AI readiness and use case strategy for a global medical technology manufacturer

The situation

A global manufacturer of medical devices and surgical products, with business units across Europe, the US and Asia, and a board that had been asked the same question at three consecutive meetings: what is our AI strategy? Every unit had an answer. None of them matched.

Forty candidate ideas had accumulated over eighteen months. Some were real: predictive maintenance on manufacturing lines, demand forecasting across a complex supply chain, document intelligence for regulatory submissions. Some were a vendor's slide deck with the company logo added. The data team, twelve people serving a group of thousands, was being pulled in every direction at once and had started saying no to everything.

What was in the way

Three things made this harder than a normal prioritisation exercise. The first was regulatory. Anything that touched a device, a patient or a clinical claim fell under MHRA in the UK, the FDA in the US and the EU MDR, and the quality and regulatory function was, rightly, nervous about AI anywhere near a product. The second was data. Every unit believed it had the data to support its idea; almost none had checked. The third was human. The forty ideas had forty owners, and most of them had already told their teams it was happening.

The board did not need to understand large language models. They needed to see six things, ranked, with a price and a risk next to each.

What we did

Navigator starts with the business, not the technology. The first fortnight was structured discovery with every business unit: what the idea was, what it would change, what it would need, who would own it. Each conversation was drawn as it happened, and the drawing went back to the unit the same day so they could correct it. By the end of week two the forty ideas were forty one-page cards in the same format, which was the first time anyone had been able to compare them.

Weeks three and four were ARIA. Every card scored across fifteen dimensions in four groups: value, feasibility, risk and strategic fit. The scoring was done with the unit owner in the room, not to them, which meant the arguments happened early and in private. Several owners scored their own idea down once they saw the data dimension honestly. Two ideas that nobody had championed rose to the top because they were cheap, safe and reusable across units.

Weeks five and six were GRACE on the shortlisted six: data provenance, human oversight, regulatory exposure, explainability, and the controls that would need to exist before anything reached a patient or a submission. Quality and regulatory were in every session. Two of the six were re-scoped so that AI assisted a human decision rather than making one, which took them out of the highest-risk classification without losing most of the value.

Weeks seven and eight were the write-up and the board. One strategy document, one roadmap, one page per use case, and a thirty-minute presentation with the drawings rather than the slides.

What changed

The board approved the strategy in one sitting. Six use cases were funded with named owners and a twelve-month sequence. Four were parked with written reasons, so their owners knew what would need to change. Thirty were shelved, and the shelving was explained to each owner personally before the board paper went out, which is why none of them fought it.

The governance baseline became the company's standing AI policy: a scoring method the data team now runs on every new idea, a risk register the quality function owns, and a rule that nothing touches a device decision without a human in the loop and a GRACE profile on file.

What they kept

The scoring method, the templates and the register, all run internally now. A quarterly review rhythm with the board that needs no outside help. And a data team that says yes to the right things, because they have a way to say no to the rest.

All case studies