The best AI positioning barely mentions AI. It starts by finding a workflow the customer already cares about and a number they already measure, then asks whether that number could move. Your discovery job is to qualify hard: who owns this outcome, what is it worth, what data and systems does it touch, and what would 'good' look like in a quarter? Customers — especially in manufacturing — respect specificity. A vague 'AI transformation' pitch invites a vague 'send us a proposal'; a sharp 'we think we can take two points off your unplanned downtime, here's how we'd prove it in six weeks' invites a meeting with the person who owns downtime. Beware the trap of agreeing to a free-floating proof of concept with no defined success metric and no path to production. That is how presales effort evaporates into pilot purgatory. Qualify for a use case that has an owner, a number, reachable data and an appetite to ship — and walk away from the ones that don't.

Avoid the demo trap

A POC with no success metric and no route to production is a time sink for everyone. Before you build anything, agree what number it must move and what happens if it does. No metric, no POC.

1. 1.A strong discovery conversation centres on:

   The model's benchmark scoresA workflow and number the customer already owns and measuresHow many GPUs you haveThe latest research papers

2. 2.Agreeing to a POC with no defined success metric usually leads to:

   A fast path to productionPilot purgatory and wasted effortA guaranteed saleLower model costs