Name the workflow, not the technology
Scope an AI project by naming the job it removes, costing the manual version, and measuring the governed replacement against that number.
The right way to scope an AI automation project is to name the workflow it will remove: a specific job, done by specific people, at a measurable monthly cost. Missed-call recovery. Invoice chasing. Proposal drafting from discovery notes. A scope written this way can be costed, built and judged. A scope written around the technology, an agent, a copilot, a transformation, can only be believed in.
The distinction sounds small and decides almost everything that follows: what the build costs, how long it takes, and whether anyone can say afterwards that it worked.
What does a well-named workflow look like?
It has an owner, a frequency and a cost. A sentence like "quotes take five days because staff chase scope details by hand" names a workflow. So does "discovery-call notes sit for a week before anyone drafts the proposal". One of our builds took that second problem, at an Oman-based IT services firm, and cut the first proposal draft to 24 hours, down from 5 working days, precisely because the scope was that narrow and that clear.
Why do technology-first AI projects stall?
A project scoped as "deploy AI agents" has no baseline, so it has no finish line, so it can never be completed, only paused. Nobody can say what the manual version used to cost, which means nobody can say whether the replacement paid back, which means renewal decisions run on sentiment. When budgets tighten, projects that run on sentiment are the first to go, and often they deserve to be.
The framing also inverts the burden of proof. A named workflow forces the supplier to show a cost removed. A technology programme lets the supplier point at activity.
How does naming the workflow change the buying conversation?
It turns the purchase into a comparison a finance director already knows how to run: the manual version costs this much per month, the governed replacement costs this much to build, and the saving is measured against a baseline both sides agreed before work started. No belief is required at any step. A supplier who resists this framing is telling you something useful about what their proposal would survive.
Name the job. Cost the manual version. Judge the build against it. Everything else in AI buying is downstream of those three sentences.
What AI governance actually means for a board
The real test of AI governance is whether you can reconstruct what a system did and why, months later, for someone with the power to penalise you.
Why we measure every AI build against one number
Most AI programmes stall on the choice of first project. Scoring every candidate on monthly payback settles the choice with arithmetic instead of opinion.