Every engagement ends. Most end twice. Once when the invoices stop, and again a few months later when the work quietly stops too. The second ending is the one nobody puts in the closing report.
The second ending
Ask an operator about the last big transformation programme and you will get two dates. The date it launched. The date it was signed off. Ask for a third date, the day it stopped working, and almost nobody can give you one.
That is not because it never stopped. It is because nothing marked the moment. There was no alarm, no missed number, no meeting where someone said the new way of working has lapsed. It receded. People went back to the spreadsheet they trusted, the approval they always asked for, the report they actually read. The new process stayed alive in the documentation and died in the building.
Here is the uncomfortable version. A great deal of consulting works while the consultant is in the room. The room is the intervention. Attention arrives, priorities line up, decisions that had been stuck for two quarters get made in a morning. That is real value, and it is not transferable. Take the room away and you find out how much of the change was structural and how much of it was the presence of someone whose only job was to make that change happen.
The room is the intervention. Take the room away and see what is left.
Three ways a handback fails
You handed back a document. The deck, the operating model, the playbook. All useful, none of them load bearing. A document does not make a decision, does not hold a budget, and does not turn up to the meeting where the new way of working gets quietly deprioritised. Handing back a document is handing back a description of the capability, not the capability.
You handed back to a committee. Shared ownership sounds like buy-in and behaves like a vacuum. When three people own something, the honest answer to who is accountable is nobody, and everyone in the room knows it by the second meeting. Steering groups are good at approving things. They are poor at maintaining them.
You handed back something nobody is measured on. This is the most common and the least visible. The named owner is real, willing and capable. The thing simply does not appear in their objectives, their budget or their review. So it competes for their attention against the things that do, and it loses. Not immediately. Around month four, when the quarter gets tight.
None of these are failures of goodwill. In every case the client wanted it to work. Wanting is not a mechanism.
What actually transfers
Four things, and they are unglamorous.
A named owner, with it written into their objectives. One person. Not a function, not a working group. If you cannot say the name out loud, there is no owner.
A decision right. The owner can act without asking. If the new process requires permission from the same people who were the bottleneck before, you have not changed the structure. You have added a step to it.
A budget line that survives the next planning cycle. Work funded out of the transformation pot has a fixed lifespan, and everyone involved knows the date. Work carried in the operating budget has a future. The test of whether a change is real is whether anyone is willing to pay for it out of the money they already fight over.
A review date with a number attached. Not a check in. A date, a metric, and a named person who has to explain the number.
Notice what is not on that list. Training. A centre of excellence. A community of practice. Those are ways of spreading a capability that already has an owner. They are not substitutes for one.
Three that held
Makerspace, inside Ogilvy & Mather. The unit had a repeatable unit of work rather than a mandate: ten thousand pounds, under thirty days, every time. The commercial mechanic mattered more than the creative one. Rather than asking for new budget it redirected existing budget, so instead of spending twenty or thirty thousand planning an idea, a team took ten of that and made something. Five makers on staff. Fifteen strategists trained through it, three for every maker, because the unit existed to change how people worked rather than to absorb the work. It ended by being absorbed into the way of working. Not closed, not quietly defunded. Dissolved on purpose, because making early had become how the agency worked rather than somewhere the agency sent things.
A listed UK digital services group. Harder, because the recommendation was to separate a division rather than integrate it, and a separation is worth nothing if the separated business cannot stand up alone. Losing the parent’s cover means losing the parent’s reasons to be chosen. So the second half of the work was building the strategy capability the new business would need, deliberately, as a competitive moat. It launched in March 2024 with 130 people drawn from three legacy agencies, its own brand, and no dependency on us.
Aprender, where I chair the board rather than consult. No budget at all, which is what makes it the most useful of the three. The training model cost roughly five hundred pounds per educator, because every new teacher required somebody to get on a plane. The educators already trained were being treated as the output of the model when they could be the mechanism of it. Cost per educator fell from five hundred pounds to one hundred. The board was recomposed rather than enlarged, still seven trustees, now including expertise drawn from Africa in marketing, digital, fundraising and operations. And a volunteer programme grew from two people to ten, and now runs campaigns independently.
What those three have in common is not a method. It is that in each case somebody ended up owning the thing, with a budget, a decision right, and a number they had to explain.
The leaving test
We borrow the shape of the freedom test from our method and point it at the exit instead. Four questions, and the point is to ask them in week two, not week twenty.
- Who is the named owner, and is this in their objectives? If the answer needs a sentence of explanation, the answer is no.
- Who loses if this succeeds? If the answer is the owner, you do not have an owner. More on that below, because we learned it expensively.
- If our access were switched off tomorrow, what stops? Be specific. Name the reports that go unpublished, the decisions that revert, the meeting that loses its agenda. If the honest answer is that very little stops, that is good news and you should finish early.
- What is the first decision they will make without us, and when? A handback with no scheduled first solo decision is a handback with no date.
Run these early and they are a design brief. Run them at the end and they are an autopsy.
What we get wrong
We have done all three of these. We also found a fourth, the hard way.
An American pet food business was opening in China. The US head office brought us in because the market was not moving, and in their reading the reason it was not moving was the person running it, a British sales representative based in the country.
The work was good. A master brand and positioning, a segmentation model, and the design of a social media pilot built around a group the category had written off. Pet mothers, women routinely dismissed in China as old maids, who are in fact the rescuers of hard-to-place dogs, including dogs with disabilities. Heroes rather than spinsters.
Then we handed it to the sales representative to execute.
Read that again. We handed the work to the one person in the system for whom the work was an implicit criticism. He was the named owner. He had the decision right. He also had every reason in the world for it not to succeed, because succeeding meant conceding that the previous two years had been the problem. It never launched. A change of chief executive in the US later deprioritised the market altogether, which finished it, but it had stopped moving well before that.
None of the three tests above would have caught this. There was a name. There was authority. There was even a plan. What there was not was a single question that takes ten seconds to ask.
Who loses if this succeeds? If the answer is the owner, you do not have an owner. You have a hostage.
The founder problem
In founder-led organisations the handback question is harder, because the founder is not the obstacle to the system. The founder is the system. Judgment, relationships, standards and the final word all sit in one place, and for a long time that was an advantage. It is the reason the organisation exists.
The failure mode here is not that the founder will not let go. It is that there is nothing built to let go into. You cannot delegate a capability that has never been written down, priced, staffed or governed. Which means the work is not persuasion and it is not succession planning. It is building the institution the founder has been standing in for.
That takes longer than a project. It is why it sits as its own engagement with its own clock, rather than as a workstream inside something else.
Why we design the exit first
An engagement that cannot end was never a pilot. It was an outsourcing arrangement with a strategy label on it.
So we write the exit into the work at the start. Who owns it, what they will decide without us, what budget carries it, and the date we stop. That date is not a commercial concession. It is the specification. Everything we build has to survive it, which changes what we build: fewer things, more deeply owned, wired into decisions that were happening anyway.
It also means we make less money from a client who succeeds than from one who does not. We are comfortable with that trade, because an embedded unit that never leaves is not a unit. It is a dependency, and the client has ended up paying for our presence instead of their capability. We would rather be referenced than retained.
We pilot so that failure is cheap. We hand back so that success outlives the engagement.
Third in a series of short essays answering the questions that come up most often in client work. The first asked whether your problem is structure or people. The second asked why most AI pilots fail.