Most AI sales calls go badly because the buyer doesn't know what they want yet. The vendor reaches for whatever they last shipped, the buyer agrees because the demo is impressive, and four months later there's an unhappy meeting.
Here's a 30-minute exercise you can run yourself, before talking to anyone.
The exercise
Grab three columns on a whiteboard.
Column 1 — Hours per week. List every repetitive task a senior person in your org does. Be honest. Include report-writing, proposal drafts, status updates, support replies, scheduling.
Column 2 — Revenue impact. For each task, score 1–5: does doing this better/faster make us more money? A faster proposal might mean more proposals out, more revenue. A faster report might just mean an earlier coffee.
Column 3 — Risk level. For each task, score 1–5: if AI got this wrong, how bad is it? Drafting a proposal that a human reviews: low risk (3). Sending an invoice directly to a client: high risk (5).
Multiply Column 1 × Column 2, then divide by Column 3. The highest number wins.
The point
The winning use-case is almost never the sexy one. It's usually "draft this thing for a human to review" — which is exactly where current LLMs are strongest. Save the moonshots for project three.