This year my business went backwards on a number that decides our standing with our largest vendor. We earned that by doing exactly what our clients asked us to do. We rationalised their cloud estates, switched off what they were not using, and tightened their security. Their consumption fell. So did the number we are judged on.
Our head of operations review said it in one line: you rationalise them, you bring them down, you control their spend — it is good for the customer, it is not good for your face.
My own version, in the same meeting, was blunter. If our clients do their security properly, we are penalised for it.
It is budget season, and almost every AI business case I have read in the past month is a reduction case. Fewer hours in the write-up. Fewer licences allocated out of habit. Less spend on an ageing platform that AI makes it possible to leave earlier than planned. Those are the right cases to make, and I encourage clients to make them.
They are also the first time I have seen the client's business case and the provider's revenue model point in opposite directions on the same invoice, in the same quarter. Not as a philosophical tension. As arithmetic.
The channel's funding architecture has not moved. Partner standing, rebates, co-investment and the money that pays for the workshops your provider runs for you free of charge are all indexed, directly or indirectly, to growth in what you consume. Trim a client's spend and you can hold every other measure at full marks and still fall short, because the one that has not been met is the one that tracks the direction of their bill.
Meanwhile the entire managed intelligence pitch — and it is a real shift, I am making it myself — is sold on outcomes and efficiency. Every provider in the market is now promising to reduce the very thing their own programmes reward them for growing. Almost nobody says this out loud, because saying it means admitting that the incentive under your advice runs the other way.
They assume alignment because the relationship is good. It usually is good. But the relationship is not the mechanism. Buyers interrogate SLAs, certifications and roadmaps, and almost nobody asks the one question that actually predicts behaviour: what happens to your provider's revenue if this works?
You can see the consequence without ever seeing the incentive. A finance executive at a private hospital told me last week that his blocker was not the technology and not the budget. He had been asking a straightforward question about whether the AI tooling already on his staff's screens met his sector's data rules, and, in his words, no one seemed able to give him a straight answer, so he had put the brakes on it. Several people in his supply chain had a commercial reason for that answer to stay vague.
An IT leader at a community health service with around eight hundred staff had reached the same place from the other end. He was bringing in an outside adviser to write an AI plan, explicitly so that the recommendation did not come from anyone with a stake in the outcome, including himself. That is what a buyer does when they cannot tell whose interests the advice is serving.
If this programme cuts your cloud and licence spend by twenty per cent, what happens to your provider's revenue? Ask for the answer in one sentence, in writing. A provider who cannot answer it has not thought about it, and a provider who will not answer it has.
What share of what you pay them is resale margin, and what share is work? If most of it rides on the licences and the consumption, then efficiency is a cost to them however warmly it is described in the proposal.
What have they recommended in the last twelve months that reduced your bill? Not saved you time. Reduced the invoice. If the answer is nothing, that is not an accusation, but it is data, and it is data you already have.
I am on the wrong side of this, which is the only reason I am entitled to write it down. We took the hit because the alternative was to leave clients paying for capacity and licences they did not need, and I would take it again. But taking it repeatedly while complaining about the metric is not a strategy, it is a grievance.
The alternative is to change what we sell, so that the number moving when a client succeeds is a number of ours. Paid for processes that are still running in ninety days. Paid for agents that are registered, owned and governed. Paid for the judgement, not for the consumption underneath it. That revenue is smaller in year one and far harder to forecast, and it is the only version of managed intelligence that survives the question in the section above.
Do not ask your provider whether they are AI-enabled. Everyone says yes. Ask what happens to their revenue when your bill goes down. It takes one sentence to answer, and it tells you more than any capability statement you will be sent this year.