Most organisations choose an operating model for a new capability by asking how fast it can be stood up. It is the question with the cleanest answer, it is the one a sponsor asks first, and in this review it was the only criterion of four the recommended design did not win.
A global staffing and HR services group was integrating two already-approved business designs onto a single platform, one of them relationship-led and one automated end to end. Before anything could be built, the group had to settle where digital capability would sit: in a centre, in the country markets, or split between them. Four candidate designs were assembled and each was scored on the same four criteria, on a low, medium, high scale: speed to mobilise, speed to market, investment efficiency, and governance.
The challenge
The country markets were not empty. That is where the design work started, and it is what most centre-first designs quietly assume away. Local teams were already mobilised, local capability was already maturing, and good practice was already being executed in market. A design that treats the centre as the origin of all capability has to overwrite that before it produces anything of its own.
The four candidates were not four philosophies. They were four subsets of the same building blocks: a leader accountable for experience and data, a shared delivery unit, group-level forums, and a network of local advocates. Placing them took two axes rather than one, because how much capability sits in the centre and how much coordination the centre runs are independent variables. Confusing those two is what produces a centre nobody uses.
The weakest candidate stood up a central function to publish standards to the country teams, with no local advocates engaged at all. It is quick to mobilise. It also scored at the top of the low band on all four criteria, roughly a third of the way up the scale on every one, the flattest profile in the set. The reasoning given: a central function of this shape becomes very good at producing guidelines and poor at getting them adopted, and unless marketing, technology and operations are already centralised it cannot reach a local market at all.
The second candidate added a delivery unit staffed from the centre, so the centre could build tools and platforms for others to use, with close control over development and execution alike. It reads around 97 percent of the scale on speed to mobilise, tied for the highest single reading anywhere in the comparison. On speed to market it reads around 32 percent, the same low band as the design with no delivery capability at all. Around sixty-four points of scale separate how fast that design can be created from how fast it can ship. The stated reason is mechanical rather than cultural: with no engagement in the country markets, solutions miss local needs on customer, commercial and regulatory grounds.
The third candidate removed central delivery capability entirely and kept only standards, set by technical, data and content panels and enforced through local advocates. It posts the best governance reading of the three rejected designs, around 58 percent, with speed to market and investment efficiency both stuck around 32. The named defect is precise: there was no forum for enterprise-wide investment. The centre had no money to allocate, so it had no teeth, and a standards body without teeth becomes a talking shop.
The approach
The recommended design keeps real delivery capability in the centre and changes two things. The group-level forums are staffed with local people rather than central ones, and a network of local advocates connects the centre to the markets. The centre coordinates business strategy, the measures everyone is held to, and the technical and content standards. It buys once for the whole group, adopting platforms and shared skills no single market could fund alone. Capability grows locally as well as centrally, deliberately, rather than being concentrated.
On the four criteria it reads around 84 percent on speed to market, around 97 percent on investment efficiency, tied for the highest single reading in the whole comparison, and around 77 percent on governance. On speed to mobilise it reads around 40 percent, low in the medium band. The design that mobilises fastest reads 97 on that same dial. So the recommendation asks the group to give up roughly 57 points of stand-up speed to win three other criteria.
That trade has to be argued out loud, because it will not survive being glossed over. A sponsor who has promised a capability by quarter end will read 40 and stop reading. The honest counter is that speed to mobilise measures how quickly the thing can be created, while the other three measure what it does once it exists. Only one of the four stops mattering after the first quarter. Presenting the recommendation as a clean win invites the first person who finds the 40 to treat the whole comparison as advocacy.
The distance between the two strongest candidates is small enough to state in one sentence and expensive enough to fight over for a quarter. Forums staffed locally instead of centrally, plus an advocate network. The platform does not change. What changes is who has standing where standards get set, and that is what the delivery reading is tracking.
The outcome
What this phase produced was a scored comparison and a recommendation, not an implemented operating model. Everything above is readings against stated criteria, not a measured result. Alongside the recommendation came a skills landscape: the named roles across creative, technology, data, operations and management that the group would have to grow and add to the organisation before any of this could be staffed.
Hold that comparison against how groups are currently standing up a copilot capability and the same shape appears. A centre gets created quickly, because creating it quickly is the visible win. It buys the vector store, writes the retrieval guidance, assembles the evaluation sets, and publishes all of it to markets that were never in the room. Adoption then arrives as a communications problem, which it is not.
The data argument is sharper still. Retrieval quality is a function of what the local systems hold, how clean it is, and whether its lineage runs back to a source someone will stand behind. Data readiness lives in the country markets, and so does the regulatory variation. The political agreement reached on the European AI Act tells every group in this position that documentation, traceability and human oversight are coming, and that those duties will land on records held locally rather than on a central policy page. A centre with no standing in the markets can write the policy and cannot produce the evidence.
This is where an operating design turns into work we do. Deciding to coordinate rather than to own is a two-hour decision. Wiring a shared evaluation set that local teams will run against their own data, building a lineage trail that survives an audit, and making the shared platform cheaper for a market to adopt than to bypass, is a build. Our Platform work starts where the centre has to earn adoption instead of mandating it. Consult starts earlier, at the scored comparison, because a design chosen on stand-up speed alone is a design nobody scored.
The early and carefully scoped autonomous experiments coming out of these programmes make the trade more acute, not less. A system allowed to act on its own inside one market's process needs local ownership of the guardrails, local evaluation, and a governance forum that controls real money. A centre reading 32 on delivery supplies none of those.
The one criterion the recommended design loses is the one that stops mattering soonest. Say that to a board in those words, because the alternative is to bury the 40 and let someone else find it.