Strategy · Intermediate
Build an AI strategy that actually gets funded
A five-step process for going from a vague board mandate to a ranked, costed list of use cases you can start on Monday.
Most AI strategies die in one of two ways: they are too abstract to start, or too broad to finish. This playbook is deliberately narrow. It produces a ranked list of five or fewer use cases with a costed case behind each.
It takes two to four weeks with a small team. The output is a document your engineering leadership can build from and your CFO can approve.
The playbook
01 Write down the metric before anything else
Pick the single number this programme is meant to move: gross margin, cycle time, cost to serve, or risk exposure. One number, agreed with the executive sponsor in writing.
If nobody will commit to a number, that is your finding. Stop here and resolve it. Every downstream decision depends on it.
02 Map the process, not the technology
Take the three functions closest to the metric. For each, walk the actual process with the people doing it, step by step, and record where time and rework accumulate.
Do not mention AI during these interviews. You are looking for bottlenecks, and naming the solution first biases what people tell you.
03 Generate candidates and kill most of them
For each bottleneck, write one sentence describing what an AI system would have to do to remove it.
Score each candidate on three axes: value if it works, confidence it will work, and effort to get to production. Anything low on all three is dropped now rather than in month four.
04 Build the ROI model, including the failure case
For the top five, model the annual value against the fully-loaded cost: build, run, and the internal time the rollout consumes.
Model what happens if adoption is only 40 percent, because it often is. A use case that only works at full adoption is fragile and should be ranked accordingly.
05 Sequence, and name an owner for each
Order the survivors so the first one ships within a quarter. Early proof buys patience for the slower items behind it.
Every item gets a named operational owner who has agreed to run it in production. No owner, no place on the roadmap.
Start before the gap gets expensive.
Tell us where the work is stuck. We will tell you, in plain language, what AI can and cannot fix, and what it would take to do it properly.