Revenue per employee in AI-native companies vs SaaS (2026)
After my master's I took a sales job at Oracle. I was handed about 200 accounts. Before a call I looked up the company, checked whether they had used Oracle, found a person, and wrote a note that did not sound like every other email. I once counted it at about three hours an account. On that list, that is six hundred hours. Most of a quarter, gone, before the work I had been hired to do.
I think about that list when someone asks about revenue per employee in AI companies. The job did not shrink. The preparation moved. The question is what a person can do with the hours that come back, and when those hours are still a reason to hire.
Revenue per employee is the plain version of the question. A year of revenue, divided by the people. ARR per employee swaps in annual recurring revenue: a snapshot, not the money a completed year brought in. Revenue per headcount is what you get when the denominator is a headcount date, often year-end payroll, rather than the average team. I keep the formula, the bands, and the basis of each figure on the revenue per employee page.
Where private SaaS sits
SaaS Capital's 2026 benchmarks come from its 15th annual survey of more than 1,000 private SaaS companies, completed in March and published on 30 July 2026. The median is $141,125 of ARR per employee, up from $129,724 the year before. From $1 million to $3 million of ARR, the median they cite is $109,644.
Under $500,000 a person is A0 on the bands I use. Both of those medians sit in A0. The AI-native revenue per employee figures people repeat are not a small step up from $141,125. Each one needs a date and a denominator.
In the same SaaS Capital cut, bootstrapped companies come out ahead of equity-backed ones at a similar size. From $5 million to $10 million of ARR, the medians are $177,240 and $152,295. Still A0. A lower ratio can be the plan, if the plan was to hire before the revenue arrived.
What founders on the show have said
These are their claims, tied to the episode. They are not audits.
Adam Holmgren says Fibbler reached $1.2 million ARR with two people and no sales team. Divide his two figures and you get $600,000 ARR per person, in A1. The division is mine. He also describes a target of $10 million ARR with fewer than five people. A target is a plan.
Amos Bar-Joseph describes scaling one seller to 100x instead of hiring ten. The episode records three founders, more than 200 customers, and seven-figure ARR in 2025, with no further hires. Seven-figure is a range, so I will not turn it into ARR per employee. The 100x line is an ambition he described, not a result I measured. The rule under it is the one I use: extend the person you have, and hire when that capacity runs out.
Omar Ismail says Ascend reached $25 million ARR with no growth team, with an aim of $25 million to $100 million without adding people in a meaningful way. I do not have a workforce in that episode that I will divide into the $25 million, so there is no ARR per employee for Ascend here.
Adit Trivedi describes hyperGTM as a team of nine, with two engineers. On his account, much of the old headcount was work agents now cover. Hire for a problem you have already felt, not for a team drawn in advance.
Molo Lorenz talks about a four-person go-to-market engineering team, and more from each rep, against a Moss target of €100 million ARR without a hiring spree. The €100 million is a target. On the episode, clean data came before the agents, and automated outbound was tried too early.
Adam Robinson, to whom the book is dedicated, spoke about two companies he had built. He described Retention.com reaching $13 million ARR with six people. He said that, at the time of our conversation, RB2B was at $8.6 million ARR, and that the CTO had not touched the code in over a month. The revenue per employee page keeps a separate May 2026 snapshot: $9 million ARR across three people, from his account and Henry Shi's leaderboard. Two dates, two records. I am leaving both as they were set down.
Jaspar Carmichael-Jack says Artisan has about 50 people, against a plan of more than 100, with revenue growing faster than the team. He also said the token bill had reached the point where another hire could be the better spend.
The cost the ratio leaves out
Joshua Wohle describes a $75,000 monthly AI bill at Mindstone. On his account, that month did the work of about 50 software engineers, and people did not yet know where to point the extra effort. I am using $75,000 as his report of the bill, not as a benchmark.
Yannick Dillen makes the cost point from the other end. Zero employees does not mean zero costs. A €1,000 token bill is small for a company. On a student budget, he said, it can stop the project.
Jacco van der Kooij is the counterpoint I want kept in the room. Once tokens cost more than people, revenue per employee is closer to a VC metric than an operating one. Revenue per headcount then stops telling you whether the business got cheaper. I still use the metric. I read it next to the bill, and next to what the customer gets.
Epoch AI, in Luke Emberson's note, estimates about $9 million of revenue per employee at Anthropic and about $5.5 million at OpenAI. The page says the figures are approximate: annualized revenue, and staff counts estimated because the dates do not line up. I would not place either company on the A0 to A5 chart as a completed SaaS year.
Other rows on the revenue per employee page are not from the show. AKOOL, November 2024: $800,000 ARR per person, from $40 million invoiced ARR and 50 employees, an ARR snapshot. Super.com, July 2025: $1 million, from $220 million of reported annual revenue and 220 employees. Midjourney, February 2025: an estimated $12.5 million, from a $500 million annual figure and 40 employees, not confirmed accounts. Medvi, 2025: about $200.5 million, from $401 million of reported revenue divided by the founder and one employee. Outside contractors, and the clinical and pharmacy partners, are excluded. Target, 2025: about $252,000, from $104.780 billion of net sales divided by 415,000 year-end employees, including part-time and seasonal staff. Say what the numerator is. Say who is missing.
A0 to A5, kept apart from how the work runs
Comparison bands, not an industry standard. The top of each band is excluded, so exactly $1 million is A2.
- A0, under $500,000
- A1, $500,000 to under $1 million
- A2, $1 million to under $10 million
- A3, $10 million to under $100 million
- A4, $100 million to under $1 billion
- A5, $1 billion and above
A0 holds young companies and businesses that still depend on a lot of people. It does not mean a company uses no AI, or that it is badly run. L1 to L5 answer a different question: how independently the work runs. A company can move on the autonomous business scale before the ratio moves, and it can climb these bands without becoming more autonomous. A high number is not a level of autonomy.
What the hours are for
If an agent had prepared that Oracle list, I would have reached the calls sooner. A full day of conversations would still have been one person using judgment, and I would have needed time between them. Saved hours are not automatically more hours of the same intensity.
Dora Kondeva described the incumbent version at Snowflake. The episode reports EMEA sales cycles shorter by 25 to 30 percent within six months, and 2.3 times as many meetings with high-potential accounts on 38 percent less account-based marketing spend. She talks about where the time saved on sales prep goes. On her account, it went back into selling.
Bhanu Chawla says he runs marketing at FRAME alone, after 55 marketers at his previous company. He would still hire, for work that has to happen in person, events among it. Starting with AI is a sequence, not a ban on people.
Nick Knuppe's target at neno is 150 clients per accountant, against a traditional load of about 30 to 35. That 150 is the ambition. The episode does not say an accountant there already carries it. He also says the firm had passed 150 customers. Different count. I will not add them.
Jonathan Kvarfordt, from Salesforce, says that handing out AI access does not change the work. The episode is about freeing people for higher-value work, and the page has a chapter on revenue per agent. There is no revenue-per-agent figure on the page, so I will not invent one. Judgment stays with the person.
I have advised 30+ companies, most of them through Revenue Wizards, the Amsterdam RevOps and go-to-market consultancy I co-founded. Familiar work gets faster first. Then someone chooses what the time is for. More customers. A better answer for the ones you have. A way for a colleague to carry it when you are away. Or a hire, because the limit showed up.
Five Dots is the company I co-founded so I would have to follow this. Growth After Headcount, the book, is forthcoming, aimed at Q4 2027. It is not for sale.
Sources
- SaaS Capital, 2026 revenue per employee benchmarks, 30 July 2026.
- Epoch AI, Luke Emberson. Annualized revenue, estimated staff.
- Bands and company rows: the revenue per employee page.
- Guest figures: the linked episodes. Each is the guest's claim, or a result the episode page reports.