<?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 Nidhi DeSutter on Medium]]></title>
        <description><![CDATA[Stories by Nidhi DeSutter on Medium]]></description>
        <link>https://medium.com/@nidhisd3?source=rss-b688adbe474c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*ZWYB_QC27pc_zohMGRcsnA.png</url>
            <title>Stories by Nidhi DeSutter on Medium</title>
            <link>https://medium.com/@nidhisd3?source=rss-b688adbe474c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 01:12:24 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@nidhisd3/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[The Business Case for Retail Media]]></title>
            <link>https://medium.com/@nidhisd3/the-business-case-for-retail-media-4dd1329d5e23?source=rss-b688adbe474c------2</link>
            <guid isPermaLink="false">https://medium.com/p/4dd1329d5e23</guid>
            <dc:creator><![CDATA[Nidhi DeSutter]]></dc:creator>
            <pubDate>Sat, 04 Oct 2025 21:49:17 GMT</pubDate>
            <atom:updated>2025-10-07T14:43:00.508Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*_ZV4JHJsNaR_0vFDyXHAdg.png" /><figcaption><em>Adweek — Retail Media 101: What It Is, How It Works, and Why It Matters</em></figcaption></figure><p>Driven by the convergence of e-commerce growth, declining cookie-based targeting, and the need for measurable performance, <strong>Retail Media</strong> offers brands a powerful way to reach<strong> high-intent shoppers</strong> at the moment of decision. It blends advertising and commerce into one experience. It is no surprise that Retail Media is rapidly becoming one of the most important channels in <strong>digital advertising</strong>.</p><p><strong>What is Retail Media?</strong></p><p>Retail media is a form of advertising where brands purchase ad placements on a retailer’s owned platforms, such as their website, app, or physical stores, to reach shoppers directly at the <strong>point of purchase</strong>. It allows brands to target shoppers who are already in the funnel and have a higher affinity to purchase.</p><p>Retailers that operate Retail Media Networks (RMNs) — like Walmart, Target, DoorDash, Amazon, Instacart, and Best Buy — effectively function as both publishers and ad networks, leveraging their owned media and first-party shopper data to connect online and offline experiences for advertisers.</p><p><strong>Types of Retail Media:</strong></p><p><strong>Onsite</strong> — Ads that appear directly on the retailer’s owned digital properties, such as their website or mobile app. Examples include:</p><ol><li>Sponsored product ads (search, category pages)</li><li>Display banners (homepage, cart/checkout pages)</li><li>Video placements</li><li>Native brand content (articles)</li></ol><p><strong>Offsite</strong> — Leveraging retailer’s first-party data to target audiences outside of their own platforms — across the open web, social, or CTV. Examples include:</p><ol><li>Programmatic display ads</li><li>Video ads (including CTV) —<em> this could include Shoppable TV ads with QR codes or links.</em></li><li>Social media ads (via platforms like Meta, TikTok, Pinterest)</li><li>Digital out-of-home (DOOH) using retail data</li></ol><p><strong>In-Store</strong> — Advertising within the physical retail environment. Examples include:</p><ol><li>Digital signage and smart displays</li><li>Audio ads via in-store radio</li><li>Shelf Displays</li><li>Interactive Kiosks</li></ol><p><strong>What is the Business Opportunity for the Ecosystem players?</strong></p><p>According to <a href="https://www.mckinsey.com/capabilities/growth-marketing-and-sales/our-insights/commerce-media-the-new-force-transforming-advertising">McKinsey</a>, in the United States, the growth of RMNs could represent as much as $100 billion in ad spending by 2026. With the overall operating margins of RMNs in the 50 to 70 percent range and over $1.3 trillion of enterprise value (EV) at stake.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IhUNYIDAZhDgQA1rCYXm5g.png" /></figure><p>In addition, based on research done by <a href="https://www.globalgrowthinsights.com/market-reports/retail-media-networks-market-116763">GGI</a>, <a href="https://www.forbes.com/sites/andreawasserman/2024/10/21/retail-media-networks-present-and-future/">Forbes</a>, <a href="https://www.emarketer.com/content/off-site-retail-media-ad-spend-growing-much-faster-than-on-site">Emarketer</a>, ~80% of revenue today is from digital sources. However, in-store is a growing option for retailers without large online presences. <em>*The %’s provided below for sources of revenue breakdown are directional based on reported analysis.</em></p><ol><li>Onsite — ~50‑65%. Steady growth, though slower than offsite. High margins. It remains the foundation of RMNs due to direct shopper intent and higher conversion. Supply (inventory) is a constraint in some cases.</li><li>Offsite — ~25‑35%. The fastest growing segment. More appeal to non‑endemic advertisers. Offers scale and reach beyond onsite inventory constraints. Margins are lower (media cost, agency fees, etc) but clear trend toward growth.</li><li>In-store — ~5%. Some estimates put U.S. in‑store spend at $500M+ in 2025; still relatively modest compared to onsite/offsite. Retailers are experimenting with digital signage, smart surfaces, immersive in‑store media. Omni-channel integration is a key trend.</li><li>Data / Audience Licensing / Insights — ~5‑15%. Less often reported separately. However, increasing in importance, especially for offsite / programmatic channels. There’s also revenue from analytics, measurement, shopper insights bundles.</li></ol><p><strong>Though, challenges still remain for advertisers and retailers to operate at scale —</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ICXaUEnz7ykCppq24zXObA.png" /></figure><p><strong>High-level: How does Databricks supports Retail Media?</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rP53KbdD0MrdbeLuzpwZVg.png" /></figure><p>Retail Media presents immense opportunities, despite the challenges outlined. These challenges are part of a fast-evolving ecosystem that continues to mature and innovate. As Brands and Retailers invest in smarter technology, better attribution models, and deeper collaborations, Retail Media is set up for continued growth. Embracing this channel not only helps companies stay relevant in a competitive marketplace but also unlocks new avenues for customer engagement, personalization, and long-term value.</p><p><em>Consider checking out these reference articles/videos for more information (</em><strong><em>not exhaustive</em></strong><em>) from Databricks and some of our partners.</em></p><p>→ <a href="https://www.youtube.com/watch?v=J98EQQQm0bk">Unlocking the power of Retail Media Networks</a></p><p>→ <a href="https://www.databricks.com/blog/search-sale-how-ai-redefining-customer-engagement-and-loyalty-retail">From Search to Sale: How AI Is Redefining Customer Engagement and Loyalty in Retail</a></p><p>→ <a href="https://www.prnewswire.com/news-releases/hightouch-partners-with-databricks-to-launch-self-service-ai-powered-audience-management-solution-for-offsite-retail-media-networks-spanning-50-media-channels-302347670.html">Hightouch Partners with Databricks to Launch Self-Service AI-powered Audience Management Solution for Offsite Retail Media Networks</a></p><p>→ <a href="https://www.youtube.com/watch?v=bEoHN4FeMvw">Accelerate your Retail Media Network &amp; Data Monetization Journey</a></p><p>→ ​​<a href="https://www.youtube.com/watch?v=vs7970uO8ok">Retail &amp; Consumer Goods Industry Forum: How AI is Transforming How Brands Connect With Consumers</a></p><p>Reach out to your account team to learn more!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4dd1329d5e23" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[One Size Doesn’t “Measure” All — Measurement 101]]></title>
            <link>https://medium.com/@nidhisd3/one-size-doesnt-measure-all-measurement-101-7e7819daf8db?source=rss-b688adbe474c------2</link>
            <guid isPermaLink="false">https://medium.com/p/7e7819daf8db</guid>
            <dc:creator><![CDATA[Nidhi DeSutter]]></dc:creator>
            <pubDate>Wed, 20 Aug 2025 17:55:50 GMT</pubDate>
            <atom:updated>2025-08-20T17:55:50.079Z</atom:updated>
            <content:encoded><![CDATA[<h3>One Size Doesn’t “Measure” All — Measurement 101</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dWSBIeMpATokJSPHHxWGzw.png" /></figure><p>In the dynamic landscape of advertising, accurate measurement is crucial for understanding campaign effectiveness and optimizing future strategies. However, the complexity of consumer behavior, diverse marketing channels, and varying business objectives mean that a single measurement solution rarely provides a comprehensive answer.</p><p>Different campaigns have different goals — brand awareness, lead generation, sales conversions, or customer retention — and each requires tailored metrics to assess success.</p><p>The diversity of channels — TV, digital, mobile, out-of-home — brings unique data challenges. Measurement tools optimized for digital platforms may struggle to capture the full impact of traditional media, where impressions and viewer engagement are harder to quantify.</p><p>Furthermore, consumer journeys today are nonlinear and multi-touch, spanning numerous devices and platforms. Attribution models vary in how they credit different touchpoints, and no single approach perfectly captures every nuance.</p><p>Thus, a flexible, multi-faceted measurement strategy that combines experimental methods like A/B testing and integrates qualitative insights along with quantitative data across channels is essential. Advertisers must select and combine multiple tools and methodologies that align with their specific objectives and audience behaviors, rather than relying on a one-size-fits-all solution.</p><ol><li><strong>Randomized Control Trial</strong></li></ol><p>A “randomized controlled trial” (RCT) in marketing refers to a research method where a group of customers are randomly assigned to either a control group (receiving the standard marketing approach or none) or an experimental group (exposed to a new marketing tactic), allowing marketers to accurately measure the effectiveness of the new strategy by comparing the results between the two groups, minimizing bias. Commonly, it’s a single channel experimentation.</p><p>Groups –</p><ul><li>Random assignment of treatment/test</li><li>Sufficiently large group, no pre-existing difference.</li></ul><p><em>Example –</em></p><ul><li>A company is running a RCT experimentation. 5% of the cohort has the probability of seeing an ad about a new version of the product (e.g. ninja blender). Of the 5% who were exposed to the ad, the company measures their purchase affinity.</li><li><strong>Business question being asked: </strong>Did it increase conversion rates?</li></ul><p><strong><em>Challenges with RCT</em></strong></p><ul><li>Certain channels can’t support it → e.g. LinearTV, Billboards</li><li>Finding a sufficiently large and representative sample population</li><li>Improper randomization techniques can lead to imbalances in important characteristics between the treatment and control groups</li></ul><p><strong>2. Marketing Mix Modeling</strong></p><p>MMM is a forecasting methodology used to estimate the impact of various marketing tactic scenarios on product sales. MMMs use statistical models, such as multivariate <a href="https://en.wikipedia.org/wiki/Linear_regression">regressions</a>, and use <a href="https://en.wikipedia.org/wiki/Sales">sales</a> and <a href="https://en.wikipedia.org/wiki/Marketing">marketing</a> <a href="https://en.wikipedia.org/wiki/Time_series">time-series</a> data. They are often used to optimize <a href="https://en.wikipedia.org/wiki/Advertising">advertising</a> mix and promotional tactics.</p><ul><li>Time series of outcomes (looking at sales, ad spend).</li><li>Regressing across revenue. It can be difficult to distinguish between correlation and causation, which can lead to misallocated budgets.</li></ul><p><strong><em>Challenges with MMM</em></strong></p><ul><li>It doesn’t accurately take into account which touchpoint actually caused a consumer to convert or if other factors were in question. For example, was the conversion due to ad spend (ad campaigns) OR seasonal (Christmas)?</li><li>Another example — companies will often measure search and attribute searches to revenue. How do you know if search is what caused the conversion? Maybe, a consumer saw an ad about a product on TV, then heard about the same product on the radio…decided to search it on google and then ultimately purchased it. A company might attribute the conversion to search when in fact, there were other touchpoints that led to the conversion.</li></ul><p><strong>3. Casual Inference</strong></p><p>Causal inference is the process of determining the effect of one variable on another. It’s a lift study/experimentation. It works by:</p><ul><li><strong>Study design</strong>: Researchers design a study to minimize bias</li><li><strong>Data selection</strong>: Researchers choose a data source</li><li><strong>Statistical modeling</strong>: Researchers use statistical methods to model the data</li><li><strong>Identify assumptions</strong>: Researchers identify any assumptions they make</li><li><strong>Calculate the effect</strong>: Researchers calculate the effect of the variable on the other</li></ul><p><strong><em>Propensity Score Modeling — Propensity score matching is a causal inference technique that attempts to balance treatment groups on confounding factors, so researchers can gauge the treatment’s causal impact on the outcome.</em></strong></p><ul><li>Build a model that predicts treatment &amp; probability of being treated (in the marketing world, this means seeing the ad).</li><li>This takes into account everything we know about the consumers (attributes, etc).</li><li>Stratify — process of dividing a study population into subgroups based on specific characteristics (like age, gender, or other relevant variables) to analyze the causal effect of a treatment within each subgroup. Note, the distribution has to overlap enough.</li><li>Enough data needs to exist to make the prediction in the first place (known exposed, known unexposed).</li></ul><p><strong><em>Challenges with Casual Inference</em></strong></p><ul><li>Confounding factors — other factors (called cofounding factors) that can influence the relationship between the variables.</li><li>Unobserved outcomes — it’s not always possible to observe all possible outcomes that can influence the relationship between the variables.</li></ul><p><strong>4. Match Market Test</strong></p><p>Matched market testing is a way to measure the impact of marketing by comparing similar markets. This is often used when you don’t know enough about the consumers or don’t have enough data needed for other measurement methods.</p><ul><li>Pre/Post tests to understand lift (sales, search, brand). For example, surveys — “Have you heard of this brand before?”</li><li>Often involves geo — cohorts based on region/country and aggregates.</li></ul><p><strong><em>Challenges with Match Market Test</em></strong></p><ul><li>Audience Leeching — The audience from one market spills over into another. For example, if you’re testing a new ad campaign in one city, but people from a nearby city also see the ads. Using synthetic controls can help, but it’s not a perfect solution.</li><li>Geographical Limitations — Not all markets are the same, and what works in one area might not work in another. Careful selection of test markets is crucial to mitigate this issue.</li></ul><p><strong>5. Multi-Touch Attribution (MTA)</strong></p><p>MTA is a marketing method that assigns credit to multiple touchpoints that contribute to a conversion. It’s a data-driven approach that helps marketers understand how effective their marketing channels are.</p><ul><li>MTA assigns values to each touchpoint in a customer’s journey.</li><li>These values are typically percentages that indicate how much each touchpoint contributed to a sale.</li><li>Marketers can use these values to allocate their budget more effectively.</li></ul><p>You can measure at the person level (via exposure data). There are different models you can leverage for MTA, such as:</p><ul><li>W-shape: This model spreads credit across the first, middle, and last touchpoints. It’s good for complex campaigns.</li><li>Time decay: This model focuses on touchpoints at the bottom of the funnel, but it also gives some credit to touchpoints in the top and middle.</li><li>Linear: This model assigns equal credit to all touchpoints.</li></ul><p><strong><em>Challenges with MTA</em></strong></p><ul><li>Privacy regulations like GDPR, ITP, and iOS 17 in Safari make it difficult to track users.</li><li>It’s not possible to track every user across every device, which can lead to less accurate attribution models.</li><li>Different vendors and platforms use different methodologies and definitions.</li><li>Unproven assumptions are often in the picture making the attribution itself less accurate.</li></ul><p><strong>6. General Attribution</strong></p><p>“Attribution measurement in marketing” refers to the process of identifying which specific marketing channels, campaigns, or touchpoints are responsible for driving conversions or sales. There are different measurement models that can be used to allocate credit for a conversion:</p><ul><li>First touch — crediting the first interaction</li><li>Last touch — crediting the final interaction</li><li>Linear — evenly distributing credit across all touchpoints</li><li>Time-Decay — giving more credit to interactions closer to the conversion</li><li>Position-based — giving more credit to interactions at key stages of the funnel.</li></ul><p><strong><em>Challenges with Attribution</em></strong></p><ul><li>Traditional models may assume that correlated events are causally linked, which can lead to overestimating the impact of some marketing activities.</li><li>There’s no one-size-fits-all solution for attribution models, so it can be hard to find the right one.</li></ul><h3><strong>How does Databricks help?</strong></h3><p>Databricks is designed to handle the complexity and scale of advertising measurement by providing a flexible platform paired with strategic partnerships in the ecosystem.</p><p><strong>1. Advanced Data Engineering and Integration</strong></p><ul><li><strong>Unified Data Lakehouse</strong>: Databricks’ Lakehouse architecture allows advertisers to consolidate structured and unstructured data — such as clickstream logs, CRM records, TV impressions, and social media interactions — into a single source of truth.</li><li><strong>Connectors and APIs</strong>: Databricks supports connectors and delta sharing to leverage data from popular ad platforms and programmatic DSPs, as well as offline data sources. This integration ensures that advertisers can continuously refresh their datasets for timely insights.</li></ul><p><strong>2. Scalable Analytics and Real-Time Processing</strong></p><ul><li><strong>Apache Spark Engine</strong>: At its core, Databricks leverages Apache Spark, a powerful distributed computing engine, to process billions of rows of advertising data rapidly. This supports near-real-time analytics, enabling marketers to track campaign KPIs and adjust strategies quickly.</li><li><strong>Delta Lake</strong>: This storage layer enhances data reliability and enables ACID transactions on streaming and batch data. For advertising, this means clean, consistent datasets that support accurate measurement and experimentation.</li></ul><p><strong>3. Experimentation Frameworks</strong></p><ul><li><strong>Built-in Experimentation Tools</strong>: Databricks supports the orchestration of A/B tests by allowing teams to run controlled experiments. Marketers can segment audiences, randomize exposure, and analyze performance metrics within the same platform.</li><li><strong>Open-Source Causal Inference Libraries</strong>: Databricks integrates with libraries like DoWhy, EconML, and CausalImpact, enabling analysts to apply rigorous causal inference techniques. These tools help estimate the true incremental impact of campaigns even when randomized tests are not possible.</li></ul><p><strong>4. Machine Learning for Attribution and Personalization</strong></p><ul><li><strong>MLflow Integration</strong>: Databricks’ MLflow makes it easier to build, track, and deploy machine learning models for advanced attribution — such as multi-touch attribution, uplift modeling, and customer lifetime value prediction. <a href="https://www.databricks.com/solutions/accelerators/multi-touch-attribution">Solution accelerator</a>.</li><li><strong>Feature Engineering and Model Training</strong>: With Databricks, data scientists can engineer features from complex multi-channel data and train models that understand nonlinear customer journeys, leading to more precise measurement of advertising ROI.</li></ul><p><strong>5. Collaboration and Governance</strong></p><ul><li><strong>Interactive Notebooks</strong>: Databricks notebooks allow cross-functional teams — data engineers, scientists, marketers, and analysts — to collaborate seamlessly, sharing code, results, and visualizations.</li><li><strong>Security and Compliance</strong>: Built-in security controls and compliance certifications help ensure that sensitive customer data is protected across the advertising measurement pipeline.</li></ul><p>To wrap this up, Databricks doesn’t operate in isolation; its power is amplified through <strong>strategic ecosystem partnerships</strong> across the advertising and data ecosystems. This includes AdTech/MarTech integrations to allow advertisers to close the loop between measurement and media planning / c360 as well as data providers to help advertisers enrich datasets with offline attribution and demographic insights, thus bridging digital and traditional media measurement. Databricks also works with other platforms that provide experimentation services out of the box that sit on top of Databricks in a composable architecture.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=7e7819daf8db" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[From Stitching to Resolution: Connecting the Dots in Customer Identity]]></title>
            <link>https://medium.com/@nidhisd3/from-stitching-to-resolution-connecting-the-dots-in-customer-identity-586bc81beacc?source=rss-b688adbe474c------2</link>
            <guid isPermaLink="false">https://medium.com/p/586bc81beacc</guid>
            <dc:creator><![CDATA[Nidhi DeSutter]]></dc:creator>
            <pubDate>Thu, 17 Jul 2025 03:34:19 GMT</pubDate>
            <atom:updated>2025-08-19T00:09:02.261Z</atom:updated>
            <content:encoded><![CDATA[<p>In today’s data-driven marketing landscape, we often talk about the need for <strong>First Party Data Stitching </strong>and<strong> Identity Resolution </strong>as a layer to build a Customer360 foundation. However, people often confuse the two because they both involve connecting data points to create an unified view of an individual. Both are fundamental and interconnected concepts but <em>have different scopes and operate at different levels of complexity.</em></p><p>Organizations ultimately need both to deliver smarter, more personalized customer experiences. Before I go into the tactical definitions, let me provide an analogy to set the stage:</p><p>Imagine you are shopping online on Brand X:</p><ul><li>You browse their site while logged in and check your order history</li><li>A week later, you buy something through their mobile app</li><li>You call customer support about a return</li></ul><p>All these actions generate different records but within Brand X’s own environment: an user ID (BrandXcustomerID), an email (steph.smith@gmail.com), a phone number (444–444–4444), maybe even an address. Brand X’s system ties all those interactions <strong>together under an unified profile.</strong></p><pre>BrandXcustomerID &lt;&gt; Steph Smith &lt;&gt; steph.smith@gmail.com &lt;&gt; 444-444-4444 &lt;&gt; 123 Main Street, Boston MA</pre><p>Now let’s say Brand X owns multiple sub brands and wants to recognize that:</p><ul><li>You clicked on their Instagram ad (cookie ID)</li><li>You bought something at a Retail store (loyaltyID)</li><li>You subscribed to their newsletter with a different email than what you use when you login into their website or mobile app</li></ul><p>Each interaction happens in a different place, often with limited shared IDs. To recognize that all these touchpoints belong to <em>you</em>, Brand X may use third party identity graphs or partnerships with data vendors. In this situation, Brand X is linking <strong>different identities across different systems and platforms</strong> to recognize you as the same person.</p><pre>BrandXcustomerID &lt;&gt; IdentityPartnerID &lt;&gt; Steph Smith &lt;&gt; Stephanie Smith &lt;&gt; steph.smith@gmail.com &lt;&gt; 444-444-4444 &lt;&gt; 123 Main Street, Boston MA &lt;&gt; stephanie.smith@yahoo.com &lt;&gt; cookieID9001 &lt;&gt; LoyaltyID678</pre><p><strong>What is First Party Data Stitching?</strong></p><p>First Party Data Stitching involves the process of linking data collected directly by an organization. This could be from mobile app, CRM, transactions, logins, etc. The focus is consolidating their known customer data within their internal systems to create an unified customer record.</p><p><strong>What is Identity Resolution?</strong></p><p>Identity Resolution refers to the process of linking various data points and identifiers associated with individuals across different systems. Identity resolution typically involves leveraging internal and external signals.</p><p>This can be done via <strong><em>deterministic</em></strong> and/or <strong><em>probabilistic</em></strong> matching techniques (more on that below).</p><ol><li>Deterministic Matching — Leverages exact matches of identifiable information such as email addresses, phone numbers, name/postal, etc. Relies on a high degree of data quality and readily available, verifiable data to establish relationships between records. <strong>Therefore, it is meant to prioritize accuracy and certainty</strong>.</li><li>Probabilistic Matching — Leverages various data points, including IP addresses, device types, browser information, operating systems, location data, etc. It relies on statistical models and algorithms to infer relationships, estimating the likelihood that they belong to the same individual. <strong>Therefore, it comes with a degree of uncertainty.</strong></li></ol><p>The question often comes down to reach vs accuracy. Probabilistic methods can often provide higher reach (more exposure) but at the cost of accuracy. In layman’s terms, I like to think of it as probabilistic = “<em>We </em><strong>think</strong><em> these data points are tied to the same individual</em>” vs deterministic = <em>“We </em><strong>have high confidence</strong><em> that these data points are tied to the same individual”</em></p><p>Let’s bring it together — First party data stitching is often the first step towards effective identity resolution. How so?</p><ol><li><em>As I mentioned, </em><strong><em>First Party stitching</em></strong> links an organization’s owned and collected data for a consistent customer view so that their internal systems are speaking the same language.</li><li><strong><em>Identity Resolution</em></strong> builds on this by resolving identities across devices, sessions, external datasets to further expand on these first party signals.</li></ol><p>To summarize:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*XiZWEFGZDEybKegybdRr1Q.png" /></figure><p>This forms the basis of a <strong>Customer 360</strong> foundation which provides a <strong>complete, centralized profile of a customer</strong> built from all the data an organization collects + third party data sources and signals.</p><p>Customer 360 is the <em>output</em> of a unified data approach, enabled by technologies like CDPs (Customer Data Platforms), CRMs, Identity resolution platforms, Data Lakes &amp; Platforms, etc.</p><p><strong>How does Databricks help?</strong></p><p>Databricks provides a <strong>Data Intelligence Platform </strong>that allows customers to to combine data engineering, AI/ML, governance, and business intelligence into a unified, intelligent system — all the layers needed to be able to stand up a scalable C360 along with the <strong>ecosystem partnerships</strong> to help accelerate. Architecture support in Databricks includes:</p><ol><li>Source &amp; Ingestion Layer</li><li>Medallion Layers — Bronze → Silver → Gold</li><li>Entity &amp; Identity Resolution — <em>Leverage ML-based approaches or composable partner solutions, both integrated into Databricks.</em></li><li>Feature Engineering, Analytics, and ML —<em>Transform features in Silver/Gold layers for predictive modeling (e.g. churn, propensity, segmentation). Use </em><strong><em>Feature Store</em></strong><em>, </em><strong><em>MLflow</em></strong><em>, and Spark to build, train, and deploy ML models directly on the platform</em></li><li>Activation &amp; Serving — <em>Expose Gold‑level C360 tables to BI tools (Power BI, Tableau, Looker), real‑time apps, and marketing systems via connectors or APIs. Optionally use </em><strong><em>Databricks Model Serving</em></strong><em> or third‑party execution layers (e.g. Mosaic AI) for low‑latency personalization.</em></li><li>Governance &amp; Security — <strong><em>Unity Catalog</em></strong><em> governs access, ensures data lineage, and enforces PII controls. Enables audibility, compliance (GDPR, CCPA), and secure sharing (e.g. Delta Sharing)</em></li></ol><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*aEawu0le9mtxRF_zcL4QBQ.png" /><figcaption>© Databricks</figcaption></figure><p>With Databricks, you can democratize data + AI across the organization to build C360 profiles at scale, keeping them updated as additional inputs/data sources flow in.</p><p>To learn more and brainstorm together, reach out!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=586bc81beacc" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[The Evolution of Customer Data Platforms]]></title>
            <link>https://medium.com/data-beats/the-evolution-of-customer-data-platforms-ae2e5c830ae2?source=rss-b688adbe474c------2</link>
            <guid isPermaLink="false">https://medium.com/p/ae2e5c830ae2</guid>
            <dc:creator><![CDATA[Nidhi DeSutter]]></dc:creator>
            <pubDate>Mon, 05 Feb 2024 14:43:08 GMT</pubDate>
            <atom:updated>2024-02-08T09:22:46.405Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/978/1*n4owT4zOKML0Gjw688fQ7Q.png" /><figcaption>Snowflake for Marketing Deck</figcaption></figure><p>The idea of bringing all of your customer data in one place, cleaning it, segmenting it, and activating against is sensible — after all, <strong><em>customer data is a key component of data-driven marketing</em></strong>. This is the logic behind the emergence of the CDP category that we see and hear of today.</p><p>The challenge with legacy constructed CDPs is that while, they accomplish the idea of becoming a data repository for customer data, they become an isolated tech stack from the rest of the data. The classic issue of <strong>multiple sources of truth aka data silos and copies of data… </strong><em>The very real problem that companies are trying to address today.</em></p><p>Not to mention, the other challenges that follow —</p><ol><li>Governance, security concerns around data that is highly sensitive (PII) and valuable for brands.</li><li>Longer TTV (time to value) — implementation is not turnkey.</li><li>Inflexibility of being able to construct the rules, segmentation, modeling needed to run dynamic activation + personalization.</li><li>Higher TCO, lower ROI over time.</li></ol><p><em>Let’s divert just a little bit and talk about TCO really quick because it’s extremely important with any technology decision you make. When you look at TCO (Total Cost of Ownership), you need to consider all three aspects — </em><strong><em>1) Acquisition Cost, 2) Usage Cost, 3) Value. </em></strong><em>Looking at just the line item (visible) costs isn’t enough and can lead to inaccuracies when making key business decisions.</em></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3V6BQdoxYvN3AbBibEINUQ.png" /><figcaption>Shoutout to Mike Dampier on my team for this chart to help sum up the TCO concept in a digestible way.</figcaption></figure><p>Alright, now the fun part.</p><p><strong><em>Modern CDPs on Snowflake</em></strong> and what that means, why it matters.</p><p><strong>Packaged CDP on Snowflake — </strong>Snowflake has a strong set of CDP partners that have constructed their architecture with Snowflake as the backbone of their solution. These CDP partners can provide an end-to-end offering on top of Snowflake’s Data Cloud Platform which eliminates the foundational problem of multiple copies of data.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*BrCtj16-IYxXIH-VnvsaJA.png" /><figcaption>Snowflake for Marketing Deck</figcaption></figure><p>The benefit of this approach is that it provides a <strong>managed solution</strong> — this is great for companies who lack support of an IT team to manage data and want to opt-in for a fully managed service that allows for flexibility, scalability but without the operational overhead. Modern packaged CDPs can still offer <em>“composable”</em> aspects but with the benefit of an out-of-the-box solution.</p><p><strong>Composable CDP on Snowflake</strong> — The best way to understand this is that each component of the CDP is represented by a different partner solution that is independent of each other. This allows a company to pull together best-of-breed applications as well as have more granular control over the actual data models, identity spine, and data transformation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*eSjKdxj9tfVe2ZHrOriD9A.png" /><figcaption>Snowflake for Marketing Deck</figcaption></figure><p>The benefit of this approach is that it provides <strong>increased control</strong> — companies can continue to leverage existing investments in Snowflake and allow their IT teams to have more control in managing the data pipelines and the applications (ie. reverse ETL partners for activation).</p><p>That said, it doesn’t always have to be <strong><em>a one-or-the-other </em></strong>approach. There are companies looking at a hybrid approach depending on their use cases, current resourcing, and technology investments.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/940/1*NSRU-YCAbpVPYRqCwcKsqw.png" /><figcaption>Snowflake for Marketing Deck</figcaption></figure><p>There are a number of questions to consider as you think about the best approach for your business, such as:</p><ol><li>What use cases matter to our business? Do we need real-time data access, live data for these use cases?</li><li>What activation or personalization channels are part of our marketing goals?</li><li>What other “data out” requirements or needs do we have?</li><li>Do we have a centralized IT team to manage additional pipelines? Do we have the resources necessary for implementation?</li><li>Requirements of an ID resolution layer within the CDP or within the data cloud platform? Are there other downstream use cases that require the identity layer?</li><li>Segmentation needs — how complex or simple?</li><li>What are the data collection and ingestion patterns we have today? Does that need to change?</li></ol><p>This is not exhaustive by any means but a good place to start.</p><p>Snowflake is uniquely positioned to support your marketing use cases. We operate across all three major public clouds thus truly providing a <strong><em>multi-cloud approach</em></strong>. We are <strong><em>neutral</em></strong><em> and </em><strong><em>conflict-free</em></strong> from a business standpoint as we are not a content platform, content publisher, or ad seller nor are we an ID partner, SSP, DSP, or measurement provider. However, we have done a lot of the heavy lifting to build partnerships in these categories so that our customers can facilitate end to end use-cases in a streamlined fashion with governance and privacy at the forefront.</p><p>When you think about the Snowflake Data Cloud, we’re providing the <strong>underlying scalable, robust infrastructure</strong> and <strong>bringing the ecosystem of partners and customers together</strong>. Not only do we have data providers in our marketplace, we also have application providers that offer their services directly where the data lives. It’s a <strong><em>paradigm shift</em></strong> where Snowflake is bringing the technology to the data, thus eliminating data movement.</p><p>This allows us to be the <strong>foundation</strong> of the modern marketing data stack, and in particular, CDPs.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ae2e5c830ae2" width="1" height="1" alt=""><hr><p><a href="https://medium.com/data-beats/the-evolution-of-customer-data-platforms-ae2e5c830ae2">The Evolution of Customer Data Platforms</a> was originally published in <a href="https://medium.com/data-beats">databeats</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Leadership is a way, not a title.]]></title>
            <link>https://medium.com/@nidhisd3/leadership-is-a-way-not-a-title-ff91d12b0221?source=rss-b688adbe474c------2</link>
            <guid isPermaLink="false">https://medium.com/p/ff91d12b0221</guid>
            <dc:creator><![CDATA[Nidhi DeSutter]]></dc:creator>
            <pubDate>Wed, 06 Dec 2023 23:23:08 GMT</pubDate>
            <atom:updated>2024-02-08T03:12:41.145Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/800/1*s4M56RYEL8fEC5_n4IHzgw.jpeg" /></figure><p><em>All viewpoints in this article are my own, based on my own experiences having worked in corporate America for 12+ years.</em></p><p>“Manager”, “Director”, “VP”, etc are all titles we are given as we move up the management chain. The reality is most can figure out the admin tasks that come with management &amp; run a team with discipline. However, organizations need more than just that when it comes to the growth of a business, growth of a team, and growth of individuals. You need vision, the ability to create buy-in, and inspire people to do the things they thought they could never do — that is <strong><em>being a leader</em></strong>.</p><p><strong>Leadership</strong> comes with:</p><ol><li>Embracing change.</li><li>Leading people that creates leaders (not just people leaders, I’m talking about thought leaders).</li><li>Setting direction that people want to follow.</li><li>Passion vs Control.</li><li>Transformational vs Transactional.</li><li>Taking Risks.</li><li>Challenging the status quo and creating new roads.</li><li>Leveraging conflict to drive change vs avoiding conflict.</li></ol><p><strong>Leadership</strong> starts with people. Actively listening to understand the strengths, areas of opportunities, motivation factors, career goals, and importantly, their journey. You can’t be prescriptive to a “playbook” — these are humans, not robots. If the focus becomes about you vs the people, you’re no longer a leader — and people will see right through that over time.</p><p><strong>Leadership</strong>, then extends to vision + strategy. The ability to set a direction for your team that encourages others to follow, work towards that common goal, and make a difference. Having a clear vision means being able to see the bigger picture <strong><em>but</em></strong> also all the parts that make up the bigger picture. Understanding the way things work — or don’t work — allows a leader to make the right decisions, sometimes tough ones to move forward.</p><p><strong>Leadership</strong> requires being out of your comfort zone because a true leader is open to doing things differently and taking risks. Every step forward comes with stepping out of your comfort zone because for most things to change, they have to be challenged. With that, comes taking responsibility and being accountable to yourself &amp; others.</p><p><strong>Leadership</strong> is about balancing intuition and logic to make better decisions. There is a time and place to leverage intuition to help you act quickly, trust your instincts, respond to different challenges effectively however, logic is what helps you stay objective, evidence-based so you can avoid biases &amp; errors. As humans, we all have biases and when those biases influence our decision making, we are not doing justice to our organization. Leaders have to be able to analyze the facts presented and use logical reasoning.</p><p><em>To close this off, the way I like to think about it is Leadership isn’t a title, it’s a responsibility, a mindset, a way of thinking that you embrace and that drives an organization &amp; people forward.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=ff91d12b0221" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Enable personalized customer experiences with Snowflake]]></title>
            <link>https://medium.com/@nidhisd3/enable-personalized-customer-experiences-with-snowflake-3d142734865b?source=rss-b688adbe474c------2</link>
            <guid isPermaLink="false">https://medium.com/p/3d142734865b</guid>
            <dc:creator><![CDATA[Nidhi DeSutter]]></dc:creator>
            <pubDate>Sun, 12 Nov 2023 04:26:43 GMT</pubDate>
            <atom:updated>2023-11-13T18:44:17.470Z</atom:updated>
            <content:encoded><![CDATA[<h3><strong><em>Overview</em></strong></h3><p>As consumers, we know first hand that personalized experiences matter. It matters more today than ever before. The value of getting personalization right– or wrong– is multiplying.</p><p>Covid was a period that changed what and how we purchased. Let’s face it — we live in a world of options today. For Retailers, customer experience is key and personalization sits at the heart of that experience.</p><p>There are four big trends to consider:</p><ol><li>Macroeconomics impacting discretionary spend — Spend has increased in travel but reduced elsewhere. More consumers are looking towards discounts or shopping opportunities with lower cost providers.</li><li>For first time shoppers, the most important consideration was making it easy for them to navigate in store and online through a personalized experience.</li><li>GDPR/CCPA really accelerated the focus on consumer privacy and it will be at the core of how we engage with consumers. In addition, 3P cookie deprecation has and will continue to change how we capture the customer’s journey and run the targeted ads on digital properties.</li><li>Emergence of channels such live commerce and metaverse — well, this is a whole new level of customer engagement. Think, a live video while shoppers interact with the brand in real-time.</li></ol><p>So, we built a personalization engine in the Snowflake Data Cloud by leveraging core Snowflake capabilities such as Streamlit, Snowpark and Snowpark Container Services, Unstructured Data Processing. We abstract image embeddings through a Deep Learning model trained with transfer learning, customer profile data, and clickstream data to optimize customer experience with personalized recommendations based on past use (recommender) or user uploaded image (image search).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*2TInTo3FD2KHHHESUQ8M8g.png" /></figure><h3><strong><em>How does this work?</em></strong></h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*AOGqlMcL-AEEO1_YtFu4hA.png" /><figcaption><em>Recommendation Engine built w/Snowflake</em></figcaption></figure><ol><li>For our recommendation system, we start off with the image data that gets preprocessed, encoded and stored into a Snowflake table. A pre-trained model (Resnet 18) is fine-tuned offline using transfer learning to classify the images &amp; the model is deployed into snowflake.</li><li>We call a UDF to extract the second to last layer of the model to get our image embeddings and store it into the snowflake database.</li><li>We leverage cosine similarity to determine how similar the images are to each other to make recommendations. When adding advanced personalization to our recommendation engine, we add a further layer of combining the embeddings with clickstream data to feed into a tree-based model that is hyper-parameter tuned and trained using Snowpark Container Services.</li><li>The final model is saved into a snowflake staging using Snowpark.ml, and recommendations are made using a UDF calling the model to produce interaction scores for each person/clothing product combination. All the results are stored into snowflake tables for further analysis.</li></ol><p>We leverage the base64 python library to encode the images without any loss to resolution or information in order to eloquently store them into our snowflake database. The UDF decodes the data, and calls the model to perform inference on the image data.</p><p>In the search functionality of the UI (built with Streamlit), the model is called each and every time we upload a new image of clothing or a “vibe” for recommendations of products.</p><h3><strong><em>The Retail Data Cloud</em></strong></h3><p>Retailers and Consumer Goods companies use Snowflake as their single source of truth for modern customer data activation by leveraging our virtually unlimited elastic compute scaling capabilities and our cross-cloud, cross-organization, the Retail Data Cloud. The Retail Data Cloud allows organizations to unify and connect a single copy of all of their data with our entire ecosystem of suppliers, vendors, customers, and partners, as well as bring any workload to their customer data across data engineering, data science, machine learning, and analytics workloads.</p><p>Customers can securely share and access governed data, tools, applications, and services — while preserving customer privacy. Instead of sending dozens of copies of your data to different SaaS applications and services, those applications can run on top of your Snowflake instance, or be built on top of it, or both. Bringing the applications to the data completely eliminates the silos, latency, cost, and complexity that’s been an unintended consequence of SaaS delivery.</p><p>Created in partnership with Jen Prusa, <a href="https://medium.com/u/f8e4a56f4202">Jenllieu</a></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3d142734865b" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>