Two weeks of back-to-back AI adoption BD calls forced three gut instincts of mine into written rules. Writing them down.
A deal has two values
There was one call I scored low at first: every comparable case had been the networking kind, you talk, you part ways, no stated need and no quote. But it was my first time meeting someone at the founder level of a top-tier tech company in Taiwan, so the channel value was high.
That day I overrode my own “30 minutes of prep, maximum” rule and built a full research report from scratch, but I also wrote one line into the opportunity file: this spend counts as channel investment, and may not be charged back to this deal’s pre-sales cost or effective hourly rate. Without that line, one networking-type case blows up the hourly math and I end up concluding “don’t touch these,” which would be wrong.
Two days later I downgraded his channel value by a notch. Semi-retired, winding down investments, told me himself that if nothing comes of it we can just be friends. No skin in the game. He’s a one-time bridge, not a channel; the real node was the other person in that room, the one still running a company.
One trap I stepped into the same week: I estimated my own hours by counting the documents I’d produced, treating a few AI-generated reports as eight hours of work when the whole lot took under an hour. The effective hourly rate came out nearly half of what it was, and I pulled the wrong lever because of it. My conclusion read “stop investing in pre-sales” when the right answer was “raise the price.”
Take it, but take only a third
For another one I ran the client-sorting chapters of The Pumpkin Plan over the deal. What the client sent wasn’t a wish list of AI ideas. It was a baseline requirements spec for an enterprise system: twenty-odd chapters, twenty domains, ten consulting deliverables listed out.
The verdict was take it, but take only a third. His needs split three ways: how to build the system, which belongs to an engineering team, not me; how to set the policy, which belongs to lawyers and accountants, and his own document already said so; and how people work with AI, which is the one I take.
He passed three of the six desert-island criteria, on par with another deal I’m happily running. So the real gate wasn’t whether he’s worth serving, it was the service boundary: all ten of his listed deliverables sit outside it.
The thing that goes with it: shrinking scope means raising price density at the same time, not scaling proportionally down from the previous vendor’s monthly-fee anchor. I’ve written up one similar case of letting go before (Clients Ask for Training When They Want a System); that one was a single post-mortem, this time it becomes a rule.
Where the ceiling on free sits
A decision-maker volunteered advice: when you approach a company, just give the top of the house a few sessions for free. “The only way it moves is if you teach the top layer and they find it worth something.” She is the buyer persona herself.
I took the diagnosis and left the price. First, she came in through a referral and paid from the start, never walked the free-course path, so she was describing an imagined route rather than a lived one. Second, the economics of a free course run backwards against channel quality: the low-grade channels that genuinely need the net cast wide can’t be saved by a few free sessions, and the high-grade referral channels that can afford you never needed them.
So the ceiling is written down hard: free buys the decision-maker’s first contact. Once, sixty minutes at most, no deliverable that outlives the meeting.
The block I used on another call that week was “give the criteria, not the answer.” When pressed on how exactly it should be done, what I handed over was a published third-party framework: a research firm’s four levels of AI autonomy from May 2026 (read-only observation, advice only, approval on every action, autonomy inside guardrails). Attributable, genuinely useful to him, and somebody else’s framework rather than my own method.
One more rule: credit can only flow from diagnosis to delivery, never from delivery to delivery. Crediting a course fee against some larger future project repackages a thing that has its own endpoint as “phase one” and prepays a discount on downstream work I’ve already decided not to take. Besides, he had never once asked about price, so that clause was answering an objection that didn’t exist.
When I do have to concede, the order is scope, then added deliverables, then payment timing, and only then price.
That’s the three. We’ll see if the next call overturns any of them.