Every AI product that touches the real world depends on people underneath it. The model gets the credit. This is the layer that decides whether it actually works.
Most AI Ops functions don't begin as functions. They begin as a workaround — a spreadsheet, a freelancer, a vendor brought in because hiring was too slow. It works, until the experiment succeeds and the organisation discovers it has production-scale demand running on pilot-scale process.
Building and managing vendor operations for AI and annotation startups — written from the operating floor, not the boardroom.
Build-vs-buy decisions, vendor vetting, QA that actually scales, onboarding without micromanaging, budgeting for the costs nobody puts in the SOW, and the mistakes that quietly sink AI ops programmes.
Pulled straight from the book's appendices — for teams who want the working documents without reading cover to cover.
Weighted scoring across delivery, quality, commercials, and strategic fit.
An operating framework for scope, volume, quality, and change control.
The core formulas and starting ratios for staffing a human-in-the-loop operation.
Questions that surface real vendor operating evidence, not sales-deck answers.
Five stages, from manual-but-visible to adaptive, with a 90-day build plan.
We're glad to talk — a short vendor-ops readiness conversation, an introduction to a vetted delivery partner, or a look at the full template toolkit for teams who want the tools without the full engagement.
No cost, no obligation, first conversation.