What an honest pilot proposal looks like.

Most pilot proposals are written to make the buyer feel safe. The honest ones make the buyer feel informed. The difference is one section, and that section is almost always missing.

I have read perhaps two hundred pilot proposals in the last decade — written some of them, evaluated more, and watched several of the worst ones get signed without anyone reading past the cover page. They tend to look the same. Three pages of executive summary, a feature matrix, a deployment timeline, a price, two case studies that are vaguely relevant. Sometimes a Gantt chart. Always a logo wall.

What is almost never in them is the section I want to argue for in this essay. The list of things the pilot will not show, the conditions under which it will fail, and the specific ranges of outcome the seller actually expects. That section is the difference between a proposal that helps a sophisticated buyer make a decision and a proposal that helps a salesperson hit a quota. The two are not the same.

01 What the missing section looks like

Concretely, an honest pilot proposal includes a sub-page titled something like "Limits and expected ranges," and that sub-page contains:

• Three failure modes the pilot specifically will not catch. Not "this isn't designed for X" buried in fine print — a clear list of things the customer should expect not to see signal on, and why. Acoustic monitoring will not catch a fault with no acoustic signature; an optimisation framework will not produce a defensible recommendation if the objective set is wrong; an LLM-based tool will not produce a useful answer if the document corpus does not contain the answer. Say it at the front, in the customer's language.
• An expected outcome range with a wider lower bound than the seller is comfortable with. Not "we expect a 30% improvement"; rather, "we expect somewhere between zero and forty percent improvement on this primary metric, with a most-likely outcome around fifteen, and a one-in-five chance of zero or worse for reasons we cannot anticipate." A real range, with a real left tail. This is the part most sellers refuse to write because it threatens the close. Refusing to write it is the most important predictor of a pilot that disappoints.
• A list of conditions under which the seller will recommend stopping the pilot early. Specific tripwires: "if X is not true after week three, we will recommend ending the pilot rather than extending it." The customer should not have to argue you out of an extension. The proposal should commit the seller to a kill criterion in writing.

None of this is novel as a discipline. Researchers write power calculations and pre-register failure modes; clinical trials have stopping rules; underwriters quote ranges, not points. The business-software pilot proposal has somehow escaped the same hygiene. It costs nothing to import.

02 Why the section is missing

The honest answer is that pilots are usually sold to non-buyers — to internal champions who need the deal to land in order to look good, or to procurement teams whose job is to extract the lowest price for the most promised features. Either of these audiences punishes the seller for stating limits openly. The seller learns. The proposal gets cleaned up. The Limits section disappears.

The result is a market in which proposals that lie by omission outperform proposals that tell the truth. The buyer reads two proposals; one promises 30% and one promises "between zero and forty, most likely fifteen, with the following risks"; the buyer's procurement function chooses 30%. The honest proposal lost on its honesty. This happens, in my experience, about eighty percent of the time.

The proposal that promises a range will lose to the one that promises a point. It will also be the one that succeeds. — working observation, ~200 pilots

The other twenty percent of the time, something interesting happens. The buyer reads both, notices that one of them has visibly thought about failure modes and the other has not, and updates aggressively in favour of the honest one. These buyers are mostly senior technical people, often founders, occasionally CTOs whose careers were built on shipping rather than selling. They recognise what the Limits section signals — that the seller has actually deployed the thing before, has watched it disappoint a customer, and has the institutional memory to tell you about it on first meeting.

Those twenty percent of buyers are the customers worth winning. The pilots they sign turn into multi-year engagements; the projects deliver; the testimonials are real. The proposal that costs you the eighty percent is the one that earns the twenty.

03 A small worked example

Consider a sensor-optimisation pilot — the kind of work I do. The dishonest version goes: "We will deliver a Pareto front of optimal sensor configurations, identify a recommended knee, and demonstrate a typical reduction of 30% in sensor count at equivalent diagnostic capability." Three sentences, an implied promise of 30%, no failure modes named.

The honest version says: "We will deliver a Pareto front of non-dominated sensor configurations across the four objectives we agree at the start of the pilot. The recommended knee is whichever point on the front balances those four objectives at the weighting you specify; if the weighting changes after delivery, the knee moves and we will not redo the analysis without a contract amendment. We expect, based on three previous engagements of this kind, a sensor-count reduction in the range of fifteen to thirty-five percent at the same diagnostic capability — but on subsystems where the reference network was already well-designed, we have seen no reduction at all, and on one subsystem where the failure-mode list was not stable, the recommendation shifted week-to-week and we ended the pilot at week six rather than extending. If by week three of this pilot the failure-mode list is still being argued, we will recommend ending early."

That second version is approximately three times longer. It also closes a third as often. The deals it closes are the ones that succeed.

04 What this implies for buyers

The corollary, which is the part I would offer to anyone evaluating pilot proposals, is short: refuse to evaluate proposals that do not contain a Limits section. If the proposal you have in front of you has no list of failure modes, no expected range with a real left tail, and no kill criterion, send it back. Ask for those three things. The seller's response — willingness, evasion, refusal — tells you most of what you need to know about the underlying capability.

If the buyer's procurement function will not let you do this — if every proposal must be evaluated on price and feature count alone — that is a separate problem, and the right one to escalate to whoever can fix it. The cost of not fixing it is the disappointing pilots you keep signing.

05 What I will not promise in this essay

To name three: I will not promise that doing this changes the outcome of every pilot you run; some pilots are bad ideas regardless of how they are scoped, and a Limits section will not save them. I will not promise that the discipline scales to every kind of work; some categories of pilot — pure exploration, R&D collaborations, novel-method validations — genuinely cannot quote ranges, and a proposal that pretends otherwise is being equally dishonest. I will not promise that the buyers you most want to work with are the ones who reward this; some excellent buyers will read a proposal with a Limits section and pass on the discomfort of seeing the failure modes named.

The argument is narrower than that. The proposals that contain a Limits section produce, on average, better outcomes for the customers that sign them. Stronger claim than that, I cannot defend. Smaller claim than that, I do not need to make.