How to Close a Pilot Project
In AI pilots, the least-taught skill is not launching but closing. Launching has a ceremony — the announcement, the demo, the first week’s buzz. Closing has none, so pilots do not close; they wilt. And a wilting pilot consumes two assets worth more than budget: the team’s attention and the organisation’s appetite for experiments. Neither appears on a balance sheet; their loss is noticed only when the next good idea fails to reach the agenda.
This piece takes closing out of the defeat ceremony and into management routine: how the closing decision is reached, how the shutdown is executed, and how a record is extracted from what ended. The aim is to make closing ordinary — an ordinary closing is what emboldens the next attempt.
What Is the Problem?
BU BÖLÜMÜN ÖZETİ
- Suspended projects occupy space
- Shutdown authority is undefined
- The word “failure” locks the door
An unclosed pilot is an invisible expense line, and it damages in three ways.
Suspended projects occupy space
The project that is neither scaled nor closed claims one sentence in every planning meeting: “we’re still waiting on that one.” The sentence is innocent and it compounds; three suspended projects effectively block a new idea from entering the room. Agenda is a finite resource like budget, and suspended work borrows it interest-free.
Shutdown authority is undefined
A Forbes analysis of stalled deployments lists a striking item in the typical picture: no human is defined with the authority to stop the system when it drifts. Where stopping authority is undefined, the closing decision is ownerless — a project everyone may close, no one closes. Authority is defined on the pilot’s first day, because in a crisis authority is used, not searched for.
The word “failure” locks the door
As long as closing counts as declaring failure, everyone defers the declaration to the next quarter. Deferral does not remove the failure; it merely charges interest on the bill. Changing the word works: a closure is not a failure announcement but a measurement result — nobody blames the thermometer.
Why Does It Happen?
BU BÖLÜMÜN ÖZETİ
- Sunk cost does the talking
- No closing criterion was written at the start
- The closing has no addressee
The roots of not-closing are structural, not emotional.
Sunk cost does the talking
“We’ve put in so much — let’s try a little longer” is spent money steering a future decision. What is spent does not return; the only input to the decision is what will be spent from here. Everyone knows this principle, and everyone forgets it in the meeting room. The fix is mechanical rather than motivational: the decision meeting shows no past-spend column at all — only forward cost is on the table.
No closing criterion was written at the start
If launch day discussed only the success threshold, failure has no definition — and the undefined cannot be declared. The abandonment threshold inside the measure definition is precisely the key to this door.
The closing has no addressee
Scaling has a natural counterpart: whoever opens the budget. Closing, in most organisations, has none; the decision wanders from desk to desk, and a wandering decision is an untaken one. The owner line on the one page fixes the addressee of both decisions in advance — naming it loads no new work on anyone; it writes the address of a decision that already exists.
How Is It Done?
BU BÖLÜMÜN ÖZETİ
- Stage 1: declare the decision
- Stage 2: release the resources
- Stage 3: write the five-line closing record
A clean closure has three stages; together they take under a week.
Stage 1: declare the decision
On decision day, a pilot below the abandonment threshold closes the same day, and the closure is declared in one written sentence: which measure, which threshold, which result. The declaration hunts no culprit; it shows a number. Nobody loses in a project closed by a number; everybody loses in one wilted by silence. Written matters: a spoken closure lets everyone author their own version; a written sentence leaves one.
Stage 2: release the resources
Licences cancelled, access revoked, integrations unwound, the project’s blocks deleted from people’s calendars. Skip this stage and the project closes on paper while living on in expenses — a half-closed pilot costs more than a fully open one, because the open one is at least watched, while the half-closed one is merely paid.
Stage 3: write the five-line closing record
Five lines: what was tried, what was measured, what came out, what was learned, and under which condition it will be tried again. The fifth line rescues the closure from becoming a permanent verdict — the project does not die; it is bound to a condition. The record enters the organisation’s memory of experiments and is opened whenever the next one page is written, so each closure becomes the next attempt’s starting capital and no lesson is purchased twice.
How Long, at What Cost?
BU BÖLÜMÜN ÖZETİ
- One week, three documents
- Return: appetite preserved
- Return: memory accumulated
The closure’s cost is small; its return compounds.
One week, three documents
The declaration sentence, the resource list and the closing record — all written from existing knowledge, no research required. The difficulty sits in the decision, not the writing; and if the threshold was written at the start, the decision was taken long ago. A closing week that stretches usually means an unwritten threshold is being negotiated now.
Return: appetite preserved
A cleanly closed pilot shows the team the next attempt will be judged fairly too. That is how experiment appetite survives — and in a field where the crossing rate is this low, appetite is the scarcest input a company has.
Return: memory accumulated
As five-line records pile up, the organisation starts seeing its own pattern: which kinds of work cross, which stall under which condition. That pattern is the one consultancy money cannot buy — industry reports describe the average; closing records describe you.
The Common Mistake
BU BÖLÜMÜN ÖZETİ
- Shrinking instead of closing
- Muting the closure
- The record turning into a charge sheet
Closures happen — and get spoiled in three ways.
Shrinking instead of closing
Moving the project “to the background” instead of closing it is the polite form of not deciding. A background project is not measured, an unmeasured project teaches nothing; its one remaining function is not being closed.
Muting the closure
An unannounced closure spreads by corridor rumour, and rumour is always crueller than the number. One written sentence is cheaper than a hundred spoken ones.
The record turning into a charge sheet
If the closing record slides toward “who erred”, nobody writes true numbers into the next pilot. The day a record system starts judging is the day it stops collecting data.
Frequently Asked Questions
Sık Sorulan Sorular
Yes; the contract calendar and the decision calendar are separate things. The remaining term is either redeployed to another use case, or the early-exit cost is weighed against the cost of staying suspended and paid — the exit is usually the cheaper item. The next contract’s early-exit clause is then negotiated with this experience in hand.
Belief goes into the record’s fifth line: which condition, once changed, reopens the file? Converting belief into a condition does not dismiss it; it makes it retriable — the condition is written concretely: which tool capability, which data readiness, which volume of work. Unconditional belief, by contrast, melts thresholds.
In the same sentence used internally: we tried, we measured, it fell short of the threshold, we will look again under this condition. A business that closes by measurement does not hide the sentence — it is the one sentence a competitor who continues without measuring cannot construct.