A staffing and HR services group sells one argument for a living. Work can be done by the people you already employ, by people you hire permanently, by people you rent for a while, or by somebody else's organisation entirely, and a staffing firm has a commercial stake in the middle two. When such a group set out to build an integrated digital platform across national markets that had each run their own systems for years, it had to make that argument to itself, in a room where the cost of every answer landed on its own budget instead of on a client's.
The blueprint that came out of it contains two documents worth reading together. One is a placement map: every role the new capability needs, and against each role, whether it sits in the centre, in a large country market, or in a small one. The other is a sourcing evaluation, four routes to those people written against the design principles the programme had already published. Apart they look like ordinary organisation design. Together they describe a company rebuilding its own commercial product inside its own walls, then declining to price it.
The challenge
The placement map is where the argument shows. Take the technology family: architects, engineers, developers of several kinds, testers, security. Every one is placed in the centre. Not one is placed in a small market, and the few marked for a large market carry the qualifier possible. The design is unambiguous about what a small country market gets. Its software is built by the centre, or it is not built.
The data family runs the same way, with two exceptions that are precise about why they exist. The only data roles reaching a small market are the ones that have to: a compliance role, present where local regulation differs, and a stewardship role, present as a role rather than a job, so someone local carries the responsibility without the post existing. Operations, the smallest family, reaches no small market at all.
The creative family inverts all of it. Those roles are placed locally almost without exception, and two of the twelve are the only entries anywhere on the map where the centre says no outright: day-to-day digital marketing management and the management of online communities. The two jobs the design refuses to centralise are the two that require standing close to a local audience and knowing what it responds to.
Now the part that is easy to read past. Eleven central entries do not simply say yes. They say yes and then name a consumer: the centre holds the capability, sets the practice, and the smaller markets draw on it. Three of the eleven go further still: they are held centrally only for the small markets to use, and all three are content production jobs. The centre carries a copywriter, a content designer and an online editor that the centre itself does not need.
That is not a shared service in the usual sense, where a group function does work the group requires. It is a pool of capacity held for internal customers who cannot justify the post themselves, drawn on when a piece of work appears. Which is, sentence for sentence, the proposition the group sells to its own clients: do not hire a specialist for a role you need six weeks a year, rent one from somebody who keeps a bench.
The approach
The sourcing question was put separately and answered in the same spirit. Four routes were written out: retrain existing staff, hire permanently, hire on contract, outsource. The programme's published design principles became the rows, and every cell was a written sentence rather than a score, which is the right choice for an argument that has to survive disagreement. The recommendation was a hybrid, several routes running in parallel, with outsourcing held back until the operation was running rather than ruled out for good.
What is missing from that page is a price. Not one of the four routes carries a rate, a duration or a total. The cost material sitting behind the first horizon names people, internal and external, and technology as the key drivers, and it splits those costs sensibly across one-off and recurring, from design, testing, training and coordination through to licensing, support and the day-to-day running of the platform. It is a good taxonomy of where money goes. It is not a number.
The only figure anywhere in that costing material is a comparator, and it belongs to a different company. Roughly 1.7 million pounds, for a first horizon covering a hosted experience platform, a new operating model and content built to a common taxonomy, at an organisation unrelated to this one. It is there as an order of magnitude and it is honest about being that. It is also the whole finding in one slide. A staffing business, asked what a capability costs a client, answers in rates and durations, because rates and durations are the units its product is sold in. Asked what the same capability costs itself, it answered in drivers and borrowed somebody else's total.
The outcome
What this phase produced is a blueprint: a placement map with every role assigned, a sourcing recommendation, and a first-horizon plan that starts hiring while quick wins run alongside. No organisation was hired here. The counts above are counts of decisions written on paper, not of people who turned up.
Three things would be done differently if the same design ran now.
The placement map is a demand forecast produced by hand, from workshops. This group holds, in systems it already owns, the closest thing to ground truth about which skills are scarce in which country and how long a role of a given shape takes to fill. None of it reached its own map. Joining internal requisition history to the role list is a join across two datasets the company already holds, and it turns every placement from an assertion into an estimate with a source attached. Data readiness is the one real obstacle, and it is smaller here than in most industries, because this company's operational records are records of placements.
The line that says a central capability exists for the smaller markets to use is where current tooling changes the arithmetic, carefully. The constraint in a small market is access to a scarce specialist for a short task: someone to check a page against the group's taxonomy, or to read a dashboard and say what it means. A retrieval-augmented assistant over the group's own content, taxonomies and market data moves a portion of that work back into local hands and leaves the judgement in the centre. It only works where the surrounding discipline exists: consistent taxonomies, lineage on where each figure came from, and an evaluation set per task, so somebody can say whether the assistant is right often enough to be trusted. Most disappointments we are asked to diagnose start with an assistant shipped without one, which means nobody can tell a good week from a bad one. Early autonomous work belongs here too, scoped tightly, where a wrong answer costs a review cycle rather than a hire.
And the placement map should now carry one more column. Recruitment and staffing sit among the employment-related uses that the European rules on artificial intelligence now taking shape treat as high risk. Nothing is required of anyone yet, but the design decision is live: whether each role's work will feed a system that has to be explainable to a candidate. That is cheap to settle while the organisation is a diagram and expensive once it is a payroll. Our Consult work now opens on that column, because it reorders a hiring plan more often than clients expect.
The argument this group sells is that you should not own capability you use intermittently. Applied inward, it produced a centre that lends people to its own small markets, and stopped one step short of what makes the argument work outside: a price on the loan.