AI spend per employee: what the Ramp AI Index shows

Why this number matters for a smaller company

How much do other companies actually spend on AI? The question comes up in almost every conversation, usually just before or just after somebody decides to try something. The Ramp AI Index is one of the few public sources that puts a figure on AI spend per employee.

What normally passes for an answer is an anecdote. Somebody knows a firm pouring money into it; somebody else knows one that never left the free trial. A public measurement is worth having even when it comes from another market.

The index covers American companies that run their spending through the Ramp payments platform. Not Belgian smaller companies, and not a cross-section of the economy either. The figure is no benchmark you have to hold yourself against. It is a starting point for a conversation that otherwise stays vague, because it shows how far apart the companies at the front sit from everybody else.

That distance is the subject of this article. It says little about what you should spend and a lot about how the leading group works.

What the Ramp AI Index measures

Ramp is an American payments and spend-management platform. Its Economics Lab aggregates the card and bill-pay transactions of more than 70,000 U.S. companies into a monthly index of AI spending. AI spending here is whatever those companies pay AI vendors by card or bill payment.

The August 2026 edition looks at July's numbers. It ranks companies by what they spend per employee per month on AI and pulls out three points.

According to the Ramp AI Index (August 2026, Ramp Economics Lab), in July 2026 the top 1% of U.S. businesses on Ramp spent a median of roughly $7,400 per employee per month on AI, the top 10% spent about $650, and the median firm spent about $11.95 — converted at roughly 0.92 euros to the dollar, that is approximately €6,800, €600 and €11.

Exchange rates move, so read the converted figures as approximations. The dollar figures are what Ramp actually published.

The top number catches the eye, but the middle of the pack says more. The gap between the median firm and the top 10% is an order-of-magnitude difference you will not find in many other spending lines.

What the gap is not

The obvious reading is that the leaders own better tools. The numbers do not show that. No AI vendor in the index has a majority: 43.5% of U.S. businesses on Ramp paid for Anthropic in July, 39.7% for OpenAI and 4% for xAI, the last of those posting its strongest monthly growth since July 2025. No single supplier holds this market.

Nor is there a recipe to copy. The index records which vendors get paid. What the work is used for does not appear in it. Of two companies with the same amount on the invoice, one may be spending it on a string of disconnected trials, the other on a single chain that runs every day.

Ramp adds a caution of its own. With this year's most capable models, the lab found a new upper bound on what businesses are willing to pay for AI. Buying more model power does not automatically deliver more.

What the gap actually is

The index measures payments, not decisions. Who inside those companies picks which process goes first appears nowhere in the dataset. That is also why the amount on its own gives you so little to hold on to.

The shape of the distribution does say something. Spending at that level means AI has landed in the daily run of the business; a single pilot never adds up to those amounts.

That takes somebody who decides which process gets tackled first, somebody who keeps the documentation of what was built, and somebody who settles it when two departments claim the same thing. Those roles appear in no index. That is still where companies which get somewhere with automation part ways with the ones that stay stuck at a pilot.

What this means if you start with one process

Copying the budget of the top 1% makes little sense for a smaller company. Those businesses operate at a scale and in a market that have little to do with yours. Their number describes what they spent; it prescribes nothing to anybody.

What does travel is the question underneath the spending. Which process eats the most attention today from people who should be doing something else? Who decides about it? And who holds on to what gets built?

In practice it starts with the one process everybody in the building knows is grinding. The supplier invoice that has to get from the mailbox to the accounts. The request that sits waiting because one person has to check it. You recognise processes like that without any index.

Those three questions you can answer without a single decision about tooling and without a purchase. Get them sharp and the first process worth doing usually names itself.

The limits of this number

Five things travel with this figure the moment you repeat it.

It covers American customers of one payments platform. That is not a representative sample of the economy, and certainly not a picture of the Belgian market.

The dataset also leans heavily towards technology companies, the group quickest to pay for AI. That pulls the top of the distribution up.

Ramp only sees what passes through card and bill payments. A framework contract settled elsewhere is not in there, so the picture is incomplete.

The median is calculated across every employee of a company, including everyone who never touches AI. A business with many people out on the road slides down the ranking for that reason alone.

Finally, the index moves. Ramp updates it monthly; this is the August 2026 edition, on July's data. If you reuse the figure later, name the edition alongside it.