Home / Blog / Enterprise Pilots
Day 26Enterprise Pilots That Convert: Escaping Pilot Purgatory
The pilot worked. Everyone was impressed. Nine months later nothing has been signed and your champion has moved teams.
A big logo agrees to a pilot and the whole company celebrates. Three months of unpaid engineering later, the results are good, the feedback is warm, and the deal enters a state that has no name in any sales methodology: technically successful, commercially dead.
MIT's NANDA initiative reported in 2025 that roughly 95 percent of enterprise generative AI pilots produced no measurable profit and loss impact. Those pilots did not fail because the technology did not work. They failed because nobody had agreed, in advance, what a good result would trigger.
A pilot without a written conversion clause is a free consulting project with a hopeful ending.
Everything that determines whether a pilot converts is decided before it starts, in the document that defines it. Once it is running, you are executing a plan somebody already wrote, and if that plan had no owner, no metric and no price, you are executing a plan to be thanked.
The four people, by name
Gartner's research puts six to ten people in a complex B2B buying group. You do not need to manage all of them, but four roles must be identified by name before you agree to anything.
The champion. Wants this to work, will spend political capital, usually the person who contacted you. Necessary and never sufficient.
The economic buyer. Controls the budget line this would come from. If your champion cannot tell you who this is, the money does not exist yet and the pilot is a research project you are funding.
The blocker. Security, legal, compliance, IT architecture. They cannot say yes and they can absolutely say no, usually in month four, on a question you could have answered in week one.
The users. The people whose daily behaviour changes. A pilot that impresses executives while the team quietly keeps using the spreadsheet has proved nothing that will survive a renewal.
Ask your champion directly: "Who else has to be comfortable with this, and can I meet them during the pilot rather than after it?" A champion who cannot or will not answer is telling you something important about their standing in the organisation.
An illustrative enterprise funnel. Note where the largest drop sits: not at the technical stage, where most startups concentrate their effort, but at "budget found," which is decided by work that should have happened before the pilot began. The highlighted step is where founders think the deal is won.
Write the conversion clause first
Before a single engineer touches this, agree one page containing six things. If the customer will not sign that page, you have learned that this is not a purchase process, and you have learned it before spending three months.
The problem, in their numbers. "Claims assessment currently takes 6.2 days on average and 14 percent are reworked." Their figure, from their system, agreed in writing.
The success criteria. Two or three, measurable, with thresholds. "Average assessment time under 2 days and rework under 8 percent, measured over the final four weeks." Not "the team finds it useful."
Who measures it. A named person on their side who owns the number, produces it from their system, and will stand behind it.
The dates. Start, end, and the date the decision is made. All three.
The price if it succeeds. The annual contract value, agreed now. This is the clause founders skip and it is the one that converts pilots. Negotiating price at the end, once you have handed over the result and have nothing left to trade, is the worst possible sequencing.
What happens if it fails. Say it plainly: if the criteria are not met, the pilot ends and nobody owes anybody anything. This costs you nothing and buys enormous credibility, because it tells a cautious buyer you are confident enough to define your own failure.
0 to 2. This is not a sales process, it is unpaid research. Decline politely or convert it into a paid discovery engagement. 3 to 4. Risky. Close items one and four before you commit engineering time, because they decide the outcome. 5 to 6. Workable. Schedule the security review this week; it is the most common silent killer. 7 or 8. Well structured. Run it, and hold the customer to the decision date as firmly as they hold you to the criteria.
Eight preconditions. Items one and four together predict conversion better than anything about your product, because they establish that money exists and that its amount is already agreed.
Survive the security review
For a small company selling to a large one, security review is where more deals die quietly than anywhere else. It arrives late, it is run by people you have not met, and it has no interest in your roadmap.
Prepare it as an asset rather than a fire drill. Build a single document containing your data handling practices, where data is stored and for how long, your subprocessors, encryption in transit and at rest, access controls, incident response process, backup and recovery, and your position on using customer data for model training. That last one now appears in nearly every review and a vague answer will stop the deal.
Formal certification helps and costs real money and months. Before you commit to that, ask three current prospects whether certification is a requirement or a preference. If it is a preference, a clear document and a willingness to answer questions directly gets you through most mid-market reviews. If it is a requirement in your segment, treat it as a cost of entry and budget for it rather than discovering it mid-deal.
Whichever route, ask for the security questionnaire in week one of the pilot. Every week you delay is a week closer to the moment when your successful pilot sits behind a queue you have not entered.
Before you agree to any pilot, send your champion this: "Imagine it is three months from now, the pilot went well, and it still did not get bought. What happened?" It is a strange email to send and it produces the most useful answer in enterprise sales. You will hear about the reorganisation nobody mentioned, the budget freeze until April, the CTO who prefers a competitor, or the fact that your champion has no authority at all. Every one of those is a condition you can plan around if you know about it in week zero, and none of them can be fixed in week twelve. Champions answer this honestly far more often than founders expect, because you have given them permission to talk about the parts of their organisation that make their own job hard.
Run it like a delivery, not a demo
Three habits separate pilots that convert from pilots that merely finish.
Report weekly, in one page, on their metric. Not features shipped. The number from their success criteria, its movement, and what you are doing next. This trains the organisation to associate you with a number, which is exactly what the champion needs when they go to ask for money.
Make the users successful, personally. Sit with them. Do the setup. Fix the friction the same day. A pilot where five users adopted the tool enthusiastically survives a budget conversation. One where the technology worked and nobody changed their behaviour does not.
Build the business case with them, not for them. Two weeks before the end, sit with the champion and construct the internal document they will present: current cost, projected cost, evidence from the pilot, risks and mitigations, requested spend. Writing it together means it uses their organisation's language and the numbers their finance team accepts. Handing them your deck means they have to translate it, and translation is where enthusiasm quietly leaks out.
Relative conversion by pilot structure. The mechanism behind the pattern is simple: a fee forces someone to find budget, and finding budget forces the organisation to identify who owns the problem. That identification is the actual product of a good pilot.
When to walk away
Saying no to a large logo is the hardest discipline in early enterprise sales, and the cheapest one to acquire.
Walk away when nobody will name the budget owner, when they want six months free "to be sure," when the required customisation only serves them, or when the security review cannot start until after the pilot. Each of those converts a sales process into a subsidy, and small companies cannot afford to subsidise large ones.
Say it plainly and without resentment: "We can run this as a paid pilot with agreed criteria and a price on success. We are not able to run an open-ended evaluation." Some prospects will leave. A meaningful number will find the budget, because the request was reasonable and you were the first vendor to hold a line. The ones who leave were going to consume a quarter and buy nothing, and you have just recovered that quarter for a customer who will actually sign.
The takeaway
- Identify four people by name: champion, economic buyer, blocker, users. A champion alone is not a deal.
- Write the conversion clause before you start: problem in their numbers, thresholds, owner, dates, price on success, and what failure means.
- Charge something. The fee forces the organisation to find out who owns the problem.
- Start the security review in week one. It kills more deals than the product ever does.
- Report weekly on their metric, and write the internal business case with the champion, not for them.
- Walk away from open-ended evaluations. A quarter spent on a subsidy is a quarter not spent on a customer.
Frequently asked questions
Why do enterprise pilots fail to convert?
Because they prove capability without replacing a process, and because nobody agreed in advance what success would trigger. MIT's NANDA report found roughly 95 percent of enterprise generative AI pilots produced no measurable P&L impact. The pattern is an enthusiast without budget, criteria written after the fact, and a budget round already spoken for.
Should startups charge for pilots?
Almost always. A fee forces someone to find budget, which forces the organisation to identify who owns the problem. Free pilots attract curiosity, get deprioritised, and set a precedent that your work has no price. Keep it modest and credit it against year one.
How long should an enterprise pilot run?
Long enough to produce a measurable result, short enough to finish inside one budget cycle: usually six to twelve weeks. Longer pilots accumulate risk as champions move and priorities shift. Put the end date and the decision date in writing.
Who needs to be involved in an enterprise purchase?
Gartner finds six to ten decision makers in complex B2B purchases. You need four named: the champion, the economic buyer, the technical or security reviewer who can block it, and the users whose behaviour must change. Engaging only the champion means passing the pilot and losing the purchase.
Two logos, nine months, no revenue.
Kill My Startup covers the deals that consume a company's runway while looking like progress on every slide.
Buy on Amazon →Sources
- MIT Project NANDA, "The GenAI Divide: State of AI in Business 2025," on enterprise generative AI pilots and measured business impact.
- Gartner B2B buying research on buying group size in complex purchases.
- Standard enterprise vendor security review practice: data handling, subprocessors, retention, incident response.
- Common commercial structures for paid proof of concept and conversion clauses.