The one cell in your AI business case that isn't telling the truth

In 1747, a naval surgeon called James Lind ran a trial aboard HMS Salisbury. Twelve sailors with scurvy, split into six pairs, each pair given a different remedy: cider, vinegar, seawater and the rest. The pair given oranges and lemons were back on duty in about a week. Everyone else stayed ill.

Lind published in 1753. The Admiralty did nothing. Thirty years later another naval physician, Gilbert Blane, replicated the work and formally asked for lemon juice to be issued to every sailor in the fleet. He was turned down, for two reasons on record: the evidence wasn't persuasive enough, and lemons were too expensive.

The order finally came in 1795. That was forty-eight years after twelve men on the Salisbury had settled the question. Nobody was ever disciplined for the delay. There was no inquiry and no lessons-learned document, because, as Andy Logan puts it, you can't hold an inquiry into something that didn't happen.

That story opens the latest episode of AI Levels, and it sets up the question Andy spends the next nineteen minutes on: what does standing still actually cost?

Doing nothing rarely looks like doing nothing

The episode's most useful move is refusing the easy explanation. Institutions aren't standing still through inertia. They're responding rationally to a governance system that only counts one of two costs. Deciding to act produces a business case, a sponsor, a budget line and a go-live date, all auditable and all attributable. Deciding not to act produces nothing at all, and nobody has ever been performance-managed for a transformation programme that stayed in draft.

So inaction arrives in respectable costume: the permanent pilot that began as a twelve-week proof of concept and is now in its third year; the working group that has met fourteen times and changed nobody's actual job; the strategy that is genuinely good, properly argued, and changed nothing on Monday. All activity. None of it is action.

Andy gives the mechanism a name. Technical debt is what you owe for building something badly. Decision debt is what you owe for not building anything at all. Same behaviour, no invoice, paid monthly in friction, staff time and turnover.

He then does Gilbert Blane's job and prices it, across four currencies. Money: the OfS reports 35.8% of English institutions in deficit for 2024/25, forecast to reach 42.7%. Students: HEPI finds 95% of undergraduates using AI, 94% of them on assessed work. Fairness: where an institution provides nothing, access is settled by who can afford £20 a month. Staff: the Digital Education Council finds 77% of faculty using AI in their teaching, but only 31% who feel meaningfully involved in shaping policy.

Which produces the reframe worth taking to your next committee. You have already deployed generative AI at scale. It went in with no procurement, no DPIA, no accessibility review and no permissions model. The question was never whether to adopt. It was whether the adoption gets governed.

Where we come in, and where we don't

Andy is blunt that AI is not the answer to a sector-wide deficit, and that anyone claiming otherwise is selling. We agree, and we'd rather say so plainly.

He's equally blunt about the wrong response to inaction: panic-buying a thin wrapper around a general-purpose model, with no grounding in your own content and no awareness of who is allowed to see what. That will cost you more than standing still.

That is the problem the SMART: platform was built for. Answers grounded in sanctioned institutional content, permissions respected by default, and a human in the loop where the decision matters. It's also why Staff Copilot sits inside the work people already do rather than beside it.

The episode's actual homework needs none of that, mind you. Find the "option zero, do nothing" row in whatever business case is currently in your inbox. Look at the cost column. If it says zero, put a named human being against it, ideally someone in finance, who will be less charitable, and ask them to fill it in properly.

▶ Watch Episode 4 of AI Levels: https://www.youtube.com/watch?v=qpESEa-t1mU

---