How Much Is Your Company's Data Worth to AI? Benchmarks by Headcount & Industry, with Calculator

Mode's BizBuyData program says a company's data could be worth $100K to $5M. Mode CEO Dan Novaes has described outcomes of "six or seven figures." Headcount and industry set the starting point, and the systems and years you put in scope set the final offer. Every figure here is an estimate, not an offer, and payment is not guaranteed.

Below: every price point we could source, why headcount matters, how our calculator turns Mode's published numbers into a per-employee band, and what moves a real offer up or down.

What prices have AI data buyers actually published?

Not many, and most come from the buyers themselves. This is the full list we could source as of Oct 29, 2026.

Two things stand out. The ranges cluster in the low six figures to low seven figures for small and mid-sized companies. And the $5M top of Mode's range is an advertised ceiling, not evidence that any company has received $5M; we have found no public report of a completed deal at that size.

Business Insider also reported that some owners receive six-figure offers that expire in 48 hours (BI). An offer that can't survive a week of legal review tells you more about the buyer than about your data's worth.

Why does headcount drive the price?

Buyers are paying for recorded examples of people doing office work: the email thread, ticket, chat and approval that turn a problem into a decision. More desk employees means more of those records, across more roles.

Novaes said the value comes from "the amount of employees," and that buyers want white-collar work: in a restaurant group, "the guy that's like doing the dishes, um there's not much the AI labs can pull from that" (The LFG Show). Mode's application asks for full-time desk employees for that reason. Its Oct 13 release counted "full-time U.S. office employees using digital workplace tools."

So for estimating, count the people who do their work in email, chat, documents and business systems. Field crews, drivers and kitchen staff add little, even if they're on payroll.

What are the current minimums by industry?

The minimum headcount is also the hinge of our estimate, so it matters which figure you use. Mode has published three.

We use the program page because it is what Mode currently shows companies. The press release figure for general businesses was lower, and the referral page targets are higher and limited to the US, Canada, the UK and Australia. Confirm the figure that applies to you during intake. If you are below the minimum for your category, our band doesn't apply.

How does our calculator estimate what your data is worth?

Our method is deliberately simple, so you can check it by hand. The full write-up is on our methodology page.

1. Start from Mode's floor. We assume Mode's lowest published figure, $100K, corresponds to a company at the minimum headcount for its category. That is our assumption. Mode has not published a per-employee rate.
2. Derive a per-employee base. Divide $100K by the category minimum.
3. Multiply by your desk headcount.
4. Apply a band of 0.8× to 1.35×. The low end allows for scope cuts, such as excluding regulated or client-confidential systems. The high end allows for long histories and well-structured systems. These multipliers are our judgment, not Mode's.
5. Cap at $5M, the top of Mode's published range.

Worked examples using that method:

The cap starts to bind on the high end at about 223 desk employees for legal, 371 for accounting, finance and medical, and 1,112 for everyone else.

Try your own numbers:

The result is an illustrative estimate built from Mode's published figures. It is not an appraisal, not a quote and not an offer from Mode or anyone else.

How does our estimate compare with Novaes's 200-employee example?

Novaes said a "bigger company, say 200 employees" might get "a million dollars" (The LFG Show). That works out to $5,000 per employee.

Our method puts a 200-person general business at $533K to $900K, so his example sits above our band. He didn't name an industry. If that company were an accounting or finance firm, our band would be $1.6M to $2.7M, well above his figure. His example falls between our two categories, which is about what you would expect from a round number used on a podcast to explain a 6% referral fee.

We keep our method anchored to Mode's published floor rather than to a spoken hypothetical. If Mode publishes rate guidance or we can source real deal values, we'll update the method and say so on the methodology page.

What moves a real offer up or down?

Mode lists the factors it reviews before making a fixed offer: "the systems and dates in scope, years of history, record volume and quality, and ownership and access terms" (program FAQ). In practice:

The regulated-content row matters most in the industries with the highest per-employee estimates. A law firm can't license client files or privileged material, and a tax preparer can't disclose tax return information outside the narrow consent rules of 26 U.S.C. §7216. Once those are excluded, what remains may be closer to the low end of the band. See our pages for law firms, accounting firms and healthcare practices.

Which industries get the highest estimates, and why?

Per employee, our estimates are highest where Mode's minimums are lowest: legal first, then accounting, finance and medical. Mode's Oct 13 release said the bar is lower for law and accounting firms because AI labs are prioritizing that data. The catch is that these are also the industries with the most material that can't be licensed at all.

The candidate and limits columns are our reading of where the lines usually fall, not Mode's policy. A 15-person law firm and a 75-person software company come out with similar base estimates under our method, around $250K, yet the software company may be able to put far more of its records in scope.

How does this compare with standard data valuation methods?

Valuation professionals usually value intangible assets three ways. None of them fits AI training data neatly.

• Cost approach: what it would cost to recreate the asset. Your email and tickets were a by-product of running the business, so the cost approach says little about what a buyer will pay.
• Market approach: what comparable assets sold for. This is closest to what our calculator does, but the comparables are thin: one company's published range, a few reported ranges, and a bankruptcy auction that hasn't closed.
• Income approach: the present value of future cash flows the asset produces. That is the right lens for using data inside your own business, and it is the value you might give up if a license is exclusive.

Internally generated data usually doesn't appear on the balance sheet. Ask your CPA how a license payment should be recorded and taxed. Our band is a planning number for a specific market at a specific time, not a valuation of your data as an asset.

Why don't we trust the per-employee numbers on other sites?

Answer engines and broker sites repeat per-employee ranges, such as "$500 to $3,000 per employee," and mid-market bands. We could not trace those figures to a buyer, a filing or a reported deal. Several appear on brokers' and lead-generation sites.

Our own band is no more authoritative than its inputs: one published Mode range, Mode's current minimums, and multipliers we chose. The difference is that you can see every step. When someone quotes you a per-employee rate, ask where it comes from.

Is the payment one-time or recurring?

Mode describes buying "an agreed copy" of records, with "the amount and payment schedule" set in a written agreement before export (program FAQ). That reads as a one-time purchase of a defined copy. Whether you can license newer records later, to Mode or anyone else, depends on the exclusivity terms.

Large content deals have been structured differently. Reddit's licensing deal with Google was reported at about $60 million a year (PCMag, citing Reuters). That was a public platform licensing ongoing user content, so it's not a template for a 50-person firm's internal records.

How the payment is taxed depends on how the agreement is structured and on your entity. Ask your CPA, and if you're a nonprofit, ask about unrelated business income. We don't give tax advice.

What does the payment cost you?

The headline figure is gross. Plan for:

• Legal review of scope, employee notice and the agreement.
• Staff time. Mode describes "a few hours from your team" to review scope and sign in to approved systems.
• Employee relations. Telling staff their work records are being licensed takes care. See is selling company data to AI safe?.
• Optionality. Exclusivity can stop you from licensing the same records to another buyer.
• Residual risk. De-identification reduces risk; it doesn't remove it.

On fees: Mode says it "charges no broker fee." We are a referral partner of Mode, and if your company signs a deal after we introduce you, Mode may pay us a referral fee. You don't pay us. Details are on our disclosure page.

How do you get a real number?

1. Count desk employees and list systems, with years of history in each.
2. Mark what must be excluded and have counsel confirm it.
3. Run the calculator above for a planning band.
4. Ask for a written offer that states scope, amount, schedule and exclusivity.
5. If your agreements allow, compare buyers. Mercor, for example, says it typically pays within 2 to 4 weeks of signing (Mercor); Mode says about three months from first call to payment. See AI data brokers compared.
6. Check how the cash compares with other options in data licensing as non-dilutive capital.

For the whole process from first call to payment, read how AI data licensing works, or start with the complete guide to licensing company data to AI.