How Do You Build an AI Roadmap That Survives Contact?
Most AI roadmaps are a list of ambitions with dates attached. A useful one is a sequence: what you do first, what that unlocks, and what you will know afterwards that you do not know now. It should be short, honest about dependencies, and built to change — because the first project always teaches you something that revises the second.
A Wish List vs. a Roadmap
| Step | Wish List | Sequenced Roadmap |
|---|---|---|
| Ordering | By enthusiasm, or by whoever asked loudest | By dependency and by what each project teaches |
| Dependencies | Discovered mid-project | Named up front, including the ones outside AI |
| Time horizon | Three years, in equal detail throughout | Detailed for two quarters, directional after |
| Success measures | Adoption, or number of projects delivered | The business number each project is meant to move |
| Handling surprises | Roadmap quietly abandoned | Decision points where the plan is expected to change |
Sequence for Learning, Not Just Value
The obvious approach is to rank projects by expected return and start at the top. It is usually wrong, because the highest-value project is often the one you understand least.
A better first project is one that is genuinely useful, small enough to finish, and that exercises the capability everything else depends on. If four of your six ideas need clean customer data, the first project should be one that forces you to sort that out while delivering something real.
That way the second project starts from a better position than the first did, which is the entire point of sequencing.
Plan Two Quarters, Sketch the Rest
Detail beyond about six months is fiction in this field. The tooling moves, and more importantly your own understanding moves — teams routinely find their second project is not what they planned, because the first one revealed something.
So write the next two quarters properly, with scope, owners and measures. Keep everything after that as direction and dependencies. A roadmap that admits its own uncertainty is one people will keep using; one that pretends to know 2028 gets quietly ignored by March.
Where This Fits
This is one part of our work in AI for Strategic Management. See the full set of AI use cases for the equivalent in other industries and functions.
Frequently Asked Questions
How long should an AI roadmap cover?
Two quarters in real detail, twelve to eighteen months directionally. Anything more precise than that is guessing, and precise guesses are worse than acknowledged uncertainty because people plan against them. Revisit properly each quarter rather than treating the document as fixed.
Should we start with a quick win or the big opportunity?
Neither framing helps much. Start with something genuinely useful that exercises the capability your later projects depend on. A trivial quick win teaches you nothing and creates the impression AI is a toy; the big opportunity attempted first usually stalls because the foundations are not there.
How do we handle the pace of change in AI?
By committing to problems rather than to tools. The business problems on your roadmap will still be problems in two years; the specific model or vendor may not be. Roadmaps written around a named technology age badly and force you to defend a choice instead of changing it.
Who should own the roadmap?
Someone accountable for a business outcome, not for technology delivery. AI roadmaps owned purely by IT tend to optimise for shipped systems rather than changed numbers, and they struggle to get the process changes that make the systems worth having. The technical lead is essential, but ownership sits with the business.
What is the most common reason roadmaps fail?
Nobody planned for the process change. The model works, and then nothing happens, because the workflow around it never adapted and the people using it were never brought along. Budget for that explicitly — it is usually more effort than building the thing, and it is the difference between a pilot and production.
Sequence the next two quarters properly and sketch the rest honestly.
Get a Free Proof of Concept within weeks.
Let’s talk
AI is here to stay
Let’s win together
Your first 45-min alignment session — and a small PoC — are free.
Or just say hello or write us an email.