Capability transfer is normally the last paragraph of a delivery plan. It arrives as a workstream in the closing fortnight, called handover or knowledge transfer, and it consists of documentation, a walkthrough and a shared drive with the working files in it. By then the work has been done by the people who are leaving. That is exactly why there is something to hand over.
A competitive proposal I worked on made the opposite commitment, and made it in the one place that is expensive to break later. Not in the closing section, where intentions live, but in the staffing line, where the team composition and the scope boundary sit and where a buyer can hold a supplier to what is written.
The client was a European composite insurance group standing up its first group-level data analytics centre of expertise. The response to the request ran to four sentences. Read together they describe a mixed team, an overlapping schedule and one element of the work the supplier declined to own.
Everything below is a bid document. No team was mobilised in the copy I hold, no operating company was visited, and the client signature block is blank.
The challenge
The group had already decided to build the function, appointed someone to head it, and taken the plan through its management committee. The mandate was secure before the tender went out. What did not yet exist was the function itself: the people, the working habits, and the standing to tell six national operating companies something they did not already believe about their own books.
That is the position every organisation is in the first time it centralises a capability. It has a decision and a job title. It then buys a piece of consulting work to produce the first artefact, which here was a map of where data could move value across the insurance value chain and a ranked plan the board could fund against its published financial targets. The artefact is the deliverable. The function is what has to exist afterwards, and a conventional delivery plan does not produce it as a by-product.
The tender itself pushed in the wrong direction. It asked for three ordered areas of support. Prepare, meaning assess the analytics opportunity and need for each operating company against that company's own strategy and market conditions, its potential business impact in quantitative and qualitative terms, the feasibility of the elements involved and the cost of implementation. Prioritise, meaning rank on sufficiently justifiable return and identify the resource and means gaps at operating company and group level. Recommend, meaning a roadmap tied to agreed value goals. Read literally, that is a relay. Finish one leg, pass it to the next.
A relay is the worst possible shape for capability transfer, because the phase that teaches the most is the last one. Prioritisation is where the arguments happen, where a market's favourite idea gets ranked below something nobody local proposed, and where the reasoning that a permanent function needs to inherit is actually generated. Run it last and the client staff meet it as a conclusion. The governance half of this problem is published separately, in Six Operating Companies, Three Months and One Governance Body the Client Chairs, which solves it by putting the future owner of the function in the chair. This piece is about the other half: who is standing next to whom while the work is being done.
The approach
The first commitment was a highly capable and experienced cross-functional team working alongside client staff to deliver the immediate requirements whilst transferring capability to those staff from the outset. Two promises in one sentence, and the order of them matters. The immediate requirement still gets delivered, so this is not a training exercise with an artefact attached. But the people who will own the function afterwards are inside the delivery rather than downstream of it.
The second commitment is what makes the first one mechanically possible. The team would work across the three requested areas together, so that what each area taught fed the others and the same resources served all three, and would run the activity in parallel over the period rather than one area at a time.
The calendar left no alternative. Preparation was priced at five weeks run from the group centre, prioritisation at three weeks, and exploration at a week per operating company. Five and three is eight of the roughly thirteen weeks in a three-month window before a single country is counted, and six operating companies at a priced week each takes the total to about fourteen. The plan closes that gap by compressing each visit to three days on site and putting all six of them inside a single month, which only works if preparation is still running while the visits are underway and the aggregation has begun before the last one closes. Sequence was not on offer. What the bid did was write the constraint as a learning argument rather than a scheduling one, which is the honest version: if the value map is still being revised while the visits are running, the sixth operating company gets a better instrument than the first, and the people carrying it have watched it change.
A relay hands the client a conclusion. An overlapping schedule hands them the argument that produced it, which is the part that has to survive.
The third commitment was to work closely with client teams to incorporate their input across the data, systems and process elements. The fourth was to adopt a collaborative role in informing the client on the people and organisation element, at operating company and group level.
Four elements, and the fourth is treated differently in writing. On data, systems and process the supplier co-produces. On people and organisation it informs, and the client decides. That is the most commercially restrained line on the page, because organisation design is the follow-on every consultancy wants. It is also the line that makes the transfer promise credible rather than decorative. A supplier that draws the client's organisation chart has written itself into the operating model it is about to leave, and every subsequent conversation about whether the function is adequately resourced becomes a conversation with a party that benefits from the answer.
The outcome
There is no outcome. This is a competitive response, and the commitments above are offered terms. Nothing here was built, no capability was transferred to anyone, and the durations are planned durations from the same document's commercial section rather than elapsed ones.
What survives as a finding is the placement. Capability transfer written into the staffing composition and the scope boundary is a contractual fact. Written into a closing section it is a sentiment, and sentiments lose to schedule pressure in the final fortnight of every engagement I have seen.
That matters more now than it did then, because the artefacts are harder to inherit. A value map is a document, and a document can be read by whoever ends up owning it. An evaluation set, a retrieval index over policy wordings and claim files, a prompt suite with its regression history, and a graded autonomy boundary between what an agent may do and what it must route to a person, none of those are documents. They are operating assets with a maintenance burden, and the knowledge that makes them safe to change sits in the arguments that produced them rather than in the artefacts themselves. If the supplier's engineers are the only people who have ever changed the retrieval configuration, the capability the insurer thinks it has stood up is a procurement relationship.
The fourth element is where the current version goes wrong most reliably. Suppliers arrive with an opinion about the AI operating model: whether there is a central function or a federated one, who owns model risk, where the platform team reports, which decisions need a human in the loop and which need only a log. That opinion is worth hearing and it is not the supplier's to settle. The organisation design determines who can be held accountable when a model does something expensive, and accountability is not a thing you can buy in.
- 3
- Areas run in parallel
- 4
- Elements in the response
- 3
- Co-produced with the client
- 1
- Informed only
The bid answered that by naming its role on the fourth element as informing, in writing, in a document the client could quote back. On the other three it put its people beside the client's from the first week. The pack does carry a knowledge-sharing section further back, four platform archetypes for putting the engagement's cook book and what the team worked out in front of staff through self-service portals, but that section is about where outputs get filed rather than about who produces them. The plan itself has no handover row and no knowledge-transfer phase at the end. The transfer commitment lives entirely in the staffing composition and the scope boundary, which is the point.
Source: the understanding slide of a competitive proposal written for the founding engagement of a European composite insurance group's group-level data analytics function, which sets the client's three requested areas beside the four sentences of the supplier's response. It is a proposal and not a delivered engagement: the mixed team, the parallel running and the scope boundary are offered terms, and the copy I hold carries no evidence that any of it was mobilised. The phase durations are priced figures from the same document's commercial section, and the three-day visit length is from its planning slide. The reading that the informing-only line on people and organisation is the load-bearing part of the transfer promise, and the arithmetic showing the calendar left no room for a sequential run, are mine.