A CTO commits to a three-year architecture roadmap with named platforms, sequencing and a funded programme, in exchange for board approval of the budget. What has been gained, what has been given up, and when does that bill arrive?

What is gained, quantified

Funding certainty, and it is worth more than architects usually admit. A three-year commitment buys a team that is not re-justified every quarter, the ability to hire against it, and supplier terms that are materially better — committed-spend discounts on cloud and licences commonly run to tens of percent against on-demand rates. It also buys attention: a board that has approved a named programme asks about it, and organisational attention is the scarce input for anything that takes longer than a quarter.

What is paid

Optionality, and the price is highest exactly where technology is moving fastest. Three specific losses:

• The named platforms become constraints. A choice that was right in year one has to be defended in year two, because changing it means reopening the approval. The commitment converts a technical decision into a political one.
• Sequencing assumes the product direction holds. Most three-year roadmaps are invalidated in the first year, not by technology but by a change in what the business is selling.
• The programme acquires its own momentum. A funded multi-year programme is difficult to stop even when its premise is gone, because stopping it is an admission and continuing it is a plan. This is the largest of the three costs and the least discussed.

When the cost becomes visible

In month 14 to 18, when the first premise fails. By then enough has been built that the sunk cost argument is available, and not enough is finished that the benefit can be pointed at. That is the moment the roadmap either bends or becomes theatre, and which one happens is determined by how it was written, not by how it is managed.

How to keep the option to reverse

The mechanics are unglamorous and they work:

• Commit to outcomes and decision points, not to platforms. "Transaction processing runs on a horizontally scalable store by Q4" survives a change of database; "migrate to product X" does not.
• Sequence so that stopping is a legitimate end state. Each step delivers something valuable on its own, and the test is blunt: if this is cancelled in month 7, is the organisation better off? If the answer depends on a later step, the sequence is wrong.
• Name the assumptions with review dates. "This sequencing assumes we remain single-region; reviewed each March." An assumption with a date is reversible; an assumption buried in a plan is not.
• Fund in tranches against outcomes. The board keeps a real decision at each tranche, which is what makes cancellation an ordinary event rather than a failure.
• Publish the exit cost of each commitment when you make it. A three-year reserved-capacity purchase and a three-year architectural direction both have exit costs; only one of them is usually written down.

When this is the right call

When the work genuinely cannot be decomposed — a core banking replacement, a data centre exit, a regulatory deadline with a fixed date — the multi-year commitment is not a mistake, it is the shape of the problem. The distinguishing question is whether the value is in the destination or in the steps. Where the destination is the only thing that pays, sequence for risk retirement and take the commitment. Where each step pays, take the money in tranches and keep the optionality; the discount you forgo is smaller than the cost of being locked to a plan whose premise expired.