2 September 2026 · My AI Stack editorial
Cut a large AI programme into jobs you can actually buy
Implementing AI and proving ROI in a large organisation is hard. Slice the problem into customer jobs. Make or buy stays strategic — a specialised tool on the shelf is often the faster, lower-risk way to learn.
The programme is the hard part
In a large organisation, the difficulty is not “which model”. It is implementation: data that lives in five systems, a works council, a security review, a customer who will meet the result in public, and a board that wants ROI before anyone has named the job.
Company-wide AI programmes hide that. They buy a platform, staff a centre of excellence, and hope use-cases appear. Use-cases that appear from a slide tend to be departmental toys. ROI stays a story because nobody can say which customer moment moved.
Smaller pieces find the applications that fit
Cut the problem the way a customer already walks it. Notice. Enquire. Buy. Get onboarded. Hit a problem. Come back. Those are touchpoints with an owner, a leak, and a job-to-be-done.
Two leaky touchpoints are enough to start. That slice is how you find the applications that belong to *this* organisation — not a vendor landscape, and not a transformation roadmap that cannot be staffed. A drafting copilot and a customer-facing agent are different products. They should not share a risk tier or a procurement path.
The CxO hub and the map exist so that conversation has a shared language before the next steering committee.
Make or buy remains the strategic decision
Make when the experience *is* the product, or when no honest specialist exists for the job. Buy when the job is standard enough that a specialised tool already finishes it. Bolt with automation when the gap is routing, not intelligence.
That decision does not go away because a suite offered “agents” in the same contract. It gets sharper when you decide per job instead of per fashion.
The shelf is a fast track for learning
On-the-shelf specialists are often the faster, lower-risk path for learning and experimentation. You discover whether the job is real, who must own the voice and the handoff, and what “good” looks like — without funding a platform that has not earned the next increment.
That is what this catalog is for: a short, opinionated shelf, not a directory. HEA-World is a labelled founding partner where a brand-owned conversational front door is the job. It is not a contact-centre replacement and not the only card on the map. Compare the specialist that actually finishes the square you circled.
When the experiment has to face a customer, read what CxOs should ask before go-live. Then buy one job. Measure it. Only then decide whether to make the next piece.