Most organizations I work with understand they need to "do something with AI." Fewer have a clear strategy for using AI to create genuinely new value in their products. They collect vast amounts of data. They invest in models. And then the data sits, the models plateau, and the expected compounding advantage never materializes. 96% of organizations investing in AI report productivity gains. Only 5% capture value at scale. The gap is a loop problem.
In the previous issue, I introduced the PIEES framework: five dimensions where AI creates value in products. Personalization, Interaction, Emotion, Experiences, and Stories. That framework tells you where to look for value. What it doesn't yet explain is how that value compounds. Many products collect behavioral signals, conversation logs, sentiment markers, journey patterns, and content engagement metrics every single day. They just never feed them back.
I think the real competitive advantage lives in the data flywheel dimension of PIEES.
The PIEES data flywheel model
A data flywheel is a self-reinforcing system where proprietary data powers AI to deliver value, and the act of delivering value generates better data. Each cycle improves the next. The flywheel accelerates because the system learns from its own output.
Every PIEES dimension is already a data flywheel waiting to be activated. The five dimensions have a collection side and a usage side. Most organizations focus on building the collection side, sometimes without even realizing it. The usage side is where the biggest value potential is.
Let me walk through what this looks like across the five dimensions.
Five flywheels, five loops
The Personalization flywheel is the most mature. It collects behavioral signals, preferences, and purchase history. It uses that data for tailored recommendations, custom journeys, and dynamic pricing. More personalized experiences drive higher engagement, which generates more behavioral data, which enables even sharper personalization. Starbucks Deep Brew is the benchmark. 30 million digital connections feed an AI system that delivers hyper-personalized recommendations. The result is a 3x increase in average spend per targeted user, 30% ROI uplift, and 15% churn reduction. The flywheel is spinning fast because Starbucks closed the loop between what they collect and what they use.
The Interaction flywheel collects conversation logs, intent patterns, and resolution paths. It uses that data for smarter responses, proactive outreach, and intent prediction. Better interactions build trust, which drives higher usage, which generates richer conversation data. Klarna compressed customer service response time from 11 minutes to 2 minutes and saved $40 million annually, because each resolved interaction trained the system for the next one. Bank of America's Erica has handled over 2 billion interactions with a 98% resolution rate averaging 44 seconds. That's a flywheel that has been spinning for years.
The Emotion flywheel is the least mature but potentially the most defensible. It collects sentiment signals, tone markers, and frustration indicators. It uses that data for adaptive tone, empathetic responses, and mood-aware recommendations. This is strategically important because competitors can copy your recommendation algorithm or replicate your chatbot flows. But your accumulated emotional intelligence, how your product handles frustration, celebrates milestones, and navigates uncertainty, is nearly impossible to duplicate.
The Experiences flywheel collects cross-channel journey data across touchpoints over time. It uses that data to build products that deepen with use and anticipate needs before the user even articulates them. Better experiences drive longer retention, which generates more longitudinal data, which enables experiences that feel almost prescient. Disney is building the most ambitious version of this flywheel, bridging digital and physical. Disney World uses 30,000+ MagicBand sensors to track and collect movement, wait times, dining, and character interactions in real time. Their Compass platform (launched January 2025) unifies Disney+, Hulu, and ESPN+ streaming data with park behavioral data. A family that binge-watches Inside Out 2 on Disney+ surfaces as a high-intent park visitor, receives a personalized itinerary, and then has every touchpoint tracked by MagicBand during their visit. That park data refines future streaming recommendations, and the cycle deepens. The flywheel works because each visit makes the next one feel more personal, and that is nearly impossible to replicate without years of cross-platform interaction data.
The Stories flywheel collects engagement data on what narratives resonate, across formats and channels. It uses that data for targeted storytelling that hits harder with each cycle. Duolingo has built a compelling version of this flywheel, and it runs year-round, not just annually. Every streak day, XP milestone, and lesson pattern becomes a narrative hook. Their Year in Review uses AI to classify each learner into one of 8 behavioral personas and generates shareable story cards that users post voluntarily. The "Death of Duo" campaign in early 2025 generated 1.7 billion organic impressions in two weeks. The results are telling: ~80% of Duolingo's users arrive organically through word-of-mouth and social sharing. Each new user generates more learning data, which trains better AI personalization (including their Lily AI tutor), which creates more engaging experiences worth sharing. In April 2025, that same data flywheel powered 148 new AI-generated language courses, the largest expansion in company history. The first 100 courses took 12 years. These took under one year. That is a Stories flywheel compounding at scale.
Your PIEES flywheel audit
Let's make this actionable. For each PIEES dimension, score your product on two scales:
- Collection maturity (1-5): How much relevant data are you already capturing?
- Usage maturity (1-5): How effectively does that data feed back into the product?
Most products I evaluate score 3-4 on collection across several dimensions but 1-2 on usage. That is your biggest immediate opportunity. You do not need more data. You need to close the loop on the data you already have.
Find your widest gap, the dimension where the distance between collection and usage is greatest, and close that loop first. One spinning flywheel creates momentum that makes the next one easier to start.
Remember that 80-90% of company data is unstructured. That is exactly the kind of data PIEES dimensions collect: conversation logs, sentiment signals, journey patterns, engagement metrics. The unstructured data goldmine from issue #15 is the fuel. The PIEES framework from issue #25 is the engine. The flywheel is how you make them compound.
Your action step
Pick your weakest loop. For the PIEES dimension where your collection-to-usage gap is widest, write a single-page plan: what data are you collecting, where does it currently live, and what is one concrete way it could feed back into the product within the next quarter? One spinning flywheel is enough to change the trajectory of an AI initiative.
Watch Mark Abraham's TED talk "The Power of Personalization in the Age of AI." Abraham makes a compelling case that the companies winning at personalization are not the ones sending more messages, they are the ones closing the loop between customer data and tailored experiences. His insight that only about 10% of businesses truly master personalization reinforces the PIEES flywheel concept: the gap is not in collecting data, but in using it.
If you'd like help turning your PIEES audit into a concrete data flywheel strategy, or want me to speak to your leadership team on the business value of closing the AI loop, I'd love to help.