We validated the whole thesis in three weeks. Here's the experiment.

James: You came in with a roadmap that ran to next Christmas. We threw most of it out in the first hour. What was the assumption everything else rested on?

Marcus: That finance teams would hand over their month-end reconciliation to software they didn't build. Everything in the deck assumed yes. We had zero evidence for it.

James: So instead of building the reconciliation engine, what did we do?

Marcus: We did it by hand. Ten companies, three weeks, a shared inbox and a spreadsheet. They sent us the mess, we sent back the reconciled ledger overnight. No product. Just the promise of the product, delivered manually.

James: The classic concierge test. What did it tell you that a survey wouldn't?

Marcus: Whether they'd actually send the data. Saying "yes I'd use this" is free. Forwarding your real, embarrassing, half-broken finance export to a stranger is not. Seven of the ten did it in the first week. That was the signal.

James: And you set the kill criterion up front.

Marcus: You made me. If fewer than half sent real data unprompted by day ten, we'd stop. I hated it at the time. But it meant when six of them did, I trusted it — I wasn't just pattern-matching to the answer I wanted.

James: What did three weeks cost you, versus the six-month build?

Marcus: About four thousand pounds and my ego. The build would have cost the better part of the round. We found out it was real before we spent the money — and we found out which half of the roadmap was fantasy.