Most proposals open by explaining to an organisation what it ought to want. The stronger move is to open with a sentence the organisation has already published about itself, and then treat that sentence as the brief. It costs nothing, it cannot be disputed, and it removes the only argument that ever really matters, which is whether the work should happen at all.
A European composite insurance group ran a competitive tender for the founding engagement of its new group data analytics function. The response that carried the argument spent its opening not on method and not on credentials, but on the group's own strategic choices, seven of them, set out close to verbatim. One of those choices had already committed the group to making data, skills and know-how travel between its businesses instead of stopping inside each one, and had already named the mechanism: centres of expertise, one per discipline. Another had already committed it to letting sharper customer insight drive decisions right along the value chain, with technology as the means of getting it.
So the function being stood up was not an idea anyone needed to sell. It was an outstanding promise. The head of it had been appointed. The plan had been through the management committee. Everything a proposal normally has to argue for had already been agreed, in writing, by exactly the people who would otherwise have had to be persuaded.
The challenge
A mandate that is secure at group level can still be argued to death one floor down. The engagement covered six operating companies as well as the group, each with its own market position, its own profit and loss, and its own view about how much a central function should be allowed to tell it. None of those local leadership teams had signed the group strategy in any meaningful sense. Each of them could reasonably ask why a central catalogue of analytics opportunities was relevant to a book of business it understood better than the centre did.
The tender itself made the standard harder rather than easier. It asked for three things in sequence: an assessment of the analytics opportunity and need per operating company, judged against that company's own strategy and market conditions; a prioritisation, on the client's own phrasing, based on sufficiently justifiable return rather than on technical interest; and a recommended roadmap tied to agreed value goals. Feasibility was to be scored on four named axes the client supplied itself: data, platform, model complexity, and impact on ways of working. That fourth axis is the one that decides adoption, and it is the one most assessments never write down.
The financial bar was equally explicit. The group had published a set of targets it had committed to: a combined ratio ceiling on the non-life book, an operating margin band on the life book, a return-on-equity range, a solvency ratio target and a dividend pay-out range. Five numbers, all of them public, all of them owned by the board. Any analytics case that could not be traced to one of them was, by the group's own definition, not a case.
The approach
The design followed from the quotation. If the strategy already named the capability, the engagement's job was not to justify the capability but to fill it in, and to do so in the group's own units.
That produced a plain crosswalk: the three requested areas laid against four named deliverables over three phases, one month each. Preparation would build a structured map of the insurance value chain and an opportunity catalogue indexed against it, deliberately exhaustive across underwriting, sales and marketing, and operations rather than curated down to a shortlist of favourites. Exploration would take that map into each operating company through a standardised three-day visit: a short, time-boxed workshop with the local leadership team, a pre-distributed capability review sent a week ahead so the session could argue about results instead of collecting them, and a validation session on the emerging local plan. Prioritisation would aggregate everything into one ranked log, harmonise value and effort estimates across the operating companies so that local optimism could not distort the ranking, and put the result to a leadership meeting with every operating company represented before it went anywhere near the board.
The group-level artefacts were to be validated before they travelled. Two workshops would tailor the value map to the products and segments the group actually sells, and five to ten interviews across six functions, from product development through underwriting to claims, would test it inside one cooperative operating company first. The map would then reach the harder audiences carrying evidence that it had already survived contact with a real book.
Every one of those steps is a variation on the same move. Bring the artefact, then let the client's own material tell you where it is wrong.
The outcome
This was a proposal and a method, and it should be read as one. What the engagement produced at this stage was a designed three-month plan, a scoring structure, and an argument the client had effectively written itself. No catalogue was delivered here, no operating company was visited, no case was built. The numbers above are plan numbers and published client targets, not measured results.
The finding worth carrying forward is about provenance rather than analysis. A recommendation has an author, and anything with an author can be re-argued by whoever takes the chair next. A commitment the organisation has published about itself has no such weakness. It survives the reorganisation, the new sponsor and the budget round, because withdrawing it means contradicting a document that is already outside the building.
That is more useful now than it was then, not less. Every insurer we work with is currently being asked to fund something in the agentic-systems column: copilots in underwriting and claims handling, retrieval over policy wordings and claim files, agent loops that draft and route rather than merely suggest. Almost none of those requests arrive with a mandate. They arrive as a pilot, with an enthusiast attached, and they die when the enthusiast moves. Meanwhile the same organisation has usually published, somewhere, a target on cycle time or expense ratio or customer proximity that the pilot would move directly, and nobody has connected the two documents.
The regulatory position has made the trick better. With the EU AI Act now in force, an insurer is producing a second body of writing about its own systems: classification, oversight, documentation, the graded autonomy it will and will not allow a model to exercise. Read alongside model risk management, that material is a second source of the organisation's own words, and it is unusually precise about which decisions matter. A programme of work whose scope is drawn from both, the strategy and the compliance record, arrives pre-argued from two directions.
The mechanics still have to be built. Our Platform work starts where the quotation stops: the evaluation sets that tell you whether the copilot is actually right, the MLOps and lineage that let an auditor reconstruct a decision months later, and the graded autonomy boundary between what an agent may do and what it must hand to a person. None of that is in the strategy document. All of it is what the strategy document commits the organisation to paying for.
Quoting an organisation back to itself is the cheapest mandate available and the only one that does not decay. The work of the engagement is everything after the quotation.