Wasted software spend is still a real issue despite optimisation efforts. The thing is, it's not really the fault of your SAM or procurement teams. It's complicated. This blog explores the challenge and what organisations can do about it

The number that made everyone look up

The facts are simple, and that's what makes them uncomfortable. A large organisation gave its employees access to Claude, set no usage limits, and a month later opened an invoice for roughly half a billion dollars.

It wasn't a hack and it wasn't a billing dispute. It was ordinary usage, uncapped, accumulating quietly until somebody finally looked.

And here's the detail that rarely makes the headline: the controls to prevent it already existed. Spend caps, access by role, live dashboards, all of it. They were simply never switched on before access was opened to everyone.

It's not a one-off, either. Microsoft pulled back most of its internal AI coding licences; Uber spent its annual AI budget months early. The tool keeps arriving faster than the discipline to manage it.

This was never really an AI problem

It's tempting to file this under "AI gone wrong" and move on. That misses the point entirely.

The same gap that lets a business pay for licences it never deploys is the gap that produced a nine-figure usage bill. Only the number is new, the mechanism scaled overnight, but the thinking behind it didn't.

What sits behind a decision to open the floodgates is almost always the same pressure: speed, and the bottom line. Move quickly, prove value, worry about the controls later. The irony writes itself, in trying to cut costs, an organisation created one of the most expensive technology failures imaginable.

Why an old problem just got more dangerous

For thirty years, software spend has been broadly predictable: you agree a price, you pay it, and the bill doesn't change because someone used the product more heavily than expected. Usage-based pricing breaks that comfort. Cost now rises directly with activity, and the more capable the tool, the faster the meter runs. That's why the old habit of reviewing technology cost once a year, at renewal, is no longer safe.

The instinct, then, is to add more controls, but the opposite failure is just as common and far less visible. Too little control, and you get the half-a-billion-dollar bill. Too much, and you freeze while your competitors move.

The organisations that come through this well aren't the ones with the most rules; they've built something structural that lets them move quickly and stay protected at the same time.

What good actually looks like

Most people respond to a story like this by buying a tool. There is a place for that. A simple scan showing how many licences a business pays for and never deploys puts a hard number on the waste, and numbers get people to listen.

But a tool only shows you the problem. It does not fix it. The fix is structural, and it is mostly about people.

It looks like one team that can see what is being spent across the whole business, not a picture stitched together from a dozen departments who each see only their own corner. It looks like decisions made across the organisation about how things get adopted, governed and signed off, rather than everything dumped on IT as "an IT problem." And it looks like treating technology spend as something someone watches every day, not a renewal event you wrestle with once a year.

That's the quiet version of this story. It's rarely one reckless decision, it's a hundred small, reasonable ones, each a sensible yes. The bill doesn't arrive with a bang. It arrives one yes at a time.

AI costs drift before anyone notices 

This is the quiet version of this story. AI overspend rarely starts with one big decision, more often, it starts with a series of reasonable ones.

One team buys a tool to move faster, then another chooses something different for their own workflow. A third signs up after a strong demo or recommendation. Each choice makes sense on its own, but across the business the cost and complexity start to build.

Before long, teams are paying for overlapping tools, training people on different systems and working with fragmented data. The bill does not always arrive with a bang. More often, it arrives quietly, one sensible yes at a time.

The bill got everyone's attention. The discipline was always there.

AI didn't invent a new discipline. It raised the stakes on an old one. Finding the gap between what a business has bought and what it actually uses is the work software asset management has done, quietly, for decades.

The $500 million bill simply made the rest of the business finally pay attention. 

If you want to understand what your own AI and software spend really looks like across the business, before the number gets anyone's attention, that is a conversation worth having. Check out 

Author

Seb Burrell is Enterprise AI expert at Livingstone, where he helps organisations adopt AI without losing control of the cost.