'There's nothing Claude can do that can't be done in Copilot'

Five external AI advisers joined the call. I had prepared a business case for Claude – use cases, workflows, the lot – and got perhaps six minutes into my pitch before Dom, one of the five, said: ‘There’s nothing you’ve said about use cases for Claude which Copilot can’t do.’

There is a particular silence that follows a sentence you believe to be wrong. Into mine arrived three responses at once. The first was the ape: what are you talking about, that is nonsense. The second was the operator, pointing out that Dom is not the decision-maker but is unquestionably a letter in MEDDPICC, and that there is no percentage in alienating him in week one. The third arrived late and rather quietly, and was the only useful one — that Dom was not talking about capability at all. He was talking about integration, and procurement, and the fact that Copilot is already bought, already deployed and already his. He was answering a different question, and answering it maybe correctly. I turned to the decision-maker, which bought me the three seconds required not to say the first thing. My face, I fear, did the rest.

Elsewhere in that first week I put a dataset into Gemini rather than doing what I would normally do, which is build a cathedral of INDEX MATCH and lose an afternoon inside it. It was, briefly, miraculous. ‘Add this variable.’ ‘Filter that out.’ ‘Change the colours.’ Charts appeared, handsome ones, with real analysis underneath them. I felt like God, or at least like a partner with a junior analyst and nothing in the diary until four.

The reckoning came the next morning. The logic did not always survive inspection. There was a field quietly poisoning an answer that I had not known to ask about. And the formulas it wrote would not copy down a column the way mine do. So back into Excel to put it right. A day’s work in the old world had taken two hours in the new one, plus half a day repairing the new one. Parity. I have rarely enjoyed parity so much.

Then to filing infrastructure, the new mecca of the AI brains and set ups (heard of Obsidian?). To get anything sensible out of Gemini you must point it at the right documents, which means an organising system, which means confronting the Google Drive of a 700-person organisation. All 700 have held a private theory of what a folder is for, and all of them have been acting on it for years. Worse, Gemini has no Projects – nowhere to park a body of context and return to it – so every query begins from nothing and must be briefed from scratch. A brilliant new colleague with no memory, and a filing cabinet the size of a barn.

The rest of the week went on remembering how to be a management consultant. The task is writing the B2B business strategy. I made myself write the storyline before the data, hypothesis first, old discipline – and produced something I was frankly delighted by: one enormous all-in-one marketplace for trust endorsements. I had invented Airbnb all over again. Two days later, having finally looked at the numbers, I could see it was terrible, and had been terrible from the moment I thought of it, and that I had by then arranged a great deal of work around it. Tearing it up sounds like a bad week but I told myself that re-work is the work.

Which is the lesson, I think. The consulting workflows I built in Claude – chunking a problem into analysis, then structure, then presentation – looked slick. But until the system knows the business and I know the system, they are best used as a source of hypotheses and methods rather than as the spine of the work. The spine is still mine, to my relief.

More soon on whether Claude gets signed off; whether I am allowed to keep Wispr Flow and Granola or must surrender them like contraband at customs; and whether in-house consulting makes it into 2026 or I spend the year back in Excel with the issue maps of 2015. Dom, I suspect, already knows.

---

Ed Davison has recently joined Which? as an internal consultant working on AI transformation. He was previously at Multiverse.