Metrics Don’t Just Measure Behavior. They Create it.
Most product organizations treat metrics as neutral — objective measures of how a product is performing. They're not. Metrics are incentive systems. They shape what gets built, how teams are evaluated, and crucially, how users behave. Choose the wrong ones, and you don't just miss a business goal. You shape people's lives in ways you never chose to.
Every metric you choose is an incentive in disguise.
The moment an organization defines success, it sets in motion a chain of incentives that runs from the boardroom to the product team to the end user. Internal teams optimize for what gets measured. Products get designed around what drives the metric. And users — consciously or not — adapt their behavior to the experience that results.
This is not a flaw in how organizations use metrics. It's how incentive systems work. The question is never whether your metrics are shaping behavior. It's whether they're shaping the behavior you actually want.
The mechanism is easier to see in hindsight. Consider watch time — a metric that seemed like a reasonable proxy for value. If people are watching, they must be engaged; if they're engaged, the product must be working. What that metric actually incentivized was content that kept people watching, regardless of whether watching was good for them. Recommendation algorithms optimized for the next click, not the next meaningful moment. The metric didn't measure value. It created a specific kind of attention — and the product was built around sustaining it.
Daily active users tell a similar story. More people, more often, must mean the product is succeeding. What it actually incentivized was habit formation over genuine utility. Products were designed around triggers, variable rewards, and re-engagement mechanics. The goal shifted from "does this help people?" to "does this bring people back?" Those aren't the same question. And the products that resulted reflect that difference.
In both cases, the metric wasn't wrong in isolation. The problem was treating it as a proxy for human value without asking whether the behavior it incentivized was actually good for the people experiencing it.
| Metric | Behavior it incentivizes |
|---|---|
| Watch time | Content optimized for continued viewing, not value |
| Daily active users | Habit formation mechanics, re-engagement triggers |
| Engagement rate | Emotionally activating content over informative content |
| Conversion rate | Friction reduction and persuasion mechanics over informed decision-making |
| Session length | Keeping people in the product rather than helping them accomplish what they came to do |
We've seen what happens when we get this wrong.
The metrics that defined the last era of digital technology weren't chosen maliciously. The teams that built the first generation of social platforms, streaming services, and consumer apps were trying to understand something reasonable: are people finding value in what we built? Engagement, watch time, return visits, daily active users — these felt like answers to that question.
But they were optimization metrics, measuring performance within a paradigm without asking whether the paradigm itself was right. Optimizing for attention turned out not to be the same thing as creating value in people's lives.
The consequences — compulsive checking, erosion of attention, algorithmic amplification of outrage, the documented impacts on mental health, particularly among younger users — weren't obvious at the time. In retrospect, they were the logical output of the metrics organizations chose. But hindsight is easy. What matters now is that we have evidence we didn't have then. We know how this plays out. And we are, whether we realize it or not, making the same kind of foundational metric decisions again — this time for AI.
The lesson isn't that those teams were negligent. It's that metrics have consequences that extend far beyond the dashboard. And once a metric becomes embedded in how an organization operates, it stops feeling like a choice, even when the consequences of that choice become impossible to ignore.
We know this now. We have evidence we didn't have then. The question — the uncomfortable one — is whether we act like we know it.
The KPIs you set today will shape human behavior for a generation.
AI isn't adding a new feature to how people relate to technology. It's rewriting the relationship entirely. The products being built today — AI assistants, agentic experiences, personalization systems — will interact with people at a scale, frequency, and intimacy that previous technologies didn't approach. And the KPIs being written into product roadmaps right now, before many organizations have fully reckoned with what they're building, will define that relationship for a generation.
Most organizations aren't approaching this as a values decision. They're reaching for the metrics they know: engagement, retention, and usage frequency. Familiar measures applied to an unfamiliar moment. The instinct is understandable — these are the metrics that worked before. But as we've written elsewhere, strong analytics show you what already happened. They cannot surface what is shifting. And the human consequences of that gap will scale accordingly.
This is the moment to ask a harder question. Not just “What do we want people to do with this product?” but “What do we want this product to do for people?” Those are very different questions that produce different metrics. Different metrics produce different products. And, ultimately, different outcomes for people’s lives.
Frequently Asked Questions
What does it mean to reach for a metric by default — and why is it risky? A default metric is one chosen not because it best reflects what success means for your users, but because it's familiar, easy to instrument, or already in use elsewhere in the organization. The risk isn't that these metrics are wrong in absolute terms — it's that they were designed for a different product, a different moment, or a different definition of value. When organizations carry default metrics into genuinely new territory, like AI-powered products, they're measuring a new paradigm with an old ruler. The metric feels safe precisely because it's familiar. That familiarity is the danger.
Why are product metrics more than measurement tools? Metrics aren’t just reporting mechanisms; they are incentive systems. When an organization defines success through a specific metric, it sets in motion a chain of decisions — about what to build, how to design it, and how to evaluate performance — that ultimately shapes how users interact with the product. The metric doesn't just reflect behavior; it produces it. This is why metric selection is one of the most consequential decisions a product organization makes, even though it rarely gets treated that way.
How do KPIs shape user behavior? Teams optimize products around the metrics they're measured against. Features get built to drive the metric. Design decisions get made to sustain it. Users then adapt their behavior to the experience that results — often in ways that were never explicitly intended. The relationship between a KPI and user behavior is rarely linear or obvious, but it's almost always present. The question product leaders need to ask isn't whether their metrics are shaping behavior — it's whether they're shaping the behavior they actually want to create.
What are examples of metrics that unintentionally shaped behavior? Watch time incentivized recommendation algorithms to optimize for continued viewing regardless of whether the content was valuable, contributing to rabbit holes and passive consumption. Daily active user metrics incentivized habit-formation mechanics — triggers, variable rewards, re-engagement notifications — that shifted product goals from utility to dependency. Engagement rate metrics incentivized emotionally activating content over informative content, contributing to the amplification of outrage and filter bubbles. In each case, the metric wasn't inherently wrong. The problem was optimizing for it without asking what behavior it was actually creating.
Why does metric selection matter more in the AI era? AI systems can influence decisions, behaviors, and workflows at a scale and intimacy that previous technologies didn't approach. An AI assistant embedded in how someone works, communicates, or makes decisions has a fundamentally different relationship with that person than a social media feed. The metrics used to define success for these products will shape that relationship at scale. And the consequences – positive or negative – will compound faster than they did in previous technology eras. Getting this right requires a different kind of human understanding than most organizations have built.
What metrics should product organizations consider for AI products? The most useful shift is from activity metrics to outcome metrics — measuring not just what people do with a product, but what the product enables them to do. For AI products specifically, this might include measures of decision quality, task success in real-world contexts, user confidence, and trust over time. It also means being honest about what you're measuring and what you're not. Engagement and retention can coexist with diminished user wellbeing, and a healthy metrics framework accounts for that tension rather than ignoring it.
What is a human-centered KPI? A human-centered KPI measures whether a product helps users achieve meaningful goals, improves their experience, or creates outcomes they recognize as valuable, rather than simply maximizing usage or attention. It's defined by asking, "What does success look like for the person using this?" before asking, "What does success look like for the business?" The two aren't always in conflict, but they're not automatically aligned either. Organizations that define success from the human perspective first tend to build products that earn loyalty rather than just capture it. AnswerLab's research on consumer loyalty consistently shows that the products people stay with are the ones that made their lives meaningfully better — not the ones that were merely hard to leave.
How can product leaders build more responsible metrics frameworks? Start by mapping the chain of incentives that flows from your current KPIs — what behaviors do they reward internally, what product decisions do they drive, and what user behaviors do they ultimately produce? Then ask whether those user behaviors are the ones you want to create. For most organizations, this exercise surfaces gaps between what metrics measure and what the organization actually values. Closing those gaps requires both the willingness to define success differently and the human understanding capability to stay connected to how people are actually living with your products. That's the work that turns insight into competitive advantage.
AnswerLab helps product, digital, and innovation leaders build the human understanding that should sit behind their most important strategic decisions — including the metrics that govern them. Start a conversation.
