Where Prep Stops Helping: Finding Diminishing Returns Before They Find You

I’ve already written about the risk math behind why people stall before starting. This is the practical follow-up, for the part nobody warns you about: knowing when to stop testing isn’t obvious even once you’re already moving, and the signal that you’ve crossed the line is quieter than you’d expect.

Where iteration actually worked

Early in developing Celeste Naturals’ products, testing was doing real work. Superfat research, cycling through oil combinations to land the right balance between moisturizing and cleansing, running new scents to build out a range of notes for customers. Every batch told us something. Each round of testing got us measurably closer to a product we’d call a real MVP, and we could run all of it in a silo, just the two of us, because the answers were things we were actually equipped to judge ourselves. Does this bar clean well without stripping skin? Does this oil ratio feel right? Does this scent hold up? We didn’t need anyone else in the room to know if we’d made something better.

Where it quietly stopped working

Then the questions got smaller and the answers got fuzzier at the same time. Should we bevel the corners of the soap? Is the bar 4.5 ounces or 5? How many citrus notes do we actually need across the lineup? Which products should share a scent and which shouldn’t?

Every one of those felt like the same kind of progress we’d been making all along. It wasn’t. It was perfectionism at work, dressed up as diligence. The tell, in hindsight, was that we could no longer say with any confidence whether a given test had actually improved the product, or changed it at all. Early testing produced a clear verdict every time. Late testing produced a shrug. That shift from “clearly better” to “really can’t tell” is the actual signal, and it’s easy to miss because it doesn’t feel like failure. It feels like diligence.

The blind spot iteration can’t fix

Here’s what it came down to: a business doesn’t operate in a silo, no matter how long you keep testing like it does. We could run batches indefinitely and never be selling to anyone but ourselves, and we are, structurally, the worst possible judges of some of these questions, because we already know what we’re holding before we pick it up.

We didn’t find out the bar was a little too big for a comfortable grip, or that the edges were sharper than they should be, from another test batch. We found out because we finally put it in customers’ hands and they told us. No number of additional rounds in our own kitchen would have surfaced either problem, the same way no amount of staring at our own packaging told us the labels were hard to read. Some information only exists on the other side of a stranger’s hands, not in another iteration of your own.

How to actually tell when to stop

The practical version of this, for anyone mid-development and unsure if they’re still improving or just stalling politely: ask whether you can still tell, plainly, if your last change made the product better. If yes, keep going, you’re still extracting real signal. If every answer starts feeling like a coin flip, bevel or no bevel, half an ounce heavier or lighter, that’s not a sign you need more data from more testing. It’s a sign you’re asking questions your own testing was never going to be able to answer, and the only way forward is getting it in front of the people you’re actually building it for.

Prep is genuinely useful, right up until the point where it starts pretending it can answer questions it structurally can’t. After that, it’s not caution anymore. It’s just a well-organized way of putting off finding out.