Somebody else has done it

"Somebody else has done it." The smartest AI strategy in the mid-market.

Last week, an IT lead at a disability services provider gave me the most honest AI strategy I've heard. When his team pitches an AI idea, the business wants to know one thing: has somebody else done it? If they have, in his words, "we copy paste."

I first heard that as a lack of ambition. I was wrong. For a mid-sized organisation, it's the most rational AI strategy there is. The problem is that almost nobody is set up to support it.

Mid-market buyers don't want innovation from their providers. They want proof. And the provider sitting on the most relevant proof, their managed service provider (MSP), is built to forget it.

Why copying makes sense

His maths explains why: "if I spend 30, 40,000 just to try something out and if it fails". At his size, a failed pilot isn't a lesson. It's the year's discretionary budget. So he does the sensible thing: he goes to peer roundtables and asks other IT leads what worked.

He wasn't the only one asking for proof last week:

• A National Disability Insurance Scheme (NDIS) not-for-profit, comparing two quotes for its first AI agent, asked: "is there a way that we can get some case studies or some references?"
• A forestry IT lead said he'd been "talking to lots of people over the last three years" and had yet to find a consultant who understood where he was coming from.
• An advisory firm acting for a prospective customer accepted our scope and fee, then asked whether we had "existing clients who would be happy to provide a reference".
• Ask your provider for three patterns it has seen work in organisations like yours, with the outcome, the cost and what went wrong. Anonymised is fine. Vague isn't.
• Offer to be a reference in exchange for learning from others. Proof works both ways.
• Copy the pattern, not the prompt. The process, controls and measures travel. The configuration rarely does.
• Treat "you'll be the first" as a warning sign, not a selling point.

Not one of them asked for ideas. They asked for evidence. Commentators call this being a "fast follower": watch, wait, then move. That's only half the answer. Watching only works if someone shows you what to watch.

The proof is closer than you think

Most organisations look for proof at conferences and in vendor case studies about businesses ten times their size. The most relevant evidence is far closer. It sits with their MSP.

An MSP sees the same problems across many similar customers every week. But it's built to forget. Every customer is a separate contract, a separate ticket queue and a separate confidentiality boundary. What one learns rarely becomes a pattern for the next.

We're no exception. When I looked at our own managed services base, my honest view was that only two or three customers would engage with us on AI right now. So we started with ourselves as customer zero: our own health checks went from eight hours to ten minutes. That's one proof point, not a pattern. A pattern needs a second organisation, then a third.

What to do about it

Copying isn't a weakness. Copying blind is.

• Ask your provider for three patterns it has seen work in organisations like yours, with the outcome, the cost and what went wrong. Anonymised is fine. Vague isn't.
• Offer to be a reference in exchange for learning from others. Proof works both ways.
• Copy the pattern, not the prompt. The process, controls and measures travel. The configuration rarely does.
• Treat "you'll be the first" as a warning sign, not a selling point.

The next generation of managed services won't be won by whoever builds the cleverest agent. It will be won by whoever can say, with evidence: here's what happened when three organisations like yours tried this. That's what the "managed" in Managed Intelligence should mean.