<?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 Claire McDaniel on Medium]]></title>
        <description><![CDATA[Stories by Claire McDaniel on Medium]]></description>
        <link>https://medium.com/@clairemcdaniel_49454?source=rss-2a5684df7f5c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/0*qJDeN7yuoDR1bmjD.</url>
            <title>Stories by Claire McDaniel on Medium</title>
            <link>https://medium.com/@clairemcdaniel_49454?source=rss-2a5684df7f5c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Tue, 17 Nov 2026 07:16:50 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@clairemcdaniel_49454/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 Next Frontier of Analytics: From Prediction to Behavior Change]]></title>
            <link>https://medium.com/@clairemcdaniel_49454/the-next-frontier-of-analytics-from-prediction-to-behavior-change-9a4d73db0d06?source=rss-2a5684df7f5c------2</link>
            <guid isPermaLink="false">https://medium.com/p/9a4d73db0d06</guid>
            <category><![CDATA[machine-learning]]></category>
            <category><![CDATA[causal-inference]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[data-science]]></category>
            <category><![CDATA[product-management]]></category>
            <dc:creator><![CDATA[Claire McDaniel]]></dc:creator>
            <pubDate>Mon, 01 Jun 2026 23:34:33 GMT</pubDate>
            <atom:updated>2026-06-02T00:34:54.975Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>The Next Frontier of Analytics: From Prediction to Behavior Change</strong></p><p>How Behavioral Theory, Machine Learning, and Causal AI Work Together to Drive Behavior Change</p><p>Every organization wants to influence behavior.</p><p>They want customers to adopt a product, employees to use a new tool, patients to follow treatment plans, or users to engage more deeply with a platform.</p><p>To support these decisions, organizations have invested heavily in analytics. They collect survey data, build dashboards, run experiments, train machine learning models, and monitor user telemetry.</p><p>Yet despite all this information, products and initiatives often fail to achieve the levels of adoption teams expect.</p><p>A new feature launches and engagement remains flat.</p><p>An AI assistant is deployed, and employees continue using existing workflows.</p><p>A customer experience initiative is rolled out and behavior changes less than anticipated.</p><p>When this happens, a simple question emerges:</p><p><strong>What should we change?</strong></p><p>At first glance, this seems like a problem analytics should already solve.</p><p>After all, we can identify the factors associated with adoption. We can build models that predict who is likely to adopt. We can measure usage with remarkable precision.</p><p>But there is an important distinction:</p><p><strong>Knowing what drives behavior is not the same as knowing how to change it.</strong></p><p>Traditional research helps us understand why behavior occurs.</p><p>Machine learning helps us predict who is likely to behave.</p><p>Neither directly tells us which intervention is most likely to create meaningful change.</p><p>Visual 1 illustrates the evolution of analytics from understanding behavior to predicting behavior and, ultimately, to changing behavior. It highlights how behavioral theory, machine learning, and causal AI each contribute a unique perspective to decision-making.</p><p><strong>Visual 1: The Analytics Gap</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/626/1*e9XN3lqScd-suaYuSK2gFg.png" /></figure><p>For decades, researchers have focused on understanding behavior.</p><p>Frameworks such as the Technology Acceptance Model (TAM) helped explain why people adopt technologies. These models revealed factors such as usefulness, ease of use, enjoyment, and social influence impact adoption decisions.</p><p>These insights are valuable.</p><p>They help us understand the psychological mechanisms that drive behavior.</p><p>But understanding behavior and changing behavior are not the same thing.</p><p>A product team cannot deploy a path coefficient.</p><p>Knowing that a variable matters does not automatically tell us which action will create the greatest improvement in adoption.</p><p><strong>The Rise of Predictive Analytics</strong></p><p>As organizations became increasingly data-driven, attention shifted toward prediction.</p><p>Machine learning allowed us to predict:</p><ul><li>who will churn,</li><li>who will buy,</li><li>who will engage,</li><li>who will adopt.</li></ul><p>This was a major step forward.</p><p>Prediction enables organizations to identify opportunities, prioritize resources, and anticipate future outcomes.</p><p>But prediction has an important limitation.</p><p>A predictive model can identify the strongest predictor of behavior without telling us whether changing that factor will actually create meaningful change.</p><p>This distinction is subtle but critical.</p><p>Many organizations implicitly assume that the most important predictor is also the best intervention target.</p><p>My research suggests otherwise.</p><p>In one technology adoption study, machine learning consistently identified Hedonic Motivation — the degree to which users found the experience enjoyable — as one of the strongest predictors of adoption.</p><p>At first glance, the implication seemed obvious:</p><p>If enjoyment is the strongest predictor, then increasing enjoyment should create the largest increase in adoption.</p><p>But when I applied causal machine learning methods, the story changed.</p><p>While enjoyment was highly predictive, it was not the strongest intervention lever.</p><p>Perceived product value produced larger estimated improvements in adoption when intervention effects were simulated.</p><p>In other words:</p><p><strong>The strongest predictor is not necessarily the best intervention target.</strong></p><p>That finding challenged one of the most common assumptions in modern analytics.</p><p>Prediction and intervention are not the same problem.</p><p><strong>The Missing Layer</strong></p><p>This realization exposed what I believe is a gap in many analytics workflows.</p><p>Traditional theory helps us understand:</p><p><strong>Why behavior occurs.</strong></p><p>Machine learning helps us identify:</p><p><strong>Who is likely to behave.</strong></p><p>But neither directly answers:</p><p><strong>How much change can we create if we intervene?</strong></p><p>That is where causal AI enters the picture.</p><p>Rather than asking what is associated with an outcome, causal methods attempt to estimate what would happen if conditions changed.</p><p>Instead of asking:</p><p>“What predicts adoption?”</p><p>we ask:</p><ul><li>Which factors are actionable?</li><li>Which interventions are likely to create meaningful change?</li><li>Which users are most responsive to those interventions?</li><li>How much improvement should we expect?</li></ul><p>These are fundamentally different questions.</p><p><strong>A Three-Prong Framework for Behavioral AI</strong></p><p>To address this gap, I began developing a technology-agnostic framework that integrates behavioral theory, machine learning, and causal AI.</p><p>Each component answers a different question.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/650/1*KuwNG1geRYZ-WkA11spcgw.png" /></figure><p>Together, these three perspectives create a more complete foundation for decision-making than any single approach alone. To test whether prediction and intervention were actually the same problem, I compared predictive importance derived from machine learning with causal importance estimated using a structural causal framework and Double Machine Learning. The results were surprising. Visual 2 compares predictive importance with causal intervention impact, demonstrating that the strongest predictor of adoption is not necessarily the best target for intervention.</p><p><strong>Visual 2: Prediction vs Intervention</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/611/1*0_UKuKmJuGCmGGiyqqTyMg.png" /></figure><p>The lesson is simple but profound.</p><p>The variables that best predict behavior are not always the variables that create the largest behavioral change when improved.</p><p>That distinction matters.</p><p>Because organizations do not improve outcomes by identifying important variables.</p><p>They improve outcomes by changing the right variables.</p><p><strong>Beyond Technology Adoption</strong></p><p>Although these ideas emerged from technology adoption research, the implications extend far beyond any single domain.</p><p>Organizations deploying AI copilots want to know which interventions will increase employee adoption.</p><p>Healthcare providers want to know which actions are most likely to improve patient outcomes.</p><p>Product teams want to understand which changes will increase engagement.</p><p>Marketing teams want to know which messages will actually influence behavior rather than simply correlate with it.</p><p>Across these domains, the same challenge appears repeatedly.</p><p>Prediction is often used to make intervention decisions.</p><p>But prediction and intervention answer different questions.</p><p><strong>The Next Frontier</strong></p><p>For years, analytics has focused primarily on understanding behavior or predicting behavior.</p><p>The next frontier is understanding how to change behavior.</p><p>Behavioral theory helps us understand why.</p><p>Machine learning helps us identify who.</p><p>Causal AI helps us determine what to change and how much impact to expect.</p><p>Together, these perspectives create a more complete approach to decision-making in increasingly adaptive and intelligent systems.</p><p>And perhaps most importantly, they remind us of a lesson that many organizations have yet to learn:</p><p><strong>The strongest predictor is not necessarily the best intervention target.</strong></p><p><strong>The Bottom Line: Moving to Decision Intelligence</strong></p><p>Understanding the distinction between prediction and intervention is a conceptual victory, but the real value lies in the shift to <strong>complete decision intelligence</strong>. If you are a practitioner, you can begin applying this framework immediately:</p><ul><li><strong>Audit Your Metrics:</strong> Stop assuming your strongest predictive features are your best intervention levers. You may be optimizing for the wrong targets.</li><li><strong>Simulate Before You Scale:</strong> Use causal inference to test the impact of interventions <em>before</em> spending engineering resources.</li><li><strong>Standardize Your Decisions:</strong> Use a “Go/No-Go” threshold to turn complex ML outputs into a clear, actionable roadmap for your team.</li></ul><p>The future of behavioral AI isn’t just in predicting what will happen — it’s in knowing exactly how to make it happen. It’s time to move beyond monitoring behavior and start creating it.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=9a4d73db0d06" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>