Why do pilots stall after going well?

Because the next step needs commitments the pilot never asked for: a named owner, a budget line, a support route, and proper handling of the cases the trial did by hand. Enthusiasm carries a trial and does not carry any of those, and no single person is usually in a position to supply them all.

What decides it before the pilot starts?

Naming who would own the result and writing down which allowances the trial is taking. Both are cheap at the start and expensive to reconstruct at the end, and together they turn the crossing from a vague next phase into a list somebody can work through.

How long does the window stay open?

Weeks rather than quarters, because the people involved move on and the result stops feeling current. A pilot that ends without a decided next step is decaying rather than paused, and restarting it later usually costs more than finishing it would have at the time.

Glossary

Pilot to production

The crossing from a trial that worked to something the organisation actually runs, which is where a large share of AI projects stop without anybody deciding to stop them.

In plain terms

The trial worked, everybody was pleased, and a year later it is still a trial. Nobody cancelled it. What happened is that the next step needed things a pilot never needed: somebody to own it, a support arrangement, a budget line, an answer about the awkward cases that were handled by hand. Each of those needs a different person to agree to something, and the enthusiasm that carried the pilot does not carry that.

Why it matters

Because the failure is invisible and therefore rarely learned from. A cancelled project gets a post mortem; a project that simply never proceeds gets absorbed into the general sense that this is difficult. The cost is real either way, and the pattern repeats because nobody wrote down what actually blocked it.

How it works

The crossing needs owners rather than approval, and the distinction is what surprises people. A successful pilot usually has sponsorship in principle and no named person accountable for the thing afterwards. Approval is easy to get and costs the giver nothing; ownership is a commitment of somebody's attention indefinitely, and it is the scarce item.

What the pilot was allowed to skip becomes the work. Exceptions handled manually to keep the trial clean, a narrow group who volunteered, material prepared by hand: each of those was a reasonable simplification and each is now a thing that has to be solved properly. Teams that recorded the allowances at the start have a list; teams that did not have to rediscover them one at a time.

The economics change shape at the crossing. A pilot's cost is small and often absorbed; production means a budget line, a renewal date and somebody defending it against alternatives. A project that was never asked what it cost is suddenly asked, at the same moment it is being asked to grow, which is a harder conversation than either would have been alone.

Momentum expires. The people who ran the pilot move on, priorities shift, and the window in which the result is fresh and somebody cares is measured in weeks. A pilot that finishes without a decided next step is not paused; it is decaying, and reviving it usually costs more than finishing it would have.

What the pilot proved, what the crossing asks

Setting the two columns side by side explains a conversation that feels baffling from inside a successful pilot. The team has answered every question it was asked, thoroughly, with evidence, and the project still does not move. That is because the questions on the right were never in scope for the exercise and are not addressed by any amount of additional proof on the left. Running a better pilot does not produce an owner. The practical consequence is that the right-hand column should be raised at the start rather than at the end, while there is still goodwill and while the answers are cheap: naming who would take it on, and writing down which simplifications the trial is taking so the list exists before anybody needs it.

Seen in the wild

An assistant trialled successfully by one team, still on individual accounts a year later because nobody owns the administered version.
ChatGPT
An automation proven on a bounded slice of records, never extended because the exceptions were handled by hand during the trial.
n8n
A search deployment that answered one department's questions well and stalled on connecting the systems the rest of the organisation uses.
Glean

Common misconceptions

People assume

A successful pilot leads to production.

In fact

It earns the right to ask, and the asking needs things the pilot did not: a named owner, a budget line, and answers for the cases the trial handled by hand. Success is necessary and it is nowhere near sufficient, which is why so many good results sit unused.

People assume

It stalled because the results were not good enough.

In fact

More often nobody would own it. The result was fine and the next step required a person to take on responsibility indefinitely, which nobody volunteered for and nobody was assigned. That reads afterwards as a technical shortfall because the real reason was never recorded.

Telling them apart

The crossing vs The pilot

The crossing

Owners, budget, support, and the cases the trial skipped.

The pilot

Whether it works with real users doing real work.

The pilot answers a question about the tool. The crossing answers questions about the organisation.

Questions

Why do pilots stall after going well?: Because the next step needs commitments the pilot never asked for: a named owner, a budget line, a support route, and proper handling of the cases the trial did by hand. Enthusiasm carries a trial and does not carry any of those, and no single person is usually in a position to supply them all.
What decides it before the pilot starts?: Naming who would own the result and writing down which allowances the trial is taking. Both are cheap at the start and expensive to reconstruct at the end, and together they turn the crossing from a vague next phase into a list somebody can work through.
How long does the window stay open?: Weeks rather than quarters, because the people involved move on and the result stops feeling current. A pilot that ends without a decided next step is decaying rather than paused, and restarting it later usually costs more than finishing it would have at the time.