Usage Can Validate Demand and Still Damage the Business

The easiest product meeting is the one where the activation graph goes up. Customers are using the new capability, conversion has improved and the team can point to visible demand. Everyone leaves with the same conclusion: we should scale it.

I have become increasingly suspicious of that conclusion.

Usage validates demand. It does not automatically validate the business behind it. This distinction matters more as products depend on AI inference, human review, third-party platforms, payment networks and physical infrastructure. Every successful interaction can carry a real marginal cost. Sometimes the customers who use a feature most are the customers the company can least afford to serve.

This is becoming visible across the industry. AI features are moving deeper into everyday business workflows while providers experiment with subscriptions and new pricing models. At the same time, the infrastructure behind those features is running into energy, capacity and local acceptance constraints. Payment products face similar pressure when services that were initially cheap or free can no longer be funded under the same conditions. Meanwhile, the growing list of discontinued AI products is a useful reminder that customer interest and technical possibility are not enough.

Yet many product teams still separate validation from economics. Product validates whether people use the feature. Growth validates whether it converts. Finance later works out whether the resulting customers are profitable.

That sequence is too late.

Consider an AI support product. A team may celebrate the number of conversations handled by the model. But each conversation could involve several model calls, retries, retrieval requests and occasional human escalation. If the product produces more contacts because customers find it easier to ask questions, the usage metric can rise while the cost per resolved problem gets worse.

The same issue appears in onboarding automation. Completion may improve, but a percentage of those completed setups might create incorrect configurations that operations teams must repair. The product dashboard records a successful activation. The operational budget records the actual result.

I do not expect product managers to become accountants. I do expect them to understand the cost structure of the behavior they are trying to create. If a roadmap is designed to increase an action, the team should know what happens economically when that action increases tenfold.

That means measuring a complete customer outcome rather than the most convenient product event. For an AI support product, I would want to connect resolution quality, revenue or retention impact, model cost, escalation and rework. For onboarding, I would connect completion with the cost of correction and the customer’s later success. The point is not to add another crowded dashboard. It is to create one joined view of value delivered and cost incurred.

This also changes prioritization. A less impressive feature that reduces retries or moves customers toward a cheaper successful path may create more value than a visible feature that drives raw usage. A pricing change may be more important than another capability. A narrow use case with healthy economics may deserve investment before a broadly appealing one that depends on permanent subsidy.

In my own product leadership, I care about revenue growth and gross-margin growth together because they expose different truths. Revenue shows whether the market is responding. Margin shows whether the operating model improves as that response grows. Teams need both signals while they can still change the product, not after the commercial model has hardened around it.

There are valid reasons to subsidize early usage. A company may be learning, entering a market or funding adoption deliberately. But that should be an explicit investment with a clear economic hypothesis. It should not be hidden inside a rising engagement graph.

A product is not validated when people merely want to use it. It is validated when the company can repeatedly deliver the promised outcome at an acceptable cost. The current generation of products makes that harder to achieve, but also harder to ignore.