One Workflow. One Skill. One Number.

Most AI plans for customer programs fail on scope. Someone draws the full picture: research, intake, agendas, recaps, analytics, targeting, all automated, live by the end of the year. It gets approved because it is impressive, and it stalls because nothing in it is small enough to finish.

The approach that works is the opposite, and it fits in one line. Pick one workflow, build one skill, measure the gap.

Monday

Pick the workflow. Not five. The one that consumes the most manual time on your team today. For most briefing programs that is executive research or the post-briefing recap. Time it for a week. That is your baseline.

Build the skill once. Whoever does the building, wrap that one workflow so an operator can run it without writing a prompt. Plan on about two weeks, including the back-and-forth with the people who do the work today. Their corrections are what make it good.

Measure the gap. Time it again. Grade the output against your written standard. Then decide what to build next, with evidence.

Ninety days, one result

The output of the first ninety days is not a platform. It is a number: hours saved, quality held or better, throughput rising, on one real program. A measured result on your own work beats any vendor benchmark, because nobody can argue with it and nobody else has it.

Define the kill signal first

Before you start, write down what failure looks like. Proof is a clear before-and-after difference in good briefings delivered, plus at least one business outcome moving. If the lift is inside the noise, a small gain with no movement in deal velocity or win rate, the first workflow was probably the wrong one. It is far cheaper to learn that on one program than on five.

Deciding this in advance is what makes a lukewarm result useful instead of a sunk cost you keep defending.

The teams that get this right are not the ones with the biggest plan. They build one skill at a time, in production, with operators in the loop.