You can build a credible AI roadmap in one week by running it as a decision sprint: two days gathering real pain points, one day scoring them for value and feasibility, one day pressure-testing the top candidates, and one day committing owners and budget to the top three. The output isn't a vision deck, it's three funded decisions.
The five days
- 1Day 1 - Pain inventory: interview six to eight team leads. Ask where hours go, not 'where could AI help.'
- 2Day 2 - Data reality check: for each candidate, verify the data exists, is accessible, and someone can grant access.
- 3Day 3 - Scoring: rate every candidate on business value and feasibility. Kill anything low on both. Be brutal.
- 4Day 4 - Pressure test: for the top five, sketch the workflow, the failure modes and the eval. Drop any that can't survive the sketch.
- 5Day 5 - Commit: pick three, assign owners, set budgets, book the 90-day review.
The three-slot portfolio
- Quick win (weeks): visible, low-risk, builds credibility, often an internal drafting or triage workflow.
- Core workflow (a quarter): a frequent, measurable process where AI changes the unit economics.
- Bet (two-plus quarters): something that could change your position if it works, sized so failure is survivable.
The traps that ruin roadmap weeks
- Letting the loudest executive's pet idea skip the scoring.
- Scoring feasibility without checking the data (day 2 exists for a reason).
- Picking eight initiatives, three funded beats eight endorsed.
- Ending without owners: an unowned roadmap item is a wish.