AI spend per closed-won dollar: the first honest AI ROI metric for sales

"What is the ROI on AI?" is a question that gets asked at board level and answered with anecdotes. A rep saved two hours a day. A team feels faster. Nobody has a number, because the two halves of the number live in different systems: the spend is at the AI provider and the outcome is in the CRM, and neither knows the other exists.

For a revenue team there is a first metric that is honest, buildable and hard to argue with. It is AI spend per closed-won dollar, per rep, over a quarter. This is how to build it and, more importantly, how to read it without fooling yourself.

The definition

For each account executive, over a period:

• AI spend is every dollar of AI usage attributed to that person: their share of API usage, their enterprise seat's reported usage, and any AI plan expensed to them, spread evenly by day.
• Closed-won revenue is the sum of the amount on opportunities they own that closed won in the period, taken from the CRM as the CRM defines it.
• The ratio is spend divided by revenue, expressed as cents of AI per dollar closed.

Three cents means the rep spent three cents on AI for every dollar they closed. The number is small, which is the first thing people notice, and its distribution across the team is the second.

Why this one first

There are better metrics in theory. Incremental revenue caused by AI, controlled for everything else, is what you actually want. It also requires an experiment you probably cannot run, and a year of data you do not have yet.

The ratio has three properties that make it the right place to start.

It uses data that already exists. Every company with a CRM already tracks closed-won by owner. Every AI provider already reports usage. The only new work is the join.

It is per person, and so it varies. A company-wide ratio is one number and it tells you almost nothing. A ratio per rep tells you that the top quartile of closers spends at one rate and the bottom quartile at another, and that difference is where the insight lives.

It is not a claim about causation. The ratio does not say AI caused the revenue. It says what was spent next to what was closed, which is a fact, and facts survive board questions better than models.

Building it

Attribute spend to people. API usage is reported by key; resolve keys to people, allocate cost across keys by token share, keep what will not resolve in a visible unattributed bucket. Enterprise seat usage is already per person. Card charges resolve by cardholder. Sum by person and by day.

Read closed-won from the CRM, by owner. In HubSpot, that is deals with an amount, a close date and an owner, where the deal is closed won. In Salesforce, opportunities with an amount, a close date and an owner, where the opportunity is won. Take the CRM's own definition of won; do not invent one. The fields that count should be configurable, because a company with a custom revenue field will otherwise get a wrong number that looks right.

Join on email. The CRM owner has an email. The person the spend was attributed to has an email. That is the whole join.

Choose a window. A quarter is the natural period for sales, because deals take time and a single big close in a month makes a monthly ratio meaningless. Show the quarter, and show the trend across quarters once you have them.

Reading it without fooling yourself

The ratio is easy to misread, and the misreadings all go the same direction.

Low is not good. A rep with a near-zero ratio might be efficient, or might not be using AI at all while closing the same book they closed last year. The question is whether the reps who spend more are closing more, not whether anyone spends less.

High is not bad. A rep whose ratio is three times the team median is either wasting money or is the person who has worked out what to do with the tools. You cannot tell from the ratio. You can tell from the ratio plus their closed-won trend against last quarter, and that is the comparison to make.

No movement is the signal. The most useful pattern is spend above a threshold with no change in the KPIs over the same window. That is not proof that the spend is wasted, but it is the first place to look, and it is the case a good tool should surface on its own.

Deal timing distorts single periods. One rep closes a large annual deal in the last week of the quarter and their ratio collapses. Look at rolling windows and at the person, not the quarter in isolation.

Beyond account executives

The same construction works for every role that has a countable output. An SDR has meetings booked and opportunities sourced. A customer success manager has renewals and expansion revenue. An engineer has pull requests merged. In each case the ratio is spend over output, per person, over a window, with the same reading rules.

The role determines the denominator. The method does not change.

What the board actually hears

Present the ratio for the team, the distribution across reps, and the two or three people at each end, with what they closed. That is a story: here is what we spend, here is what it sits next to, here is where it is concentrated, here is what we are going to look at. It is not "AI produced eleven percent more revenue", which nobody can prove yet. It is the honest first step, and boards respond well to being shown the honest first step.

Arbitir builds this ratio from the connections a company already has: AI usage from Anthropic and OpenAI, closed-won from HubSpot or Salesforce with configurable fields, joined by email, shown per person on its Performance page, with a nightly check for spend that has no output movement beside it.