Buyers outside tech should refuse every AI surcharge at renewal this year

On 1 July 2026 Microsoft raised the price of its commercial Microsoft 365 suites, taking E3 from $36 to $39 per user per month. Company officials put the rise down partly to AI features such as Copilot Chat, alongside security and management add-ons. The full Copilot licence still costs another $30 per user per month on top. If you run procurement at a hospital group, a regional bank or a freight business, I think you should treat both the suite increase and the $30 as unproven costs. Refuse them until the vendor can show savings in your own accounts.

Apollo's Daily Spark of 24 October 2026 hands you the evidence for that conversation. Its economists find revenue per worker and profit margins rising at the Magnificent Seven and flat across what they call the S&P 493, everyone else in the index. For productivity they use the San Francisco Fed's utilisation-adjusted series, which strips out cyclical swings in how hard capital and labour are being worked, and they still see no AI lift. Apollo also reports that mainstream corporate America is pulling back and starting to book token spending as a cost centre that drags on margins. I would print that paragraph and bring it to the next renewal meeting.

I went in expecting a timing story, the usual line that diffusion takes a while and the numbers catch up. What struck me instead was who sits on each side of Apollo's split. The companies with rising revenue per worker are largely the ones collecting AI subscription fees. The 493 are the ones paying them. Every AI uplift baked into a seat price moves money from the second group to the first, and I suspect a fair share of the margin gap Apollo charts is exactly that transfer.

Torsten Slok, Apollo's chief economist, has already warned that in healthcare, banking and pharma, process overhauls and privacy rules could delay any productivity gain by years. He framed the risk as five years instead of five months. For a buyer in those sectors, that means paying a premium today for a return that, on Apollo's own reasoning, may not arrive inside this contract term or the one after.

The MIT NANDA preliminary report from July 2025 points the same way. It reviewed more than 300 public initiatives, interviewed people at 52 organisations and surveyed 153 senior leaders, and found that 95% of organisations saw no measurable return from generative AI. Newsweek has cautioned that the methodology deserves more careful reading than the headline gets, and I accept that. Even read generously, though, you are being asked to pay per seat, across the whole company, for a category where most pilots could not produce a number.

The serious pushback is that gains are real and simply concentrated. Brynjolfsson, Li and Raymond studied 5,179 customer support agents and found an AI assistant raised issues resolved per hour by 14% on average, and by 34% for novice and low-skilled staff, with little effect on experienced workers. Census data shows 18% of firms using AI in a business function in the Nov 2025 to Jan 2026 period, but 32% on an employment-weighted basis and 50% to 60% among very large firms in information and professional services. Averages can bury pockets of real value. I agree, and I think it strengthens the case for withholding the premium. The support study measured a hard output per agent per hour, which is precisely the kind of figure a vendor can pull from your own ticketing data. If the gain exists in your business, it can be shown. The same study also says the gain went to juniors, so paying the AI rate on every senior seat looks like waste on the vendor's own evidence.

Vendors bend when pushed. One pricing analysis lists Copilot Business at $21 per user per month, with an $18 promotion running to 30 September 2026, after a reported cut from $30 in December 2025. On the consumer side, a Microsoft customer complained on the company's forum about a jump from $6.99 to $9.99, adding: "MS says it's to pay for Copilot AI." A volunteer moderator pointed them to Microsoft 365 Classic, without Copilot, at the original fee, so Microsoft already sells a version without the AI to customers who think to ask for it.

Before your next renewal, request the commercial equivalent of that Classic plan at last year's rate, and write the AI tier into the contract as an option that switches on only for named teams once the vendor has shown, from your own data, a measured fall in handling time or cost per case. Any account manager who will not sign that clause is telling you which side of Apollo's chart your money is meant to end up on.

Prompted by No Signs of AI Payoff Outside the Tech Sector | The Daily Spark, apollo.com.