Writing an AI roadmap that survives its first quarter

Most organisations now have some form of AI strategy. Far fewer have one that is still guiding decisions six months after it was approved. The document gets presented, applauded and filed, while the actual work drifts towards whatever is most urgent or most exciting that week.

Having helped many leadership teams through this process, we see the same reasons for failure again and again, and a handful of principles that reliably prevent them.

1. Anchor every initiative to a metric someone already cares about

An initiative described as “deploy a generative AI assistant” will struggle for attention. One described as “cut average claims handling time from three days to one” will not. Before anything goes on the roadmap, it should be tied to a business metric with an existing owner: someone whose performance is already judged by that number.

2. Score feasibility as rigorously as value

Ambitious use cases are easy to list. The hard part is being honest about which ones are achievable with the data, systems and skills available today. We score every opportunity on four dimensions:

• Value: the size of the prize, in the metric’s own units
• Data readiness: whether the necessary data exists, is accessible and is of usable quality
• Integration effort: how many systems and teams must change
• Risk: regulatory, reputational and operational exposure

The best first projects are rarely the most valuable ones. They are the ones with strong value and high readiness, which build credibility and capability for what follows.

3. Sequence for learning, not just return

A good roadmap is a portfolio, not a ranked list. Early initiatives should be chosen partly for what they teach the organisation: how to get a model through security review, how to measure quality, how to manage change with frontline teams. Those lessons make every later project faster and cheaper.

4. Budget for operating, not just building

AI systems are not finished when they launch. Models drift, prompts need tuning, usage grows and costs move with it. A roadmap that only funds the build phase sets up a predictable crisis twelve months later. As a rule of thumb, plan for ongoing operation and improvement from the outset, and make someone accountable for it.

5. Revisit it every quarter

The AI landscape changes quickly: capabilities improve, prices fall, regulations evolve. A roadmap should be a living document with a quarterly review that asks three questions:

1. What did we learn from the initiatives in flight?
2. What has changed in the technology or market?
3. Should anything move up, down or off the list?

The test of a good roadmap

A useful test is simple: six months from now, will the people doing the work still refer to it when deciding what to do next? If the answer is yes, it is doing its job.

Our AI Strategy & Consulting engagements are designed to produce exactly this kind of roadmap.