Why the constraint matters more than the sector name
At the level of models and tooling, AI work looks similar everywhere: retrieval, evaluation, agents with bounded permissions, workflow enablement for staff. What differs is the constraint that decides whether the work is acceptable.
A knowledge assistant for a law firm and one for a SaaS help centre use the same components. In the law firm, the hard requirement is that one client’s documents never answer another client’s question, and that a partner reviews anything that leaves the firm. In the SaaS product, it is that one tenant never sees another’s data and that the feature does not cost more than the account pays. Same architecture, different acceptance tests, different owners, different first engagement.
That is how these pages are organised. Each sector page names the data involved, the controls the work must respect, how acceptance is judged and who owns the result, and says what is out of scope.
How to use these pages
Start from the row in the table above that matches your hardest constraint, then read that sector page. Each one describes a specific first engagement:
- Financial services: a pilot with every AI step tiered, approved and logged, with compliance sign-off left to your functions.
- Regulated operations: translating an agreed policy into workflow controls staff can use.
- Retail and marketplaces: search, recommendations, catalogue and support, measured against your baseline.
- B2B SaaS: tenant-aware AI features with a per-customer cost model.
- Professional-services firms: firm-level sequencing across client confidentiality and partner review.
- Manufacturing and robotics: AI planning connected to machines, with safety outside the model.
- Property and construction: bounded administrative workflows with provenance and decision gates.
- Education operations: staff administrative workflows, never decisions about learners.
Two earlier sector pages remain available as they are: AI for defence and dual-use work and AI for healthcare.
When to skip the sector page
If you already know the technical problem, the sector page adds little. Go straight to the service: RAG quality rescue for wrong answers from an existing assistant, governed AI agent implementation for agents that act in your systems, or the contract AI engineer route if you need senior capacity under your own manager. If people have the tools but are not using them, start from AI enablement.
What no sector page offers
None of these pages offers regulatory assurance, legal advice, safety assessment or certification. In every sector, decisions with legal, financial or safety consequences for individuals stay with people, and AI work is designed so that those decisions are routed to them rather than made by a model. Sector experience is stated only where it is in the evidence register, with the scope of what I did.
If none of these fits
Describe the workflow, the data, and the constraint you work under in the brief below. I will reply with the engagement I think fits, the questions I would need answered first, or a plain statement that it is not something I should take on.