The boring AI win is paperwork

On June 8, NHS England announced it would give Microsoft's Copilot to 505,000 staff in a deal worth about £120 million. Read the headline and you might expect something dramatic — AI spotting tumors, reading scans, diagnosing the undiagnosable. It's not that. The job is paperwork: notes, letters, summaries, emails. The unglamorous stuff.

And the unglamorous stuff is where the payoff is. In trials, the average participant saved 43 minutes a day — about five working weeks a year — and one acute-medicine ward cut its backlog of outstanding discharge letters by 62% within a month. Multiply 43 minutes across half a million people and you get an amount of recovered time no diagnostic demo comes close to. That gap — between the AI story we tell and the AI work that pays — is worth understanding, because it'll shape where you should point AI in your own work.

The exciting use case and the valuable one are rarely the same

AI marketing runs on the spectacular: the model that cracks a hard diagnosis, writes the app, passes the bar exam. Those make great keynotes. But the spectacular cases are usually rare, high-stakes, and hard to deploy — they need careful oversight, they carry real risk if wrong, and they show up a handful of times a day.

The boring cases are the opposite. Writing a discharge letter isn't impressive, but it happens thousands of times a day, it follows a predictable shape, and a mistake gets caught by a human who's reading it anyway. High volume, low stakes, predictable structure — that's the exact profile of work where AI delivers value safely and constantly. The NHS didn't buy half a million licenses for the miracle. It bought them for the 43 minutes, every day, per person.

Why the dull work pays and the dazzling work stalls

There's a reason the impressive pilots so often stall while the boring rollouts ship. Doctors in the UK reportedly spend something like four hours on admin for every hour with patients. That's an enormous, boring, expensive pile of work — and boring, repetitive, high-volume work is exactly what AI is good at and safe at. The diagnostic moonshot has to be near-perfect to deploy. The paperwork assistant just has to be a bit faster than a tired human, with that human still in the loop.

So the value isn't hiding in the hardest, most impressive problem. It's sitting in the dullest, most repeated one — the task so tedious nobody thought to brag about automating it. That's not a consolation prize. At scale, it's the whole prize.

How to find your version

You probably have an NHS-discharge-letter equivalent — some dull, frequent, structured task quietly eating your hours. Go find it:

• Follow the volume, not the glamour. What does your team do hundreds of times a week that follows a pattern? That's the candidate, however unexciting.
• Favor low-stakes, human-checked work. The best first targets are tasks where a person still reviews the output, so an occasional miss is caught, not catastrophic.
• Measure the minutes. "Saves 40 minutes a day per person" is a boring sentence and a huge result. Count the time, not the wow.

Resist the urge to point your AI at the most impressive problem you have. Point it at the most repeated one.

The bottom line

The NHS just made one of the biggest AI bets in healthcare, and it's a bet on letters and notes, not on diagnosis. That's not a lack of ambition. It's an accurate read on where AI actually pays right now.

The durable AI win is usually the boring, high-volume, low-stakes work — not the demo that makes the keynote. Chase the tedious task everyone overlooks, count the minutes it gives back, and you'll capture more real value than any moonshot. The unglamorous use case is the one that ships.