Essay
Innovation Doesn't Need Courage. It Needs Smaller Bets.
The job of a design leader is not to convince an organization to take more risks. It is to make the next risk worth taking.
Haim Repael · 3 min read

STAR AT A GLANCE
- Situation
- Ambitious ideas kept meeting a rational "maybe not now" from leaders accountable for revenue and reliability.
- Task
- Shrink each commitment until the next step was safe to take.
- Action
- A prototype that could be wrong in public, a bounded experiment for designers touching code, research allowed to disagree, and a written stop condition.
- Result
- The plugin got staffed, the designer-code model spread from one squad to the others, and a counter-intuitive test kept us from a 12% conversion loss.
I've brought ambitious ideas into product organizations many times. A new technology, a different interaction model, a change to a workflow that seemed to work. The pattern was familiar: design got excited, then the questions began. What will this do to conversion? How long will it take? What if nobody uses it? Eventually someone said, "maybe not now."
For a long time I read that as resistance to innovation. I no longer do.
Situation"Not now" is often the rational answer
A successful product is not an empty canvas. It's a living system with customers attached. Product leaders answer for revenue, adoption and retention. Engineering leaders answer for reliability, architecture and the cost of every dependency. When engineering says a prototype hides more than it shows, it is usually right.
The problem starts when avoiding uncertainty becomes the default strategy. And nobody has ever been argued out of a legitimate concern about their own responsibility.
TaskReduce the decision, not the ambition
The work isn't to argue harder or raise the team's courage. It's to understand what the fear is telling you, then shrink the decision until the next step is safe to take.
ActionFour smaller bets
A prototype that could be wrong in public
For two years, an AI-powered accessibility plugin for Figma lost to safer investments. Nobody disagreed with it. It bundled adoption, feasibility, trust, infrastructure cost and roadmap fit into one question nobody could answer with a yes.
What changed was not the pitch. AI coding tools became good enough for a non-engineer like me to build with. I spent nights learning the Figma API. The first version confidently called a radio button an icon button. From there to the version I finally showed, accuracy improved by 41%.
Once the prototype was on the table, the questions got specific: where do false positives cluster, how much control does the designer keep, what's the smallest version worth shipping? One large bet became a sequence of cheap questions.
Designers touching production, with boundaries
AI tools were collapsing the cost of trying things, and designers could go beyond mockups into working code. The concern was legitimate: a designer changing CSS can delete logic or introduce invisible regressions. "Trust designers" would have deserved to fail.
So we designed the experiment with engineering, not around it. Separate branches. Small, bounded tasks. Interactions tested, not eyeballed. Design-system tokens, not arbitrary values. Every change through code review, the development lead owning the merge, and the option to shut it all down within a day.
Research allowed to disagree
When building gets cheaper, judgment gets more expensive. We tested one feature at the moment the product delivers value and asks for something in return. The obvious hypothesis: deliver value first, ask later. The data said otherwise.
Decide when you'd stop, before you start
Teams are fluent in defining success and uncomfortable defining what would make them stop. Now, before starting, we answer: what metric must not decline, what adoption justifies more investment, how much complexity we'll carry, how long the test runs, and what evidence would make us kill the idea. If that last one has no answer, it isn't an experiment yet.
ResultSmaller bets, real movement
- The plugin went from a slide nobody wanted to own to a staffed effort, and it stopped being mine a while ago, which is how these stories should end.
- The designer-code experiment passed, became routine in its squad, and the model was copied to the other squads. Not because everyone became optimistic, but because the boundaries held.
- The counter-intuitive test held. Delaying the ask cut conversion by roughly 12% relative; waiting for an even stronger engagement signal cut it nearly in half. The original control won, again. The broader rollout multiplied the target rate roughly thirtyfold and was associated with six-figure attributed purchase value.
The honest caveats: that before-and-after wasn't randomized, so it can't prove the change alone caused the lift, and attributed value isn't verified incremental revenue. It was a team result. The caveats aren't fine print. They're what makes the result usable in the next prioritization meeting.
What travels
Product owns prioritization. Engineering owns reliability. Design owns making a possible future concrete enough to react to. Innovation leadership is the three agreeing on what must be learned before the investment grows.
Reversibility is not a lack of ambition. It is what lets ambition survive contact with reality.
What moved the plugin wasn't a better argument. It was a smaller bet, a prototype that could be wrong in public, and a team willing to let it be.