Marty Cagan: "100% predictability = 0% innovation," and why clean apps still lose users

Your app tests clean. People finish the task. Nobody complains. And they still leave. That gap between "works fine" and "worth coming back for" is where a lot of design teams are stuck right now. Let me catch you up on what the smart folks are saying about it.

The deep cut

• Polish hides a weak reason to switch. Maya liked the clean grocery app, then went back to her old one by Friday.
• Surprise only delights when the user feels in control. Nir Eyal's "reactance" explains why an Uber scenic detour is a bug, not a feature.
• Predictability is the enemy you keep rewarding. Marty Cagan says the real trade-off is outcomes versus the comfort of a shippable promise.

The familiar habit you are really competing against

The most dangerous competitor is not the other app. It is the routine the user already trusts. In the grocery-app story, Maya had no trouble searching, ordering, or checking out. She left because switching meant learning a new system for no real gain.

Easy is table stakes. A smooth interface removes obstacles, but removing obstacles is not the same as giving someone a reason to change. If your product only matches the old habit, the old habit wins, because it costs nothing to keep.

So before the next redesign, ask what the user actually came to do and whether you make it better, not just do-able. Features are easy to show in a review. A real reason to switch is harder to point at, and it is the thing that keeps people.

A feature is not a benefit, and a count is not a conclusion

Teams reach for more because features are tangible. You can sketch a saved list, demo a progress bar, ship a dashboard. But a saved list is the feature. Not re-searching for the same items every week is the benefit. If you design the feature and skip the benefit, you polish something nobody needed.

The same blur shows up in how you present data. In the sustainability platform case, the factory with the biggest emissions number looked like the biggest problem, until the team looked at emissions per unit of production and the ranking flipped. The number was accurate. It just answered a different question. One specialist put the real need plainly: "I just want to be able to scan quickly where is the most impact when I need to know."

Prominence is a design choice that assigns meaning. Before you make something the biggest thing on screen, name the question it is answering, and ask what someone might wrongly conclude from it.

Variable rewards only work when people feel in control

If you want habit, the pull of a slightly unpredictable payoff is real. Reading Nir Eyal's Hooked, one designer broke rewards into three kinds: the tribe, the hunt, and the self. Instagram sells the tribe. Swiggy sells the hunt. Duolingo made a cartoon owl protect a 47-day streak, which is more than most gym memberships pull off.

The catch is what Eyal calls reactance. People queue happily for a roller coaster but panic when a plane shakes. Same bumps, different feeling, because one was chosen. Surprise delights only when the user picked it. When the task is already uncertain, like booking a ride or tracking an order, pile on clarity instead. Swiggy and Uber do not add mystery to delivery. They show a live map so you feel in the loop.

And ask the honest question that designer landed on: does this reward help the user, or just hold their thumb? Those are different products, and users can feel the difference by week eight.

The pull toward predictability

Here is the structural version of the same problem. Marty Cagan spent a decade saying outcomes beat output, then decided he was framing it wrong. The real tension is outcomes versus predictability. Teams default to shipping what was promised because a promise feels safe, even when the result does nothing.

Cagan quotes Henrik Kniberg: "100% predictability = 0% innovation." The uncomfortable part is admitting that most ideas on your roadmap will not work, which means you have to say what you do not know. That is hard to do in a stakeholder review built around dates.

There is a training version of this trap too. Teresa Torres gets asked to teach continuous discovery in a one-hour offsite slot and says a flat no. "That's like saying, can you teach me how to be a marketer in an hour." Skill lives in a gym of repeated practice, not a single download. A talk plants a seed. It does not change behavior on its own. Polish on the training, like polish on the product, can hide that the hard work has not started.

Three questions for your team

• For the feature at the top of our roadmap, what benefit does it deliver, and what does the user gain over their current habit? If the answer is only "it is easier," that is not enough.
• Where are we adding surprise to a flow that is already uncertain, and where should we be adding visibility and control instead?
• In our next review, are we rewarding a shipped promise or a real outcome? Name one roadmap item we are keeping only because it is predictable.