There are two versions of this question, and they get very different answers.

The first is: we want evidence, and a small study is cheaper, so can we start there? The second is: we are going to run a real study eventually, and we want to de-risk it first.

The second question has a good answer. The first one runs into a structural problem, and it is worth understanding before you spend anything.

What a pilot is actually for

A pilot is an operational and statistical instrument. It answers questions about your study rather than about your product.

Each of those is worth real money, because each is a way the expensive study fails if you get it wrong. That is the case for a pilot, and it is a strong one.

Notice what is not on the list.

What a pilot cannot buy you

It cannot buy you a claim, and the reason is structural rather than about size.

The FTC's guidance sets out the basic principles it expects human clinical research to satisfy, and the first is a control group. The efficacy of a product should be demonstrated by comparing the treatment group to the control group. Improvement over time in the treatment group alone, the guidance says, could result from a placebo effect, from spontaneous changes in subjects' health, from improvements on a test measure that come purely from practice or repetition, or from other variables unrelated to the product.

A single-arm design can be the right choice for the questions a pilot is built to answer. It is also the design that cannot separate your product from those alternative explanations.

A pilot is not a regulatory shortcut either

Founders sometimes reach for a pilot on the theory that a small study attracts less scrutiny. It does not, at least not in the way they mean.

FDA's guidance addresses this directly. The number of subjects enrolled has no bearing on whether a study falls under the IND regulations, and the definition of a clinical investigation reaches studies with as few as one subject.

The same logic runs through the IRB question. Whether review is required turns on the nature of the investigation and where its results are headed, not on how many people are in it. We wrote about both separately: the IND question here and the IRB question here.

Small does not mean exempt. It just means small.

Three ways a pilot becomes a liability

1. Analyzing your way to a result. When the pre-specified endpoint does not move, the temptation is to go looking. The FTC guidance names this: analysis that departs from the original protocol can indicate data mining, and the more comparisons you examine, the more likely a significant difference is just chance. It says such analysis may identify areas for future exploration but does not generally provide reliable evidence to substantiate a claim.
2. Measuring everything. A pilot may carry a wide panel of outcomes, which is reasonable when you are learning what moves. It becomes a problem when one of eight measures hits significance and gets reported as the finding. The guidance works through exactly that scenario and points out that with eight outcomes and no statistical correction, a single positive result could easily be chance.
3. Reporting selectively. The guidance is explicit that studies using multiple outcome measures should report all of them rather than only the positive ones, and that public trial registration is a generally accepted practice partly because it reduces the chance of that happening.

Each of these turns a pilot from a useful internal instrument into something that damages you if anyone reads it carefully. And on a claim, someone will.

The risk nobody prices in

Here is the part that gets left out of the case for running a pilot, and it is the one worth sitting with.

The FTC guidance works through an advertiser with two controlled, double-blind studies showing a modest but statistically significant loss of body fat at six weeks. There is also an equally well-controlled twelve-week study showing no statistically significant difference between treatment and control. Taken together, the guidance says, the studies suggest that if the product has any effect it would be very small and may not persist. Given the totality of the evidence, the claim is unsubstantiated.

The same section states that advertisers should consider all relevant well-conducted research relating to the claimed benefit, rather than focusing only on research that supports an effect while discounting research that does not.

Which means a promising pilot followed by a null confirmatory study does not return you to where you started. It leaves you with a documented inconsistency you now have to account for, and you cannot quietly set it aside because it came out the wrong way.

That is not an argument against pilots. It is an argument for being clear-eyed that you are generating evidence, not generating an option on evidence.

That last one costs nothing. A pilot described accurately is an asset. The same pilot described as proof is the thing a competitor screenshots.

So: should you?

Run a pilot when you are genuinely uncertain about feasibility, when the protocol has a piece nobody has tested operationally, when you need to know whether your endpoint can move at all in the window you have, or when you need internal evidence to justify a bigger spend.

Skip to the controlled study when you already know the study runs, you have a defensible basis for sizing it, and the actual goal is a claim. In that case the pilot is a delay wearing the costume of prudence.

Do neither yet when the honest answer is that you do not know what you would claim. That is a positioning problem, and no study design fixes it. Work out the claim first, then work backward to the evidence it needs. We wrote about that half of the problem here.

Where we sit in this

We run pilots. So read the above knowing we have a commercial interest in you deciding to run one, and notice that the article argues a pilot cannot deliver a claim.

That is the point. The reason to run a pilot is that it makes the real study cheaper, faster, and more likely to work. The reason not to run one is to get a number for the label. If that is what you need, a pilot is the wrong instrument, and we would rather say so before you spend the money than after.