<?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 ciValue on Medium]]></title>
        <description><![CDATA[Stories by ciValue on Medium]]></description>
        <link>https://medium.com/@civalue?source=rss-bf9a24cd68e7------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*OeL5uqdndgLczfG35Q2zzg.jpeg</url>
            <title>Stories by ciValue on Medium</title>
            <link>https://medium.com/@civalue?source=rss-bf9a24cd68e7------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Wed, 18 Nov 2026 06:18:05 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@civalue/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[Sifting Through Retailer’s Data With Smart Alerts]]></title>
            <link>https://medium.com/@civalue/sifting-through-retailers-data-with-smart-alerts-68a8e68594f7?source=rss-bf9a24cd68e7------2</link>
            <guid isPermaLink="false">https://medium.com/p/68a8e68594f7</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[retail-ml]]></category>
            <category><![CDATA[retail]]></category>
            <category><![CDATA[retailtech]]></category>
            <dc:creator><![CDATA[ciValue]]></dc:creator>
            <pubDate>Sun, 13 Mar 2022 09:27:51 GMT</pubDate>
            <atom:updated>2022-03-13T09:27:51.764Z</atom:updated>
            <content:encoded><![CDATA[<p>By Gilad Cohen, CTO, ciValue</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ynU_ehzNBxoTo8VfIM5vIw.jpeg" /></figure><h3>The problem: marketers are drowning in data</h3><p>Big data gives marketers access to a wealth of information, that’s the good news. The bad news is that there is so much data that it is increasingly complicated to utilize it effectively. Say I’m Kellogs and I’m looking at the performance of Special K, Frosted Flakes and Rice Krispies in three regions, and each of those brands has two segments, that means my marketing team has 720 data combinations to follow! Even if they have the capacity to keep track of 720 options, it can be hard to understand what’s important and what isn’t, especially in real-time. An additional challenge is that simply tracking change in sales isn’t enough to generate actionable insights. For some items, a 10% change from week to week isn’t significant. For other items, even a 2% change is significant. That’s because some changes are expected or seasonal — things like growth in sales of notebooks before school starts and a boost in swim diapers in the summer. But keeping track of what is significant and what isn’t for every item can be extremely challenging.</p><p>With this in mind, we set out to help retailers cut through the noise and quickly recognize which changes in turnover or other key metrics are expected and of no concern and which could represent a new trend or an aggressive move from the competition. <strong>We wanted to empower marketers to become “citizen data analysts” and give every type of customer the tools to make the most of the data we offer.</strong></p><h3>The objective–understanding trends in market share</h3><p>When we talked to customers, we understood that they were interested primarily in tracking trends in market or wallet share. That’s because a rise or drop in their own sales for a specific product is less significant if the same change has occurred across the entire field. But market share isn’t a simple thing to calculate. For example, how do you determine what comprises a significant period of time? Is it a week, a month, or a quarter? And how do you define the relevant businesses for comparison?</p><p><strong>Leveraging data science to cut through the noise.</strong></p><p>At ciValue, we serve dozens of retailers, thousands of brands, and millions of customers everyday with our solution. Leveraging AI, we help marketers across the globe make better and faster decisions to grow their market share. To further support them, we wanted to add alerts to cut the time it takes them to go from insights to action and focus them on the insights that matter the most for their business.</p><p>We used a quantile regression algorithm to analyze brands over different periods of time (week, month, quarter). We then built a chart of trends, identified trends for every brand in the system, and created a wallet share trend based on the past week/quarter/year. The threshold is automatic, and there are mechanisms built in to cancel out the noise. When we are confident that a trend is statistically significant, we send the customer an alert, and the customer can go to the report and better understand what is taking place. The alerts don’t replace the reports–they just help marketers understand where to focus their attention.</p><p>There are defaults in the system, but all alerts can be customized according to filters that the customer is interested in and ciValue helps tailor the configuration according to the customer’s specific needs. For example, the customer determines the period to look at and whether the model should be configured on a regional or national basis. The customer also chooses the brand and determines the parameters–some customers only want to see their top three brands or prefer to focus on the product that has changed most in comparison to the previous month. The goal is to give a brand no more than 5 or 6 alerts a week so marketers can focus on the top alerts, the information that is most relevant to the company’s business goals. The alerts can even be triggered across categories and brands when relevant.</p><h3>Data science for all</h3><p>Data science is being democratized like coding is with “code-free programming”. With tools like ciValue’s ML alerts at their disposal, marketers can become “citizen data scientists” and mine the information they need, identify opportunities, solve problems and share crucial data to support business decisions without becoming overwhelmed.</p><p>If you want to learn more or brainstorm about creating better processes for your marketing users feel free to<a href="https://www.linkedin.com/in/gilad-cohen-353948/"> reach out to me</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=68a8e68594f7" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Best Practices for Churn Prevention in Retail]]></title>
            <link>https://medium.com/engineering-at-civalue/best-practices-for-churn-prevention-in-retail-db5f502e0dba?source=rss-bf9a24cd68e7------2</link>
            <guid isPermaLink="false">https://medium.com/p/db5f502e0dba</guid>
            <category><![CDATA[retail]]></category>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[retail-technology]]></category>
            <category><![CDATA[category-management]]></category>
            <category><![CDATA[churn-prediction]]></category>
            <dc:creator><![CDATA[ciValue]]></dc:creator>
            <pubDate>Tue, 16 Nov 2021 07:50:54 GMT</pubDate>
            <atom:updated>2021-11-17T07:45:43.840Z</atom:updated>
            <content:encoded><![CDATA[<p>By <a href="https://www.linkedin.com/in/gilad-cohen-353948/">Gilad Cohen</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*ku24W66dqTB4m9AZ.jpeg" /></figure><p>Photo by <a href="https://unsplash.com/@fikrirasyid?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Fikri Rasyid</a> on <a href="https://unsplash.com/s/photos/supermarket?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></p><p>Customer churn is a major concern for retailers. Anyone who has been involved in retail knows that churn hurts, as organizations spend time and resources keeping their customers loyal and engaged. To keep them coming back and prevent them from shopping elsewhere, marketing, brands, category, merchandising and data teams need robust insights into their customers’ preferences, lifestyles, and likelihood to grow or churn.</p><p>The most basic indicator in churn prediction is the time elapsed since the last visit. In most industries, if a customer hasn’t been seen for three months, it’s safe to assume that they’ve either churned or gone on a trip around the world. But identifying churn three months after the fact is usually too late to prevent it from happening. Even traditional models that predict churn before it happens aren’t always accurate, as they often look at churn risk at the store level only, which isn’t always enough to provide the full picture.</p><p>Recently, we have worked with several teams from different verticals (both food and non-food) to help them optimize their churn detection and grow their wallet share. What we discovered is that, if a customer is still active but significantly reduces activity in a category with high recurrency such as pasta or soda, there’s a high chance that he or she is buying that item elsewhere. Even if he/she is still buying other items from the store, the fact that some purchase activity dropped is a strong predictor of churn in the future, as your competition is starting to nurture and grow engagements with these specific shoppers. That’s precisely the point where you need to intervene.</p><h3>The benefits of looking at the category level</h3><p>Given the limits of traditional models, best practices in churn prediction today drill down beyond the store level to the category level to gain the insights that can actually predict and prevent churn effectively.</p><p>Let’s look at retailers that sell multiple products, like a Health and Beauty store. Customers often come in looking for something specific, for example, to fill a prescription, but they’re likely to pick up additional items while they’re in the store. That’s a good thing, not only in terms of revenue but for churn as well — as we know when customers buy from more than one category, they come in more often and have a lower risk of churn.</p><p>However, multi-category stores are harder to assess since customer behavior varies greatly between categories. Customers run out of different types of products at different cadences — diapers go faster than makeup, perishable food items need to be replenished more often than cleaning supplies.</p><p><strong>That’s why effective churn prediction models must go beyond the store level and assess churn at the category level in order to gain smart insights into customer behavior and predict churn with greater accuracy</strong>. By leveraging ML models, the teams can now look at many customer behavior features in a specific category to generate important insights for retailers.</p><p>It’s important to use models that take the unique characteristics of each category into account — it’s not one size fits all. For example, let’s say a customer used to buy baby food and suddenly stopped — that’s not necessarily cause for concern, it’s the nature of the category. Babies grow up, and there’s no use wasting valuable marketing resources trying to get a customer to buy their preschooler baby products. But other categories, like fresh food and toiletries, don’t have a natural ending and should be ongoing purchases. When a customer suddenly stops buying in those categories, there’s an opportunity to take action and prevent churn. For one grocer, we managed to identify roughly 70 categories/products in which inactivity indicates a high potential for churn and 40 such indicators for our beauty customers.</p><p><strong>The tests we run at </strong><a href="https://civalue.com/"><strong>ciValue</strong></a><strong> show that churn models that include categories identify early stages of churn 3–5 times more effectively than store-wide models, giving retailers a sufficient window for effective intervention. </strong>For our churn prediction models who use category level information (vs. purely store level information) we found that the “category model” is significantly more effective in raising the warning flag than store level only model.</p><p>To do that we run our prediction model on active shoppers (i.e. ones that visited in the last month) and checked if the shoppers actually churned 6 month later. What we found is that our predictive churn scoring model calculated a churn propensity in the range of 51% to 81% (25%/75%) for the customers who actually churned and a score of 18% to 53% (25%/75%) for the customers that did not (see picture).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*_5n7GVs2GcoD_lFp" /></figure><h3>Build or buy?</h3><p>Most data teams we meet with are in a build or buy conundrum for their churn predictions. If you and your team are considering such options here are a few parameters that we recommend you look at:</p><p><strong>In-house development:</strong> One option is to build a solution in house, but that requires a major investment of both time and finances — hiring a data science team, managing real-time data at scale, and keeping the data secure. It’s difficult and time-consuming. If you’re considering this option, think about the time and resources you have to allocate to the project. That includes not only building the models, but maintaining them and working closely with your downstream users.</p><p><strong>If buying an existing solution</strong>, it’s important to consider how you’re going to test the models. If it’s a generic solution, how will you extract value and ROI, or how fast can you customize the algorithm? Can the solution support the fast-paced FMCG environment? How adaptable is the solution to the complex domain?</p><p>If you are in the retail world and want to brainstorm around best practices for your ML activities, feel free to <a href="https://www.linkedin.com/in/gilad-cohen-353948/?originalSubdomain=il">reach out </a>— we’re here to help!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=db5f502e0dba" width="1" height="1" alt=""><hr><p><a href="https://medium.com/engineering-at-civalue/best-practices-for-churn-prevention-in-retail-db5f502e0dba">Best Practices for Churn Prevention in Retail</a> was originally published in <a href="https://medium.com/engineering-at-civalue">engineering at civalue</a> on Medium, where people are continuing the conversation by highlighting and responding to this story.</p>]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Best Practices for Churn Prevention in Retail]]></title>
            <link>https://medium.com/@civalue/best-practices-for-churn-prevention-in-retail-d9e0f9b674c8?source=rss-bf9a24cd68e7------2</link>
            <guid isPermaLink="false">https://medium.com/p/d9e0f9b674c8</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[retail]]></category>
            <category><![CDATA[retail-technology]]></category>
            <dc:creator><![CDATA[ciValue]]></dc:creator>
            <pubDate>Tue, 16 Nov 2021 07:31:50 GMT</pubDate>
            <atom:updated>2021-11-16T07:35:49.407Z</atom:updated>
            <content:encoded><![CDATA[<p>By <a href="https://www.linkedin.com/in/gilad-cohen-353948/">Gilad Cohen</a></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Q8SAse-I3eJLti2LMjg5KA.jpeg" /></figure><p>Photo by <a href="https://unsplash.com/@fikrirasyid?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Fikri Rasyid</a> on <a href="https://unsplash.com/s/photos/supermarket?utm_source=unsplash&amp;utm_medium=referral&amp;utm_content=creditCopyText">Unsplash</a></p><p>Customer churn is a major concern for retailers. Anyone who has been involved in retail knows that churn hurts, as organizations spend time and resources keeping their customers loyal and engaged. To keep them coming back and prevent them from shopping elsewhere, marketing, brands, category, merchandising and data teams need robust insights into their customers’ preferences, lifestyles, and likelihood to grow or churn.</p><p>The most basic indicator in churn prediction is the time elapsed since the last visit. In most industries, if a customer hasn’t been seen for three months, it’s safe to assume that they’ve either churned or gone on a trip around the world. But identifying churn three months after the fact is usually too late to prevent it from happening. Even traditional models that predict churn before it happens aren’t always accurate, as they often look at churn risk at the store level only, which isn’t always enough to provide the full picture.</p><p>Recently, we have worked with several teams from different verticals (both food and non-food) to help them optimize their churn detection and grow their wallet share. What we discovered is that, if a customer is still active but significantly reduces activity in a category with high recurrency such as pasta or soda, there’s a high chance that he or she is buying that item elsewhere. Even if he/she is still buying other items from the store, the fact that some purchase activity dropped is a strong predictor of churn in the future, as your competition is starting to nurture and grow engagements with these specific shoppers. That’s precisely the point where you need to intervene.</p><h3>The benefits of looking at the category level</h3><p>Given the limits of traditional models, best practices in churn prediction today drill down beyond the store level to the category level to gain the insights that can actually predict and prevent churn effectively.</p><p>Let’s look at retailers that sell multiple products, like a Health and Beauty store. Customers often come in looking for something specific, for example, to fill a prescription, but they’re likely to pick up additional items while they’re in the store. That’s a good thing, not only in terms of revenue but for churn as well — as we know when customers buy from more than one category, they come in more often and have a lower risk of churn.</p><p>However, multi-category stores are harder to assess since customer behavior varies greatly between categories. Customers run out of different types of products at different cadences — diapers go faster than makeup, perishable food items need to be replenished more often than cleaning supplies.</p><p><strong>That’s why effective churn prediction models must go beyond the store level and assess churn at the category level in order to gain smart insights into customer behavior and predict churn with greater accuracy</strong>. By leveraging ML models, the teams can now look at many customer behavior features in a specific category to generate important insights for retailers.</p><p>It’s important to use models that take the unique characteristics of each category into account — it’s not one size fits all. For example, let’s say a customer used to buy baby food and suddenly stopped — that’s not necessarily cause for concern, it’s the nature of the category. Babies grow up, and there’s no use wasting valuable marketing resources trying to get a customer to buy their preschooler baby products. But other categories, like fresh food and toiletries, don’t have a natural ending and should be ongoing purchases. When a customer suddenly stops buying in those categories, there’s an opportunity to take action and prevent churn. For one grocer, we managed to identify roughly 70 categories/products in which inactivity indicates a high potential for churn and 40 such indicators for our beauty customers.</p><p><strong>The tests we run at </strong><a href="https://civalue.com/"><strong>ciValue </strong></a><strong>show that churn models that include categories identify early stages of churn 3–5 times more effectively than store-wide models, giving retailers a sufficient window for effective intervention. </strong>For our churn prediction models who use category level information (vs. purely store level information) we found that the “category model” is significantly more effective in raising the warning flag than store level only model.</p><p>To do that we run our prediction model on active shoppers (i.e. ones that visited in the last month) and checked if the shoppers actually churned 6 month later. What we found is that our predictive churn scoring model calculated a churn propensity in the range of 51% to 81% (25%/75%) for the customers who actually churned and a score of 18% to 53% (25%/75%) for the customers that did not (see picture).</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*K6Mh69ecbwDbNShw" /></figure><h3>Build or buy?</h3><p>Most data teams we meet with are in a build or buy conundrum for their churn predictions. If you and your team are considering such options here are a few parameters that we recommend you look at:</p><p><strong>In-house development:</strong> One option is to build a solution in house, but that requires a major investment of both time and finances — hiring a data science team, managing real-time data at scale, and keeping the data secure. It’s difficult and time-consuming. If you’re considering this option, think about the time and resources you have to allocate to the project. That includes not only building the models, but maintaining them and working closely with your downstream users.</p><p><strong>If buying an existing solution</strong>, it’s important to consider how you’re going to test the models. If it’s a generic solution, how will you extract value and ROI, or how fast can you customize the algorithm? Can the solution support the fast-paced FMCG environment? How adaptable is the solution to the complex domain?</p><p>If you are in the retail world and want to brainstorm around best practices for your ML activities, feel free to <a href="https://www.linkedin.com/in/gilad-cohen-353948/?originalSubdomain=il">reach out </a>— we’re here to help!</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d9e0f9b674c8" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>