AI Discovery — Selling Into Fear

Discovery is the most underrated skill in technical sales — the general craft of it (finding the real problem, quantifying it, mapping who cares) is covered in the Sales for Engineers discovery material and, canonically, in John Care’s Mastering Technical Sales. This post adds the AI-era layer: discovery when the thing you’re selling is probabilistic and the buyer is afraid of it.

The two jobs of AI discovery

AI discovery has to accomplish two things at once:

1. Find a problem AI should actually solve — with a measurable outcome and a real path to production. (This is classic discovery, and it’s also the first stage of the Deploy-to-Win Loop: qualify.)
2. Surface the fear — the specific risks this buyer is carrying about adopting AI, so you can retire them deliberately instead of being ambushed by them later.

The first job is familiar. The second is what’s new, and skipping it is how AI deals that “went great in discovery” collapse in the POC.

The four fears

Behind almost every AI purchase sit four anxieties. Good discovery names them out loud:

• Hallucination / trust — “What happens when it’s confidently wrong? Can we trust it in front of a customer or a regulator?” The buyer needs to know how errors are caught and bounded.
• Cost — “This is cheap in the demo; what does it cost at our volume?” Token and inference economics terrify buyers who’ve been surprised before. Unquantified cost kills more deals than people admit.
• Security & data — “Where does our data go? What about prompt injection, data exposure, compliance?” For regulated buyers this is often the real gate.
• Lock-in — “If we build on your model/platform, how trapped are we when prices change or a better model ships?”

You will not close by pretending these don’t exist. You close by raising them first — which signals you’ve done this before — and showing, concretely, how each gets retired on real data. Naming the fear is disarming; letting the buyer discover it mid-POC is fatal.

Qualify hard — the best discovery says “not yet”

The AI era is full of enthusiasm that outruns readiness. The most valuable move in discovery is often to disqualify: to recognize a use case that shouldn’t be built yet — no measurable outcome, no owner who’ll adopt it, data that isn’t ready, a bar no current model can clear — and say so. This protects the customer from an expensive, impressive failure and protects you from a doomed engagement. A reputation for honest qualification is worth more than any single deal.

Use-case selection is where AI value is won or lost: a flawlessly executed project on the wrong use case produces nothing. So discovery should weigh candidates as a portfolio — real business value against honest feasibility — and point at the one with a measurable outcome and a credible path to production, not the one that demos best.

Quantify the value, in their terms

The other half of discovery is turning a vague “AI could help here” into a number the buyer’s organization recognizes: hours saved, cost avoided, revenue unlocked, risk reduced. An AI project competes for budget against everything else; “it’s cutting-edge” loses to “it saves ₹X / $X a quarter, here’s the arithmetic.” Quantifying value is also what later lets you prove the technical win against something that matters, rather than against a vibe.

Key takeaways

• AI discovery has two jobs: find a problem AI should solve (measurable outcome, path to production, an owner) and surface the fear the buyer is carrying — the second is the AI-era addition, and skipping it is why deals that aced discovery die in the POC.
• The four fears behind most AI purchases: hallucination/trust, cost at volume, security & data (exposure, prompt injection, compliance), and lock-in. Raise them first yourself — it signals experience and disarms them — then show how each is retired on real data.
• The best discovery disqualifies: recognizing a use case that shouldn’t be built yet (no outcome, no owner, data not ready, bar unreachable) and saying so protects the customer and your reputation — honest qualification outvalues any single deal.
• Use-case selection decides AI value: flawless execution on the wrong use case produces nothing; weigh candidates as a portfolio (value vs honest feasibility), not by which demos best.
• Quantify value in the buyer’s terms (hours, cost, revenue, risk) — an AI project competes for budget, and a number beats “it’s cutting-edge”; it also gives you a real bar to prove the technical win against later.

Further reading

• Value proposition — articulating and quantifying the value a solution delivers
• Sales process engineering — treating discovery and qualification as a designed process