Every large programme is offered the same false choice at kickoff. Either you detail the architecture, the operating model and the governance first, and start building once they are agreed, or you put something visible in front of users now and fit the foundation around it afterwards. The first route buys coherence and loses the room. The second buys attention and leaves you holding a drawer of pilots nobody can operate.
A global staffing and HR services group planning a multi-country platform programme refused the choice on the face of its own plan. The first horizon was drawn as a single page with two tracks running through the same calendar. The page said the programme drives collaboration across the country businesses through quick wins while building the required foundation, and then it laid out both sets of work in the same window rather than in sequence.
The challenge
The first horizon was cut into three stages. The opening stage is labelled the first ninety days. The second is labelled quick wins. The third is labelled develop. Nothing about that is unusual until you read across the rest of the page and see what is scheduled beside the quick wins rather than after them.
On the slow track, inside the same horizon: detail the platform and enterprise architecture. Detail the operating model and the ways of working underneath it. Detail the governance for that operating model, including the measures it would be run on. Write the service, solution and project designs and their business cases. Select the customer experience and automated workflow platforms, then procure them and start the implementation. Mobilise the delivery teams, write their charters, stand up the programme office and its reporting, begin hiring, and run the communications and cultural work alongside.
On the fast track, inside the same window: run the as-is capability gap assessment in three markets. Identify candidate quick wins. Implement them as growth-hacking experiments. Launch the resulting services. Track and measure success.
Launch and measure are the load-bearing words on that page, and they are the reason this plan is worth reading years later. A quick win that ends at a demonstration is a communications tactic. It buys a sponsor three weeks of patience and produces nothing that can be cited in an argument. A quick win that is launched as a service to real users and then measured produces something else entirely: a result, with a population and a direction attached, admissible in the business case being drafted twenty rows further up the same page.
That is the dependency almost every two-track plan gets backwards. The fast track is usually justified as cover for the slow one, a way of keeping the board interested while the architects work. Here the arrow points the other way. The business cases for the first horizon were being written in the same weeks the experiments were running, which means the experiments were the only source of first-party evidence available to them. Everything else going into those cases was vendor material and assumption.
The approach
The baselining decision is the second thing worth copying, and it is quieter. The as-is capability gap assessment was scoped to three markets, not to the whole country estate the group ran. Three assessments can be done inside ninety days by people who talk to each other. An estate-wide assessment cannot, and by the time it lands the platform selection has already happened without it.
Choosing three carries a real cost and the plan does not hide it. Markets left out of the baseline get a design informed by markets that are not theirs, and the group had already recorded elsewhere that its country businesses sat at different levels of maturity. The bet is that a representative sample delivered in time beats a complete survey delivered late, because the survey's only customer is a decision that will be taken with or without it.
The staffing logic follows the same clock. The sourcing evaluation behind the same programme put speed at the centre of the argument: contract staff were considered specifically because the transition needed to accelerate, some specialists were wanted only long enough to transfer skills, and outsourcing a brand new operation was judged risky enough to hold back rather than to buy. Retraining existing staff was kept on the table for its knowledge rather than its cost. The recommendation was a hybrid of all of it, which is what you choose when the calendar, not the budget, is the binding constraint.
One item on the plan reads as an afterthought and is not. Growth-hacking and agile training were scheduled as a deliverable of the first horizon. The fast track was not expected to run on enthusiasm. Someone had to teach the method for framing a hypothesis, shipping a scoped service and reading the result, because the people who would run the experiments had spent their careers running a placement business instead.
The outcome
Set the tense honestly. This is a planning page from a programme design. It is a sequenced intention, not a delivered outcome, and no result in the first horizon can be claimed from it. What the group had already demonstrated, on a separate and earlier run, was a single pass from problem framing to a scoring service in live use inside six weeks, worked by twenty staff, with the movement in hit rate and in user satisfaction recorded as direction rather than as a percentage. One run is not a programme. It is proof that the fast lane was not decorative.
The structure problem this plan answers has not aged, and anyone scoping AI work now is standing inside it. The foundation work is genuine and slow: data readiness, lineage you can show an auditor, the MLOps to get a model into production and keep it there, a vector store and a retrieval design that survives contact with real documents, an evaluation set that means something. None of it is optional, and none of it produces a visible result inside a quarter. Meanwhile the organisation wants copilots in front of people, and a handful of carefully scoped autonomous experiments are worth running while the ground rules for them are still being written and while the European rules that have reached political agreement are still being turned into obligations.
Run both. Put the copilot in front of one team, in one market, as a service with an owner and a measurement, not as a demonstration in a boardroom. Keep the platform work at its own pace and let the measured pilots supply the numbers its business case would otherwise have to invent. Baseline your data and capability in two or three representative units instead of surveying all of them, and accept the sampling error as the price of being ready in time. The Platform work we do and the Consult engagements that scope it are both built on that split, because the alternative is a foundation with nothing standing on it or a pile of demonstrations with nothing underneath.
The quick wins in this plan were never the point. The measurements they produced were.