Hooli). Rather than sidelining these individuals’ creative ambitions, channel their energy and focus by giving them opportunities to explore ways to improve the business or explore data differently. But, most importantly, give them room to create and help bring their ideas to fruition in a business-relevant way.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8eebd88d7b90" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Insights vs Product vs Engineering Data Science, and how each provides value to your business]]></title>
            <link>https://thedatastrategist.medium.com/insights-vs-product-vs-engineering-data-science-and-how-each-provides-value-to-your-business-a52785ea58b6?source=rss-182bce5c9ee8------2</link>
            <guid isPermaLink="false">https://medium.com/p/a52785ea58b6</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[data-strategy]]></category>
            <category><![CDATA[data-engineering]]></category>
            <category><![CDATA[data-job]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Gordon Silvera]]></dc:creator>
            <pubDate>Thu, 22 Apr 2021 15:55:27 GMT</pubDate>
            <atom:updated>2021-04-22T15:55:27.953Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/925/1*TPqsFkN8KRWbPgvUZCaLTA.jpeg" /></figure><p>Data science roles in tech businesses fall into three categories:</p><ul><li><strong><em>Insights</em></strong><em>.</em> Using data science to understand users, products and businesses.</li><li><strong><em>Product</em></strong><em>. </em>Using data science to test and optimize a product or feature.</li><li><strong><em>Engineering</em></strong><em>.</em> Building models and data, which are then integrated into product features or used by other stakeholders in the organization.</li></ul><p>These approaches can be very different in practice, requiring different tools, expertise, and operational processes. Nonetheless, businesses are only starting to differentiate these forms of data science. By properly differentiating these roles, businesses can more effectively hire, cultivate, and retain data science talent.</p><p>In this article, we will discuss the differences between Insights, Product, and Machine Learning data science. Below is a summary of the key differences and examples of each application. We will view these practices through the following dimensions: goals, deliverables, and tech stacks.</p><ul><li><strong><em>Purpose</em></strong>. What is the fundamental role of the data scientist? What stakeholders do they partner with? How do they provide value to the organization?</li><li><strong><em>Deliverables</em></strong>. What types of outputs do these roles produce? How does staffing and sprints differ?</li><li><strong><em>Tech Stack</em></strong>. What types of tools and programming languages are used?</li></ul><h3>1. Purpose</h3><h4>Insights Data Science</h4><p>Fundamentally, the goal of insights is to use data to gain a deeper understanding of our business — its products, employees, customers and anything else that interacts in the business’ ecosystem.</p><p>This understanding is guided by stakeholders’ business questions, hypotheses and information needs. Insights stakeholders include executives, marketing teams, product managers, HR, customer service, or any other function that would benefit from understanding their business from a qualitative and quantitative lense. As data scientists produce insights and outputs that satisfy their stakeholders’ questions or needs, they contribute to the collective knowledge of the business.</p><p>Insights practitioners at Uber or Lyft may analyze questions such as:</p><ul><li>How do changes in price impact <em>user retention</em>? How does this change based on competitive intensity and price elasticity within a given market?</li><li>Does sending push notifications to riders when the driver arrives reduce <em>time to ride start</em> (i.e. the time between a driver arrives and the ride actually starts)?</li><li>How will launching in Croatia impact MAU (monthly active users) and trips?</li><li>How does surge pricing impact user retention?</li></ul><h4>Product Data Science</h4><p>The goal of product data science is to (a) identify potential improvements to a product or feature, (b) work with engineering to instrument an experiment, (c) execute the experiment, and (d) present results to stakeholders. This process leverages some of the same skills as insight scientists, but is far more focused on A/B testing.</p><p>Problems the data scientists within a ride sharing business may include:</p><ul><li>How can we make ride hailing faster once users have opened the app? How, in turn, does this impact user engagement and retention?</li><li>How does a recent UI design impact user engagement and retention?</li></ul><p>The stakeholders for Product data scientists are different from their Insights counterparts. They work closely with engineering teams and present their findings to Product Managers and Owners, who then give a green/red light on the final product rollout.</p><h4>Engineering Data Science</h4><p>Conversely, engineering DS focuses on producing data and algorithms for product features. While <em>business questions</em> and <em>hypotheses</em> guide strategic analytics<em>, feature use cases</em> guide product data science. We should note that engineering data science is now being further differentiated into:</p><ul><li><strong><em>ML engineers</em></strong>: those who build models that are integrated into products</li><li><strong><em>Data engineers</em></strong>: those who produce massive datasets for product use</li><li><strong><em>Analytical data engineers</em></strong>: those who produce massive datasets for analytical use (for instance, by other data scientists)</li></ul><p>The key difference between these and other data scientists is that they are primarily focused on <em>producing data</em> as an output rather than producing the <em>synthesis of data</em>.</p><p>Within this ride sharing business case, these are some potential use cases:</p><ul><li>Build a model that calculates optimal price-surges based on factors such as demand, time of day, and location.</li><li>Produce a dataset of Manhattan distances for every trip, which can be used for analysis by data scientists.</li><li>Build a model used to personalize email communications to users.</li></ul><p>Since these processes require large, fast-moving data, the underlying infrastructure used is quite different from strategic data science. Also, these data scientists mostly work with technical stakeholders — engineers, data scientists, and tech leads.</p><h3>2. Deliverables</h3><h4>Insights Data Science</h4><p>Insights deliverables tend to fall in these categories:</p><ul><li>Presentations — e.g. decks and reports</li><li>Dashboards</li><li>Data and data pipelines</li><li>Notebooks — e.g. Jupyter, Google Collab</li></ul><p>The end results of the analyses are stand-alone — digestible, value-added information that helps stakeholders make informed decisions about their business. Ideally, learnings from earlier work streams are incorporated into future analyses, resulting in an iterative, evolving collection of insights. The key to these sprints is focus — that is, to ensure the business questions and resulting insights are confined as possible to maximize the value of the results.</p><p>Delivery cadences depend on the complexity of the data and analysis, ranging from a couple weeks to a year. Often, data scientists will work on multiple analyses concurrently. A seasoned data scientist can effectively manage communications and expectations with their stakeholders.</p><h4>Product Data Science</h4><p>Deliverables for Product data science tend to have the same format as Insights data science; however, their outputs are less strategically-oriented and more defined in scope. Specifically, the scope of the analysis will be the experiment the data scientists and team execute.</p><p>In product data science presentations, it’s important to ensure (a) the experiment results account for any operational or technical challenges encountered during the experiment and (b) the results are properly interpreted by stakeholders. We should ensure uplifts and statistical values are properly understood as they are communicated up the company’s chain of command. We should note — a small change to an app can have a substantial impact on the business, so validating one’s work in product data science is paramount.</p><h4>Engineering Data Science</h4><p>Conversely, deliverables and delivery cadences for Engineer data scientists more closely resemble that of an engineer (of course). These deliverables may include:</p><ul><li>Datasets and data pipelines</li><li>APIs</li><li>Automated processes (for instance, Kubeflow, Airflow, or Serverless pipelines)</li><li>Code containing an machine learning models</li></ul><p>These outputs often become part of the product and, therefore, should be built with the engineers who own the feature or product. In many scenarios, a data scientist will embed or partner with an engineering team to support the implementation of statistical processes.</p><h3>3. Tech Stack</h3><h4>Insights &amp; Product Data Science</h4><p>These data scientists share basically the same toolkit, which generally includes:</p><ul><li>Notebook — e.g. Jupyter, Collab</li><li>Python or R</li><li>SQL and a relational database</li><li>Data visualization software — e.g. Tableau, Periscope, Data Studio</li><li>Github</li></ul><p>However, depending on the requirements of the analysis, these data scientists may use a broad suite of tools, such as from Excel, Cloud Platform Services (compute, storage, …), or machine learning software (e.g. TensorFlow).</p><h4>Engineering Data Science</h4><p>Engineering data science has a broader set of tools and languages to cover. It also requires a boarder domain knowledge, since it incorporates aspects of engineering. These are Engineering data scientists’ primary domains and related tools: *</p><ul><li><em>Big Data query languages</em>: SQL, Spark, Scala, Hive</li><li><em>Databases</em>: Relational, NoSQL</li><li><em>Machine Learning</em>: Python, Tensorflow</li><li><em>DevOps</em>: Docker, Kubernetes, Git Actions, Jenkins</li><li><em>Cloud Infrastructure</em>: compute, storage, access management</li></ul><p>* Note that, depending on the company, systems used, and use cases, these tools can vary widely. (Also note that I’m not an Engineering data scientist, so I can only speak to the tools I know).</p><p>The three main cloud computing platforms are Google Cloud Platform, Amazon Web Services, and Microsoft Azure. Though they are different platforms, most of the services are quite similar and even interchangeable (with some well-planned code).</p><h3>In Conclusion</h3><p>We’ve now discussed some of the key differences across the forms of data science. As the data science practice progresses, we should consider how these different roles play in our business. Businesses have started to question the value data science provides, but much of the shortcomings stem from not using the right people in the right place. By having a cleared understanding of these roles, I hope you can better understand how each of these roles provide value within your organization.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=a52785ea58b6" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>