A large customer agrees to a pilot. Your team builds it, it works, the users are complimentary, the demonstration goes well — and then the conversation goes quiet and reappears six months later as a request for another pilot with a different department. This is the most common way enterprise deals fail, and it almost always traces back to what was agreed before the work started.
The defining question is whether the pilot has a decision attached to it. A pilot that ends with a report is an evaluation exercise; a pilot that ends with a named person deciding to buy or not buy, by a date, against criteria written down in advance, is a purchase process. If nobody has agreed what result would cause them to buy, then no result will, and the pilot is a free proof of concept you are funding for a customer who has committed to nothing.
So write the success criteria down, with the customer, before starting. Not aspirations — measurable statements. If this reduces processing time below a stated threshold across a stated volume, we proceed to a paid deployment for this department at this approximate value. Customers who are genuinely evaluating will engage with that conversation. Customers who cannot or will not name a criterion are usually not buying, and finding that out in week one rather than month six is the most valuable thing the exercise can produce.
The second failure is running the pilot in a sandbox. A pilot on synthetic data, with a handful of friendly users, in an environment disconnected from the systems the work actually flows through, proves that your software runs. It does not answer the question the buyer needs answered, which is whether this survives contact with their real data, their real volume, and their genuinely reluctant colleagues. It also means the integration work — the part that determines the real cost and timeline — is entirely still ahead of you when the decision is made.
The third is not knowing who signs. Enterprise purchases involve a champion who wants it, an economic buyer who funds it, and several people who can stop it — security, procurement, legal, the platform team, sometimes an incumbent vendor with a relationship. A pilot that delights the champion and never meets the others is a pilot that will die quietly in a review nobody invited you to. Ask early who else must be comfortable, and get in front of them while goodwill is high.
Finally, charge for it, even a modest amount. A paid pilot changes the internal conversation on the customer's side: money spent needs a decision, and an unfunded pilot can be abandoned without anyone having to explain themselves. It also filters, in your favour, for buyers with budget and intent. Firms that move from free pilots to paid ones typically report fewer pilots and considerably more of them converting, which is the trade you want.
