AI inside the product. AI inside the workflow. Either way: useful, measurable, honest.
Embedding AI into the product and the workflow — where it earns its keep.
Why teams hire me for this
There are two ways AI lands inside a business badly. The first is AI theatre: a chat widget bolted to the homepage, no one uses it, the bill arrives. The second is AI everything-bagel: a vague mandate to “be AI-first” that diffuses across forty surfaces and lands nowhere.
The boring third way is the only one that pays back: pick the one workflow where the maths is obvious, ship something narrow that does it well, measure honestly, expand only when it’s earned.
That’s what I do.
How it usually goes
Week 1 — discovery. I sit with the team that does the work. I read the tickets, the SOPs, the support transcripts. I write up the three or four AI opportunities that are clearly worth doing, the two that probably aren’t, and one that nobody had thought of.
Weeks 2–6 — build. Pick the top opportunity. Ship the thinnest end-to-end version. Get it in front of real users. Wire evals so we know if it regresses. Lock in a cost ceiling.
Handover. Documentation, runbook, observability dashboards, and — critically — a team that knows how to keep it alive without me.
Recently shipped
- A bespoke SMS distribution + reply-classification system that cut a 15-hour-a-week ops loop to 3 hours (ClickMeCars case study).
- Internal copilots for fintech ops teams — drafting customer responses with their own brand voice, grounded in their own knowledge base.
- Eval suites for production LLM features so the team knows immediately when a model update changes behaviour.