<?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 Maciej Piwoni on Medium]]></title>
        <description><![CDATA[Stories by Maciej Piwoni on Medium]]></description>
        <link>https://medium.com/@maciej.piwoni?source=rss-e5e269bf081d------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*MENWtmPqPMxv0S1A8_DtSA.jpeg</url>
            <title>Stories by Maciej Piwoni on Medium</title>
            <link>https://medium.com/@maciej.piwoni?source=rss-e5e269bf081d------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 00:23:37 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@maciej.piwoni/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[Transforming Digital Data at Sage.com]]></title>
            <link>https://medium.com/datashop/transforming-sage-digital-data-4433b777a926?source=rss-e5e269bf081d------2</link>
            <guid isPermaLink="false">https://medium.com/p/4433b777a926</guid>
            <category><![CDATA[digital-transformation]]></category>
            <category><![CDATA[big-data]]></category>
            <category><![CDATA[google-cloud-platform]]></category>
            <category><![CDATA[analytics]]></category>
            <dc:creator><![CDATA[Maciej Piwoni]]></dc:creator>
            <pubDate>Tue, 07 Nov 2017 20:08:36 GMT</pubDate>
            <atom:updated>2017-11-07T20:34:32.980Z</atom:updated>
            <cc:license>https://creativecommons.org/licenses/by-nc-nd/4.0/</cc:license>
            <content:encoded><![CDATA[<h4>Going from zero to data hero in 100 days</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*L4IVj0yFsWH98WG5L7gA6A.jpeg" /><figcaption>Photo by Angie Harms <a href="https://goo.gl/SfqQbV">https://goo.gl/SfqQbV</a> / Creative Commons</figcaption></figure><p>We were given the following task: make sense of digital data in Sage. Fixed budget and 100 days. Let me share what went right, what went wrong and how we managed to succeed.</p><h4>Spaghetti junction</h4><p>Imagine a company that owns 10,000 websites, 20+ mobile apps and 1,200+ analytical accounts. Managed by a number of independent teams that usually don’t talk to each other.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Kbo-FtuSAss8EzbhQ9j4_Q.png" /><figcaption>10,000 live websites owned</figcaption></figure><h4>Understand Marketing cloud</h4><p>Sage as many companies created its own marketing cloud. It happened by picking the best in class solution for each problem. This looks like a good idea until you will try to connect systems together.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*cxsWkwFQTLszhOwZ." /><figcaption>Sage custom marketing cloud</figcaption></figure><p>Individual teams had skills and understanding one or two solutions (vertical) but lacked understanding and skills across the entire portfolio solutions (horizontal). For example, media teams were good at media but not necessarily at the CRM and vice versa.</p><p>User journey data was sitting across multiple platforms with very little interconnectivity at best.</p><p>This challenge was amplified by technology vendors itself. Each provider is pushing its own ecosystem. At the same time, they are not very interested in connecting with competitor’s products. This is why you can see the most successful companies switching to a complete stack by a single vendor: Adobe marketing cloud, Google Analytics 360 suite, SalesForce marketing cloud to name a few.</p><h4>Create strategic data layer</h4><p>As for many enterprise businesses, replacing existing platforms was out of the question. The impact on short-term revenue and <em>business as usual</em> (BAU) would be too high. We had to find another solution.</p><p>Data layer to the rescue! Don’t change the system but change input and output instead. In short, unify data independently of the platforms.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*YT7KpNHtYp0y_1F6." /></figure><p>Data layer had obvious advantages for the marketing and data science teams. It allows modelling the data in a much closer way to the customer behaviour without impacting IT.</p><p>In marketing area, instead of simplistic groups, each individual could be a member of multiple groups, interested in multiple products or services with different propensity to buy. Data layer also allowed to create models where each user move from one group to another.</p><h4>Separate apps and data</h4><p>Data layer allowed to detach information from the applications. It also allowed the creative and agile approach to data collection and processing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*1ta42egJl6i6Lsiw." /></figure><p>This was done by creating groups of attributes, taxonomies, triggers for data collection and processing — with minimal changes to the core platforms (for example SalesForce). No reliance (or minimal) on the internal IT team and minimum disruption to the business.</p><h4>How to win adoption? Data vs. analytics vs. reporting</h4><p>The success of the project depended on the adoption.<br>Different stakeholders required a different view of the data.They also consume data in a different way. We were inspired by the<a href="https://hbr.org/2017/05/whats-your-data-strategy"> HBR approach of a single source but multiple versions of truth.</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*mP8HG0dYKYIINeHx." /><figcaption>Combining existing BI tools with more specialised applications (Google stack)</figcaption></figure><p>Business leaders were looking for daily and weekly strategic performance indications like revenue while optimisation managers required access to near real time-data.</p><h4>Define KPIs</h4><p>Key to the success was to focus on each audience and understand what will drive impact for that group.<br>The first step was to map data and expected outcomes. In some cases it was in the form of KPIs in other it was set of triggers and automation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*fMs9EcObfssUGnbW." /></figure><p>Things to consider: complexity, depth of information, speed, required training.</p><h4>What’s under the hood?</h4><p>This can be described as an iceberg — what is visible to internal <em>customers </em>versus data pipelines and technology that delivers the information. Input — output model is quite useful to illustrate high level data flows.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*t93nJoex2wn0lt86." /></figure><h4>Create data pipeline</h4><p>From epics to stories and sprints. We have used conceptual phase to help communicate data flows and data pipelines before implementation work began. The diagram below shows high-level data pipeline — with no specific technologies being used. It helped to find the best tech solutions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*nfKJOwYRvjy2-eMH." /></figure><p>Considering the timeframe of 100 days and all the requirements we have decided to go with <a href="https://cloud.google.com/">Google Cloud Platform</a>. We became one of the first users of <a href="https://cloud.google.com/dlp/">data loss prevention API</a> which allowed us to provide real-time compliance at scale.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*BWCpKugqRuh2HV0V." /></figure><p>Ability to immediately scale up, native integrations and connectivity within Google Cloud were very a huge winners.</p><h4>Immediate results</h4><p>What matters to business users are results that can be seen and <em>felt</em>. To win hearts and minds we started prototyping dashboards and reports parallel to the work on the data pipelines. Integrations within GCP helped us to accelerate this part of the project</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uU_rW28lv2khjWFL5i7o1w.png" /><figcaption>Reports, dashboards. Democratised data and automation</figcaption></figure><p><a href="https://www.google.com/analytics/data-studio/">Data Studio</a> connects natively to the rest of Google stack (Big Query, Google Analytics 360 etc). While not as powerful as Tableau or Qlik it wins where the connectivity, agility and the speed are the key.</p><p><a href="https://cloud.google.com/datalab/">DataLab</a> and <a href="https://research.google.com/colaboratory/unregistered.html">Colaboratory </a>were another two native Google tools we used for at scale data analysis. Again, both natively connects to data pipelines and data sources created within <a href="https://cloud.google.com/">GCP</a>.</p><h4>Lessons for the future</h4><ul><li>Data is bigger than analytics. Get data right first. Analytics will naturally follow,</li><li>Consider mental model used for data analytics — probabilistic approach vs. counting buttons. Some serious stakeholder education may be required,</li><li>Stick with agile.</li><li>Completeness kills. Be aware of diminishing rate of return. You have 80% of data right, getting the next 10% make take a year and $$$. Is it worth it?</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4433b777a926" width="1" height="1" alt=""><hr><p><a href="https://medium.com/datashop/transforming-sage-digital-data-4433b777a926">Transforming Digital Data at Sage.com</a> was originally published in <a href="https://medium.com/datashop">DataShop</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Data Science in eCommerce — Part 5]]></title>
            <link>https://medium.com/datashop/data-science-in-ecommerce-part-5-1cb3d54b971c?source=rss-e5e269bf081d------2</link>
            <guid isPermaLink="false">https://medium.com/p/1cb3d54b971c</guid>
            <category><![CDATA[customer-experience]]></category>
            <category><![CDATA[ecommerce]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Maciej Piwoni]]></dc:creator>
            <pubDate>Fri, 05 May 2017 07:36:14 GMT</pubDate>
            <atom:updated>2017-05-05T07:36:35.229Z</atom:updated>
            <content:encoded><![CDATA[<h3>Finding patters</h3><p>In the last chapter we discovered the following insights:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/883/1*2KzmMkvUMiUJYxW7IaShWQ.png" /><figcaption>Cumulative conversion value (percentage) growing with each channel.</figcaption></figure><ul><li>Customers with more than 4/5 touchpoints are highly likely to do not generate any substantial revenue. Should there be a trigger to stop paid marketing activities for those users?</li><li>Journeys with the short customer path — one, two three touchpoints are the most successful from the conversion perspective.</li></ul><p>Let’s uncover some patterns! We will analyse customer journey made out of one, two or three touchpoints. Those customer journeys accounts for almost 80% of the revenue.</p><ol><li><strong>Customer journeys made out of one touchpoint</strong></li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/857/1*_4fBhqQ6P2to2xyJY6448w.png" /><figcaption>Customer journeys made out of one touchpoint vs. conversion value</figcaption></figure><p>How to interpret above graph? It can be presented in different form:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/875/1*ayBgetVu8a4HTdEshgwGdw.png" /></figure><ul><li><em>Organic search</em> is responsible for more than 50% of revenue within one touchpoint customer journeys,</li><li><em>Organic search</em> and <em>direct </em>are responsible for almost 90% of revenue within one touchpoint customer journeys.</li></ul><p>Let’s summarise it. One touchpoint journey is responsible for 40% of the revenue. Within this length of the customer journey, organic search and direct (combined) are responsible for 90% of the revenue.</p><p>Numeric values:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/394/1*qza_jMgjv923nk7BbhOLtg.png" /></figure><p><strong>2. Customer journeys made out of two touchpoint</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/855/1*MDmXxqFhF0WTpMKdlKHKwQ.png" /><figcaption>Customer journeys made out of two touchpoint vs. conversion value</figcaption></figure><p>The chart above shows 30+ permutations (9 distinctive channels, two channels to conversions). <a href="http://stattrek.com/online-calculator/combinations-permutations.aspx">You can calculate all permutations using this tool.</a></p><p>Let’s take a look at the cumulative conversion value of the journeys:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/370/1*Y5rQJsfk9gsqMFr9QdWr1Q.png" /></figure><p>The first two permutations:</p><ul><li>Organic search &gt; Direct</li><li>Direct &gt; Direct</li></ul><p>makes up almost 90% of all conversion values for the two touch point journeys. Can we continue the discover going through one, two, three, four, touch point journeys. Chart below shows customer journeys made out of three touchpoints with corresponding conversion value (sum).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/803/1*RTMGXKs9SmhvnSZ6kphZkg.png" /><figcaption>Customer journeys made our of three touchpoints</figcaption></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/802/1*DShfA-3Z4WAVJTuVwj_R1w.png" /><figcaption>Customer journeys made out of four touchpoints</figcaption></figure><p>Is there a different approach? Let’s take a look at a quicker way to find the most valuable journeys without examining all lengths of the paths to conversion. I have setup a cut off point at the certain conversion value to restrict number of paths shown.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/777/1*4SoksHzXFl6o0KvxkRA5Ag.png" /></figure><p>Let us take a look at he cumulative distribution:</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1cb3d54b971c" width="1" height="1" alt=""><hr><p><a href="https://medium.com/datashop/data-science-in-ecommerce-part-5-1cb3d54b971c">Data Science in eCommerce — Part 5</a> was originally published in <a href="https://medium.com/datashop">DataShop</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Data Science in eCommerce — Part 4]]></title>
            <link>https://medium.com/datashop/data-science-in-ecommerce-part-4-79eee0f5a0?source=rss-e5e269bf081d------2</link>
            <guid isPermaLink="false">https://medium.com/p/79eee0f5a0</guid>
            <category><![CDATA[ecommerce]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[customer-experience]]></category>
            <dc:creator><![CDATA[Maciej Piwoni]]></dc:creator>
            <pubDate>Fri, 05 May 2017 07:35:54 GMT</pubDate>
            <atom:updated>2017-05-05T07:40:57.167Z</atom:updated>
            <content:encoded><![CDATA[<h3>Deep dive into customer journey analysis</h3><p>In this part we will perform analysis to understand customer journey to conversion. We will uncover paths that brings the most valuable conversions. We will also analyse the impact of different marketing channels on the customer journey and conversion value.</p><p>So far we learnt two things: both conversion value and path length are not <a href="https://en.wikipedia.org/wiki/Normal_distribution">normally distributed</a>. In fact both follows <a href="https://en.wikipedia.org/wiki/Pareto_principle">Pareto distribution</a>. This will be our starting point for the analysis.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/923/1*DdV64MGnrGjBNbcJ2KLaUQ.png" /></figure><p>We will try to answer some questions:</p><ol><li><strong>How conversion value changes depending on the length of customer journey?</strong></li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/784/1*ETQF_lk71_Xa0xaSxc3m1A.png" /><figcaption><strong>Conversion value: </strong>Cumulative conversion value versus length of customer journey</figcaption></figure><p><em>Shorter customer journeys generated higher conversion value</em></p><p><strong>2. How number of conversions changes depending on the length of customer journey?</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/880/1*97kVPmss0_Uzdm4HM9kQKg.png" /><figcaption><strong>Number of conversions:</strong> Cumulative number of conversions versus length of customer journey</figcaption></figure><p><em>Takeout: Shorter customer journeys generated more conversions</em></p><p>Above two charts can be nicely summarised by this diagram.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/706/1*1_Nxu6-fg9QWKMQeuu_LMQ.png" /></figure><p>Note outliers on above diagram— check the previous part to learn how outliers skewed statistical summary of data. Visual analysis helps to identify outliers in the dataset.</p><p><strong>3. Customer journeys deep dive</strong></p><p>So far we uncovered the relationship between length of the customer journey and the value and the number of conversions. Couple of questions remain:</p><p>What is the actual volume of our customer journeys?</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/895/1*-CVM7lDPJtus7hGQPivxOw.png" /><figcaption>Distribution of the length of customer journey</figcaption></figure><p>Above diagram shows number of customer journey for each type of path length. Let’s zoom in:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/895/1*RweBhTlbvtdUtv3SkErqrw.png" /><figcaption>Distribution of the length of customer journey</figcaption></figure><p>Takeout: Majority of the customers had a customer journey between 4 and 8 touchpoints.</p><p><strong>4. Customer journeys deep dive — Grand finale<br></strong>Let’s combine our findings together: map journey length histogram with impact on conversion value</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/902/1*eL9xIHXQ48b8CZSOkOOLoA.png" /><figcaption>Combining journey length histogram with impact on conversion value</figcaption></figure><p>Takeouts:</p><ul><li>Disparity between customer journeys that brings the highest conversion value (one touchpoint) and the most common customer journeys (five touchpoint).</li><li>Customer journey with single touch points generated the highest conversions value (40% of all conversion value). On other hand this is a small cohort of the overall number of customer journeys (diagram on the top).</li><li>The most common types of journey are made out of 5 and 6 touchpoints. Unfortunately, this type of paths yields relatively low conversion value (cumulative).</li></ul><p>Above takeouts could be nicely illustrated be the graph below:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/883/1*2KzmMkvUMiUJYxW7IaShWQ.png" /></figure><p>It shows cumulative conversion value (percentage) growing with each channel.</p><p>Some very important questions for the business:</p><ul><li>Customer journey with one, two and three touchpoints are responsible together for almost 80% of all conversion value.</li><li>Customer journeys with up to 5 touchpoints generate 90% of all conversion value</li></ul><p>What we can learn from the optimisation perspective?</p><ul><li>Customers with more than 4/5 touchpoints are highly likely to do not generate any substantial revenue. Should there be a trigger to stop paid marketing activities for those users?</li><li>What is the channel makeup for the most successful journeys? Journeys with the short customer path — one, two three touchpoints?</li></ul><p>We will find answers to the above two questions in the next part:</p><p><a href="https://medium.com/datashop/data-science-in-ecommerce-part-5-1cb3d54b971c"><strong>Part 5: Finding patterns</strong></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=79eee0f5a0" width="1" height="1" alt=""><hr><p><a href="https://medium.com/datashop/data-science-in-ecommerce-part-4-79eee0f5a0">Data Science in eCommerce — Part 4</a> was originally published in <a href="https://medium.com/datashop">DataShop</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Data Science in eCommerce — Part 3]]></title>
            <link>https://medium.com/datashop/data-science-in-ecommerce-part-3-c61d250de5bf?source=rss-e5e269bf081d------2</link>
            <guid isPermaLink="false">https://medium.com/p/c61d250de5bf</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[statistics]]></category>
            <category><![CDATA[ecommerce]]></category>
            <dc:creator><![CDATA[Maciej Piwoni]]></dc:creator>
            <pubDate>Fri, 05 May 2017 07:35:23 GMT</pubDate>
            <atom:updated>2017-05-05T07:40:37.743Z</atom:updated>
            <content:encoded><![CDATA[<h3>Summary Statistics</h3><p>Let’s take a closer look at the summary statistics of out transformed data set. It will reveal some interesting information:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*ow5gvrMdbVPGWv7ObrSUMg.png" /><figcaption>Summary statistics</figcaption></figure><p>In this article, we will focus on the two variables: conversions and path length.</p><ol><li>Conversions<br>It has a mean of 9.8 but a standard deviation of 125.8. This looks quite suspicious. Box plot confirms distribution of the values — majority of the observations has number of conversions equal to 1 or 2 (read more about quartiles and <a href="https://en.wikipedia.org/wiki/Box_plot">box plot</a>).</li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/634/1*sBgn1nqXRiq3joiCQUU2Qg.png" /></figure><p>Number of observations (1,851) and the maximum value in the set (4,062) provides some clues. Let’s visualise this data:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*e3qed-HvLG0tTh6zq8Cq4Q.png" /><figcaption>Number of conversions (x axis — conversions, y axis — path to conversion)</figcaption></figure><p>Look at the distribution of the values points into <a href="https://en.wikipedia.org/wiki/Pareto_distribution">Pareto distribution</a>. It gives us a hint about the use of the <a href="https://en.wikipedia.org/wiki/Pareto_principle">80/20 rule</a> know as <a href="https://en.wikipedia.org/wiki/Pareto_principle">Pareto principle.</a> In above case we can translate into the following statement: <strong><em>‘Majority of the conversions comes from the limited number of customer paths’.</em></strong><em><br></em>Definitely worth a further exploration.</p><p>2. Path Length<br>It has a mean of 7.6 and standard deviation of 6.1. Maximum value of 161 may indicate outliers. Box plot and distribution plot will help to understand distribution.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/573/1*a3H-So1Pkf96RhyRITgcOw.png" /></figure><p>Again, we have a case of Pareto distribution. More properties of this type of distribution will be shown in the next part.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*KUet_wZU2GXII5PY0-VaxA.png" /><figcaption>(x axis — number of touchpoints to conversion, y axis — path to conversion)</figcaption></figure><p>Some business takeouts:</p><ul><li>Disparity in the number of conversions: some paths had 4,062 conversions while 75% of all paths had only up to two conversions. We may want to take a closer look at those outliers to understand where bulk of conversions takes place.</li><li>50% of observations has a path length equal or shorter to 6 touchpoints,</li><li>75% of observations has a path length equal or shorter to 9 touchpoints,</li><li>There some some outliers skewing the summary statistics — observations with up to 161 touchpoints to conversion.</li></ul><p><a href="https://medium.com/datashop/data-science-in-ecommerce-part-4-79eee0f5a0"><strong>Deep dive into customer journey analysis in Part 4</strong></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c61d250de5bf" width="1" height="1" alt=""><hr><p><a href="https://medium.com/datashop/data-science-in-ecommerce-part-3-c61d250de5bf">Data Science in eCommerce — Part 3</a> was originally published in <a href="https://medium.com/datashop">DataShop</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Data Science in eCommerce — Part 2]]></title>
            <link>https://medium.com/datashop/data-science-in-ecommerce-part-2-b3d4dce47103?source=rss-e5e269bf081d------2</link>
            <guid isPermaLink="false">https://medium.com/p/b3d4dce47103</guid>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[python]]></category>
            <category><![CDATA[ecommerce]]></category>
            <dc:creator><![CDATA[Maciej Piwoni]]></dc:creator>
            <pubDate>Fri, 05 May 2017 07:34:46 GMT</pubDate>
            <atom:updated>2017-05-05T07:40:16.386Z</atom:updated>
            <content:encoded><![CDATA[<h3>Data Preparation</h3><p>In previous article, we started with customer journey dataset that looks as this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/706/1*wdU2BkhgzdCBIRXtHuQBew.png" /><figcaption>Multi channel funnel (MCF) report in Google Analytics</figcaption></figure><p>Google analytics allows to extract MCF data in csv format.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/543/1*VMCfqTAXqDjJzJcNJFfgSw.png" /><figcaption>Original three columns MCF dataset</figcaption></figure><p>There is couple of problems with above data set:</p><ul><li>Path is saved as a text (string). It is not the most useful format for the analysis,</li><li>We are missing couple of additional information like path length.</li></ul><p>We will transform above data to more useful format. Using Python and Pandas we can expand our variables to make it easier for analysis:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/916/1*Cb-lIWIPXkeOT94WjO2JlQ.png" /><figcaption>Adding additional variables</figcaption></figure><p>New data set contain number of additional variables:</p><ul><li>path length: showing number of touchpoints,</li><li>Count of touchpoints by each channel: example — 1x direct, 2x social,</li><li>flag if the particular channel was the last touchpoint (example: paid media as the last touchpoint before the conversion).</li><li>Average conversion value (calculated as sum of conversion values divided by the number of conversions)</li></ul><p>Above transformation we increased number of columns from three to 24.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/437/1*evbEoHsMGIvdQk1UIItFKQ.png" /></figure><p>Python script automatically identified all unique marketing channels that were present in customer journeys and created appropriate columns:</p><pre>&#39;Direct&#39;, &#39;Affiliates&#39;, &#39;Email&#39;, &#39;Organic_Search&#39;, &#39;Paid_Search&#39;, &#39;Other&#39;, &#39;Social&#39;, &#39;Referral&#39;, &#39;Display&#39;</pre><p>Click here for the python script used for data load and transformation.</p><p>We can take a look at the summary statistics for our transformed data set:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/813/1*upEIDOCyfAyJ3wlhLukc9w.png" /><figcaption>Summary statistics</figcaption></figure><p><a href="https://medium.com/datashop/data-science-in-ecommerce-part-3-c61d250de5bf"><strong>Learn how to interpret summary statistics in Part 3 — Summary Statistics</strong></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b3d4dce47103" width="1" height="1" alt=""><hr><p><a href="https://medium.com/datashop/data-science-in-ecommerce-part-2-b3d4dce47103">Data Science in eCommerce — Part 2</a> was originally published in <a href="https://medium.com/datashop">DataShop</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Data Science in eCommerce — Part1]]></title>
            <link>https://medium.com/datashop/data-science-in-ecommerce-part1-cc34a102e249?source=rss-e5e269bf081d------2</link>
            <guid isPermaLink="false">https://medium.com/p/cc34a102e249</guid>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[ecommerce]]></category>
            <category><![CDATA[marketing]]></category>
            <dc:creator><![CDATA[Maciej Piwoni]]></dc:creator>
            <pubDate>Fri, 05 May 2017 07:32:29 GMT</pubDate>
            <atom:updated>2017-05-05T07:39:36.571Z</atom:updated>
            <content:encoded><![CDATA[<h3>Making sense of digital touchpoints — Visual exploration</h3><p>Did you ever wonder how to optimise customer journey to get more conversions? Data science and advanced analytics will help to achieve that. In this series of articles, I am going to show how to bring an understanding of touchpoints on customer journey and its impact on the conversions.</p><p>We can start with one of the most popular online analytical tools — Google analytics. It provides eCommerce data insight called multi channels funnels report( <a href="https://support.google.com/analytics/answer/1191180?hl=en">MCF</a>). Report show unique paths to conversion. Each path is made out of individual touchpoints.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/706/1*wdU2BkhgzdCBIRXtHuQBew.png" /><figcaption>Multi channel funnels report for eCommerce</figcaption></figure><p>For each path we can see corresponding number of conversions and conversion value. Report is based on the default channel grouping set for the account (<a href="https://support.google.com/analytics/answer/3297892?hl=en">read more about default channel grouping</a>).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/548/1*GRR_oFW6ndh-kMomOS-vgw.png" /><figcaption>MCF Report — Data extract</figcaption></figure><p>Above reports provides good starting point for the visual analysis. Let’s plot all all unique paths with corresponding conversion value:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/764/1*GQgUxgcyNrn2tShMi-2h0Q.png" /></figure><p>Plotting all values on the single diagram shows distribution of conversion value against all unique paths. In this case, we have a high number of unique paths to conversion (1,800+) — axis x.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/772/1*ph8QpA5WLCTQ_PhAMoAMEQ.png" /></figure><p>Only small subset of all paths produces meaningful results. In this case, we set a threshold for viable conversion value at $1,000. Setting threshold will help us with the analysis. Conversion value is a cumulative number made out of individual conversion over certain period of time.</p><p>Let’s zoom in into the area of the conversion.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*1ojn7DvAyu8Bj1dMTI9QfQ.png" /><figcaption>Let’s zoom in — only small subset of paths produces ‘meaningful’ conversion values</figcaption></figure><p>There is couple of interesting take outs and questions we can asked:</p><ul><li>We have more than 1,800 unique paths to conversions but only a relatively small subset is producing conversions. Can we identify them?</li><li>Is there a pattern in the paths that produces the highest conversions?</li><li>Impact on conversions between short path and the long path?</li><li>What is the impact paid media, social, email marketing activities? Do they increase conversions, shorten the path or maybe are ‘killing’ the conversion value?</li></ul><p>To answer above questions, we will have to transform our data.</p><p><a href="https://medium.com/datashop/data-science-in-ecommerce-part-2-b3d4dce47103"><strong>Learn how to transform data in Part 2 — Data Preparation</strong></a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=cc34a102e249" width="1" height="1" alt=""><hr><p><a href="https://medium.com/datashop/data-science-in-ecommerce-part1-cc34a102e249">Data Science in eCommerce — Part1</a> was originally published in <a href="https://medium.com/datashop">DataShop</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[eCommerce: How to use statistics to increase sales?]]></title>
            <link>https://medium.com/data-science/ecommerce-how-to-use-statistics-to-increase-sales-1830cf7eb684?source=rss-e5e269bf081d------2</link>
            <guid isPermaLink="false">https://medium.com/p/1830cf7eb684</guid>
            <category><![CDATA[clustering]]></category>
            <category><![CDATA[towards-data-science]]></category>
            <category><![CDATA[ecommerce]]></category>
            <category><![CDATA[analytics]]></category>
            <category><![CDATA[data-science]]></category>
            <dc:creator><![CDATA[Maciej Piwoni]]></dc:creator>
            <pubDate>Sun, 12 Mar 2017 13:49:39 GMT</pubDate>
            <atom:updated>2021-04-10T22:27:10.753Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*UCglB2_LtcSLaxJx" /><figcaption>Photo by <a href="https://unsplash.com/@benchaccounting?utm_source=medium&amp;utm_medium=referral">Bench Accounting</a> on <a href="https://unsplash.com/?utm_source=medium&amp;utm_medium=referral">Unsplash</a></figcaption></figure><p>Case of regression analysis and clustering for sales analysis.</p><p>In the previous part we took a look at the performance of eCommerce store. We started with the combination of the vanity KPI — sessions and a reasonably good business KPI — transactions. It has quite interesting relationship.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/641/1*FLcHxlAHO_erU6l_tvrSjw.png" /><figcaption>Number of sessions (horizontal axis, blurred) by number of transactions (vertical axis)</figcaption></figure><p>How to make sense of the data and understand the relationship? In previous part we explored correlation which lead to even more questions. Next step will involve the use of two statistical methods: regression analysis and clustering.</p><p><strong>Predicting sales — regression analysis<br></strong>In statistical modelling, <strong>regression analysis</strong> is a statistical process for estimating the relationships among variables. This is exactly what we are looking for.</p><p>We will use Python and stat models library to build a model to predict the number of transactions based on the number of sessions.</p><p>The outcome of the model can be represented as a straight line (linear regression).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/701/1*OBxpwr2k9AFWU0hb8yn3qw.png" /><figcaption>Predicting number of transactions based on number of sessions</figcaption></figure><p>There are two immediate challenges with our model:<br>1. It doesn’t make sense logically. It fails to explain scenario when there is no or very low number of sessions. If sessions equals zero model predicts four transactions (Intercept has value of 3.99).<br>2. <a href="http://blog.minitab.com/blog/adventures-in-statistics-2/regression-analysis-how-do-i-interpret-r-squared-and-assess-the-goodness-of-fit">R-squared value is 0.232</a>. Our model explains <em>only 23%</em> of the variability of the data. It leaves 77% of the data unexplained which is not acceptable from the business perspective.</p><p>Let’s try to tweak our model to accommodate scenarios when number of sessions is low. We will force the intercept to be equal zero (zero sessions means zero transactions):</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/591/1*Q2x5_8ZIsH0C1Ostv_lthA.png" /></figure><p>New model makes more sense logically but is <em>significantly worse. </em>It explains only <em>13% of variability</em> of the data which is disappointing. We can always try to improve r-squared parameter by using different regression analysis methods like <a href="https://en.wikipedia.org/wiki/Lasso_(statistics)">lasso</a> or <a href="https://en.wikipedia.org/wiki/Stochastic_gradient_descent">gradient descent</a> instead of least squares) but it is outside the scope of this article. We will explore another interesting method instead.</p><p><strong>Understanding audience — cluster analysis<br></strong>Cluster analysis or clustering is the task of grouping a set of objects in such a way that objects in the same group (called a cluster) are more similar (in some sense or another) to each other than to those in other groups (clusters).</p><p>Why this is important for eCommerce? There is no average user or customer. Instead we can try to distinguish different groups (clusters) who will have some things in common.</p><p>We illustrate the idea of clusters by dividing our dataset in a couple of different groups:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/978/1*IB_VFYSEVmxNMt12fj2UQQ.png" /><figcaption>Examples of clusters</figcaption></figure><p>We can use a statistical package like SPSS or BigML to do the number crunching for us.</p><p>In the example below we took two variables: number of sessions and number of transactions. Number of clusters were set to four. Each circle represents single cluster, size represents number of instances within.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/491/1*spOFwDV4wvSQNgpG_iXT0g.png" /><figcaption>Diagram showing distance between clusters — level of similarity.</figcaption></figure><p>In this simplified example we have finished up with four clusters. While it won’t provide a wealth of insights it will help to understand idea of clustering.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*VNLjKOHVwKdSHABUe2gfag.png" /></figure><p>Above four clusters may explain our audience better than regression analysis. For example cluster 2 (very high transactions number) is made out of outliers not captured by regression analysis.</p><p>Each cluster is described by number of parameters: number of instances (observations), central point for transactions and sessions. All instances in the cluster relates to his central point. Distance from this point is shown in the histogram.</p><p><strong>Getting insights from cluster analysis. Real life scenarios.</strong></p><p>Having industry knowledge helps. Good practice will dictate including more relevant data into analysis. Where should we start? We may want to consider the following variables:<br>-Customer acquisition method (paid, email, social etc.),<br>-Transaction details (like product category)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/885/1*vHZ1Ldf4fLmEwynR4uTMng.png" /><figcaption>Adding more variables may brings better insights</figcaption></figure><p>Cluster analysis is a fine balancing act. Our aim is to get a better understanding of eCommerce audience while we try to keep number of clusters low and make sure they make business sense.</p><p><em>Next: Can we predict future sales based on the past performance?</em></p><p><a href="https://medium.com/@maciej.piwoni/demystifying-vanity-kpis-e59557b41ab1#.47vq84xnq">Back to part 1 — Demystifying Vanity KPIs</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1830cf7eb684" width="1" height="1" alt=""><hr><p><a href="https://medium.com/data-science/ecommerce-how-to-use-statistics-to-increase-sales-1830cf7eb684">eCommerce: How to use statistics to increase sales?</a> was originally published in <a href="https://medium.com/data-science">TDS Archive</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Demystifying Vanity KPIs]]></title>
            <link>https://medium.com/datashop/demystifying-vanity-kpis-e59557b41ab1?source=rss-e5e269bf081d------2</link>
            <guid isPermaLink="false">https://medium.com/p/e59557b41ab1</guid>
            <category><![CDATA[ecommerce]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[data]]></category>
            <dc:creator><![CDATA[Maciej Piwoni]]></dc:creator>
            <pubDate>Sun, 12 Mar 2017 13:49:19 GMT</pubDate>
            <atom:updated>2017-03-12T21:49:00.713Z</atom:updated>
            <content:encoded><![CDATA[<h4>Online store visitors vs sales</h4><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2DD4ncFkvl3WHyTUIkC3VQ.png" /><figcaption>Photo by <a href="https://www.flickr.com/photos/dbakr/">Damián Bakarcic</a> / flickr (Creative Commons 2.0 license)</figcaption></figure><p>One of the most common KPIs used to measure online performance is a number of sessions. Usually quite useful to gauge high-level view of incoming traffic. A quite typical scenario is for the business to stop at measuring sessions and use it as an indicator of ‘success’. This is a place when common assumption goes as follow “more people in — more sales”.</p><p>Let’s verify this assumption using statistics. We will test it against quantifiable business KPI — number of transactions in eCommerce.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*byjgRjPe5UGkzQx3OjqErA.png" /><figcaption>Number of sessions over time (actual values blurred)</figcaption></figure><p>If “more people in — more sales” assumption is correct we would expect the correlation between sessions and sales. Something like this:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/212/1*vZZrw0yF9AIz_SzbRZKLfg.png" /><figcaption>Do more sessions lead to more sales?</figcaption></figure><p>Let’s take a look at the real data from the eCommerce store:<br>1. horizontal axis represents sessions,<br>2. vertical axis represents transaction,<br>3. each dot represents transaction. Its colour represents transaction value.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/641/1*FLcHxlAHO_erU6l_tvrSjw.png" /><figcaption>A number of sessions (horizontal axis, blurred) by the number of transactions (vertical axis)</figcaption></figure><p>There is no simple intuitive relationship between a number of sessions and number of transactions. A couple of takeouts, though:</p><ol><li>Increasing number of sessions beyond certain point does not increase the number of transactions (right hand side)</li><li>Low number of sessions up to certain level which seems to produce up to 10 transactions per session but it is very inconsistent,</li><li>There is another group of sessions /transactions (middle of the chart) seems to produce higher number of session but again it is very inconsistent. Perhaps there is a number of other factors that have an impact (acquisition channel, day of the week, user device?)</li></ol><p>Is there a more <em>scientific</em> way to determine if there is or there isn’t linear relationship between sessions and transactions?</p><p><strong>Gather more data<br></strong>Is there more information available? In our case, we quickly discover that we have not only sessions and transactions but also transaction values and number of users visiting the online store. This looks like a promising start!</p><p><strong>Correlation analysis<br></strong>Start small with correlation analysis. This is a method of statistical evaluation used to study the strength of a relationship between two, numerically measured, continuous variables. It is important to remember that correlation analysis does not determine cause and effect.</p><p>How do we measure correlation? This diagram will be helpful:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/646/1*UJVrE7jj8d4VcxI1hOmEwA.png" /><figcaption>Strong negative (-1) vs strong positive (+1) correlation</figcaption></figure><p>Let’s take a look at our data using correlation methods in Python:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/668/1*Sg_iBbWuXBKPoDknfPZrIQ.png" /><figcaption>The strength of the relationship between variables</figcaption></figure><p>Note strong positive association between the number of sessions and the number of users (+0.99) and medium positive between number of sessions and transactions (+0.48).</p><p>How should we interpret the result?<br>This is the first step towards getting an understanding of what drives performance of the online store. The positive relationship between sessions and transactions is worth further investigation! Another interesting point is the relationship between the number of transactions and the revenue</p><p>What questions should we ask based on above findings?<br>Imagine a scenario when price reduction campaign triggers more visitors to the online store, reduced prices results in more transactions. Sessions and transaction will have positive correlation but causation agent is a campaign.</p><p>Keep in mind that correlation does not imply causation.</p><p><strong>One step further — Pair plot<br></strong>Another way useful way of looking into the data is a pair plot. In our case, it shows every possible pair of the variables we have. Similar to correlation plot but with much more details. We also get an additional piece of information for each variable: histogram. It shows the distribution of the values.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/735/1*IEmXQ98zlGSumulhGG9SyA.png" /><figcaption>Pair plot showing the relationship between variables</figcaption></figure><p>Compare the strength of the relationship with visual representation. In this case number sessions, users and transactions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/934/1*upFxOBz7Cy16Ppb5pzHeOg.png" /></figure><p>Sessions and users show almost perfect linear relationship. No surprise — it’s correlation value is +0.99. Transactions and sessions are more challenging to interpret. Could it be done? Absolutely!</p><p><em>Read more about advanced analysis and predicting sales value in the next part.</em></p><p><strong>Bonus Question: Does more transactions mean more revenue?</strong></p><p>Intuitively many of us would say yes. In our case, the short answer is — it depends on the particular case. Pair plot below shows the relationship between transactions and revenue.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/448/1*xF4sUxx6sYvmrXlGv-wPFA.png" /><figcaption>Another interesting relationship: transactions vs transaction revenue</figcaption></figure><p>We can create the following hypothesis: 1 to 15 transactions per day will result in a linear relationship ( the increase of 1 transaction will lead to increase of x units in revenue). Beyond the value of 15 transactions per day, this relationship becomes much more volatile.</p><p>Why? Again, we can verify a couple of assumptions:<br>- A Higher number of transactions may be driven by diverse marketing tactics bringing a mixed type of clients. This will result in more diverse transactions,<br>-Coupons and promotions are used to drive extra sales. A number of transactions increases but transaction values is lower than usual.<br>Read more about advan</p><p><em>Read more about advanced analysis and predicting sales value in the next part.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e59557b41ab1" width="1" height="1" alt=""><hr><p><a href="https://medium.com/datashop/demystifying-vanity-kpis-e59557b41ab1">Demystifying Vanity KPIs</a> was originally published in <a href="https://medium.com/datashop">DataShop</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[We are moving]]></title>
            <link>https://medium.com/@maciej.piwoni/we-are-moving-1bc04143abd1?source=rss-e5e269bf081d------2</link>
            <guid isPermaLink="false">https://medium.com/p/1bc04143abd1</guid>
            <dc:creator><![CDATA[Maciej Piwoni]]></dc:creator>
            <pubDate>Wed, 28 Dec 2016 10:32:39 GMT</pubDate>
            <atom:updated>2016-12-28T10:32:39.559Z</atom:updated>
            <content:encoded><![CDATA[<p>Dear Follower,</p><p>Please note I have changed the name of my Medium blog to Datashop.</p><p>You can now access it at <a href="https://medium.com/datashop">https://medium.com/datashop</a></p><p>Thank you,</p><p>Maciej</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1bc04143abd1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Holy Grail of Online Marketing]]></title>
            <link>https://medium.com/datashop/the-holy-grail-of-online-marketing-5d8028d266b8?source=rss-e5e269bf081d------2</link>
            <guid isPermaLink="false">https://medium.com/p/5d8028d266b8</guid>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[econometrics]]></category>
            <category><![CDATA[statistics]]></category>
            <category><![CDATA[analytics]]></category>
            <dc:creator><![CDATA[Maciej Piwoni]]></dc:creator>
            <pubDate>Wed, 05 Oct 2016 10:27:18 GMT</pubDate>
            <atom:updated>2016-10-05T10:27:18.865Z</atom:updated>
            <content:encoded><![CDATA[<p>I just came across this excellent presentation by Prof. Dr. Herman A. Bruno —<a href="https://image-store.slidesharecdn.com/d587a9e8-82a1-40e3-ac06-4401b486b3af-original.png"> “How to innovate with data analytics in retail”</a>. It has a very good summary of 5 steps of making the most of your data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BZcPGsOMh8ycDebb9w8tVQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*xwpFDfhmWDSklFWnBgMfDA.png" /></figure><p>Take a closer look at the step no. 5 — counterfactuals. The Holy Grail question of the marketing.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/716/1*Hw_MwTzp0PXs83G00GPzPA.png" /></figure><p>Short term experiments are usually well mastered by marketing teams. This include controlled experiments, A/B or MVT testing. But the bigger question remains — what is the impact on the company bottom line? Revenue, new customers, items sold. Medium term effect is notoriously difficult to measure.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_idBKWREi4O8a7Z6bFLbOw.png" /><figcaption>Short term approach</figcaption></figure><p>Medium term impact of the advertising campaign is notoriously difficult to measure. Impact on the revenue, new customers unit sold is one of the key business indicators.</p><p>One of the approach to solve it are <a href="https://en.wikipedia.org/wiki/Econometric_model">econometric models</a>. While providing some insight they are usually difficult to implement and master.</p><p>One of the alternatives it to approach challenge of measurement and what-if scenarios from different angle.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*GR7gyc7qWIT8WEYQv502Ug.png" /><figcaption>Medium term approach</figcaption></figure><p>Going beyond econometric models: Bayesian structural time-series models</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*LjPA_Iukwrj4sQF8Khx4Cg.png" /></figure><blockquote>This paper proposes to infer causal impact on the basis of a diffusion-regression state-space model that predicts the counterfactual market response that would have occurred had no intervention taken place.</blockquote><p><a href="http://static.googleusercontent.com/media/research.google.com/en//pubs/archive/41854.pdf">Inferring causal impact using bayesian structural time-series models (PDF).</a></p><p><strong>Implementation in R</strong></p><p><a href="https://google.github.io/CausalImpact/CausalImpact.html">An R package for causal inference using Bayesian structural time-series models</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/671/1*ibm5TTIZzYoiswAiIQMekg.png" /></figure><p><a href="https://google.github.io/CausalImpact/CausalImpact.html">CausalImpact</a></p><blockquote>By default, the plot contains three panels. The first panel shows the data and a counterfactual prediction for the post-treatment period. The second panel shows the difference between observed data and counterfactual predictions. This is the <em>pointwise</em> causal effect, as estimated by the model. The third panel adds up the pointwise contributions from the second panel, resulting in a plot of the <em>cumulative</em> effect of the intervention.</blockquote><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=5d8028d266b8" width="1" height="1" alt=""><hr><p><a href="https://medium.com/datashop/the-holy-grail-of-online-marketing-5d8028d266b8">The Holy Grail of Online Marketing</a> was originally published in <a href="https://medium.com/datashop">DataShop</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
    </channel>
</rss>