The £340K AI Bill With Nothing To Show For It

Sat with a client last week who spent £340K on AI in 2024. They owned none of it. Still not sure that's sunk in for them.

We went through the receipts together. The trigger wasn't a strategic rethink — it was a renewal notice. One of their SaaS vendors hit them with a 23% price increase, 14 days before the contract auto-renewed. Their CFO flagged it, started pulling threads, and the whole thing unravelled.

Here are the actual line items.

The receipts, laid out

Five categories. One year. All recurring.

When I asked what they'd have left if they cancelled all of it tomorrow, the room went quiet.

No models. No training data. No proprietary workflows. Nothing.

"We thought we were being smart by staying flexible"

That's what their COO told me, before we'd laid out every line item. The follow-up, once we had:

"I didn't realise we were paying twice for the same capability."

They weren't stupid. They were doing what most mid-market companies do. Buy best-of-breed tools. Trust the vendor's roadmap. Assume the monthly cost is the real cost. It's sensible until you add it up.

The CapEx alternative they're building

Owned infrastructure. Models fine-tuned on their data. Workflows they control. The numbers, roughly:

They lose some flexibility on the margins. Fewer shiny integrations. More internal maintenance. That's real.

But they own the asset. The data stays theirs. The models improve for them, not for the vendor's next customer.

The transition isn't painless

There's a capability gap in the first few months. Their team is learning things they used to outsource. Some workflows that worked out of the box now need building.

But here's the bit that stings.

£340K a year. They've been on this trajectory for roughly 18 months. Call it north of £400K all in. With that money:

• They could've hired a small AI team outright. Three, maybe four people.
• They could've built proprietary tooling that actually differentiates them.
• They could've owned something.

Instead they rented. The only thing they have to show for it is a folder of invoices and a login screen that works until the contract expires.

How to actually run the audit

If you suspect you're in the same trap — and most mid-market firms are — this is where to start. It's not complicated. It's just rarely done.

1. 1
   Pull every AI line item from the last 12 months. Finance can do this in an afternoon. Don't filter — include the £200/month copilots and the £80K platform fees together.
2. 2
   Group by capability, not by vendor. Document automation. Customer support. Sales enablement. You'll find duplicates within five minutes.
3. 3
   Mark each as core or commodity. Core = it touches your differentiating workflow or proprietary data. Commodity = it doesn't. Be honest. Most things are commodity.
4. 4
   For the core items, model the build cost. Get a real number. Not a vendor-led number. If owning costs less over three years than renting, the decision is already made.
5. 5
   For the commodity items, negotiate or consolidate. Rent freely where it doesn't matter. Concentrate vendors so you have leverage at renewal.

Takeaways

• If it's core to your business, own it. If it's not, rent it. Most companies get this backwards with AI.
• Renewal notices are the audit trigger. Don't wait for one — set the calendar reminder 60 days out and force the question.
• Group spend by capability, not by vendor. Duplicates hide inside org charts. They don't hide inside capability buckets.
• A folder of invoices is not an asset. If you cancelled every AI subscription tomorrow, what would you still have?

Ask me again in 6 months whether they regret the switch.