Outrich Data buys from companies with 50 or more employees and at least two years of records, so smaller companies do not currently qualify. Small-company data can still be valuable when it is specialized, deep or tied to expert work. Other buyers publish lower thresholds, and a company can prepare now for later eligibility.
The honest starting point
Outrich Data sets a minimum of 50 employees and two years of records. A company below that line will be told so by the eligibility check, and this guide does not pretend otherwise. The threshold is a working rule about where depth of data usually begins. It is not a judgment about the quality of a smaller company's work.
Being told no by one buyer does not close the question. Buyers draw the line in different places, and the factors that make a dataset valuable are not all tied to headcount.
The rest of this page is for founders and operators of smaller companies who still want to understand their position. It covers why size minimums exist, what can make a small company's data valuable anyway, which other routes are open and how to be ready when the company grows.
Why buyers set size minimums
AI labs want volume and variety. A dataset is more useful when it shows many people handling many cases over a long time. Ten people produce fewer conversations, fewer documents and fewer kinds of problem than five hundred, and the patterns in a small team's work can reflect individual habits more than general practice.
There is also a fixed cost on the buyer's side. Each deal needs a rights review, de-identification, a negotiated license and a delivery process. That work is similar whether the dataset is small or large. Below a certain size, the cost of doing the deal properly is out of proportion to the data.
Privacy adds a third reason. In a very small company, each employee's writing is a large share of the whole and easier to recognize even after names are removed. Re-identification risk falls as the number of contributors rises.
What other buyers publish
Thresholds differ across the market, and a few buyers state theirs publicly. On their own pages, Scale AI says it prefers companies with 40+ employees, micro1 asks for 30+ full-time employees, and Mode asks for 30+ employees and accepts fewer in some fields. Troveo, a licensing marketplace, offers an estimator that starts at 20 employees.
There is also a market for the data of companies that have closed. According to MIT Technology Review, AI labs buying from closed startups have paid about $25,000 for a company code base and about $15,000 for workspace data such as chat and documents. Those figures show two things. Small-company data does have buyers, and the amounts involved are modest.
These details are taken from the companies' own pages and from the cited report on the date shown in the sources. They change, so check the current position with each buyer.
| Buyer | Model | Published minimum size |
|---|---|---|
| Scale AI | Data partnership program | 40+ employees preferred |
| micro1 | Buys operational data | 30+ full-time employees |
| Mode | Buys operational data | 30+ employees, fewer in some fields |
| Troveo | Licensing marketplace | Estimator starts at 20 employees |
| Outrich Data | Direct buyer | 50+ employees and 2+ years of records |
What makes small-company data valuable anyway
Headcount is a proxy. What buyers really want is depth, and some small companies have more of it than their size suggests.
Specialization is the main factor. A small engineering consultancy, a boutique research firm or a niche software company may hold expert work product in a field where public data is scarce. A few thousand reviewed documents from specialists can matter more to a lab than a far larger volume of general customer service chat.
Volume per person is another. A small company that runs a high-traffic product may have a support archive or usage logs far larger than its team. History counts too. A company that has kept complete records for many years can have more depth than a larger, younger one.
- Specialist work product in a field with little public data
- High record volume relative to headcount, such as a busy support desk
- Long, continuous history in one system
- Complete workflows with recorded outcomes
- Code with full review and issue history
- Data that was never public and never licensed
- Clean rights: clear ownership and no restrictive customer terms
Other routes for a smaller company
A company under the threshold has options besides waiting. The first is to approach buyers with lower published minimums, such as those named above. The second is a marketplace, where a standardized dataset can be listed for several buyers without a minimum company size, though a listing does not guarantee a sale.
A third route is a direct approach to a company in the same field. A larger firm building its own AI tools may value a small specialist's data more than a general lab would. A fourth applies to companies that are winding down. The data of a closing company can be licensed as part of the wind-down, subject to the same rights checks and to any duties owed to customers and creditors.
Whatever the route, the protections are the same at any size. Confirm the rights first. De-identify before sharing. Use a written license with a defined scope. Do not deliver the full dataset before payment is secured. The escrow service described at /data-escrow is available for data deals between other parties, where its role is to hold funds and verify delivery, not to buy.
Risks that weigh more on a small company
Smaller companies should be cautious for reasons that have nothing to do with price. A startup's data is often its competitive advantage. Licensing the code base or the product logs to a lab could help a future competitor more than the fee is worth.
Customer trust is concentrated. A small company may depend on a handful of customers, and a single customer objecting to a data deal can matter more than the revenue. Read each contract, and consider asking key customers directly.
Investors and acquirers will also ask. A license granted today becomes a disclosure item in a later funding round or sale, and an exclusive license can reduce what an acquirer is willing to pay. Founders should check investor agreements for consent rights before signing anything.
Management time is the last cost. A data deal needs exports, a rights review, legal drafting and attention from the founders. That cost is real even when no fee is charged, and it should be set against a return that may be small.
How to become eligible later
Eligibility is mostly a matter of time and record keeping. A company that expects to cross the threshold can make its future dataset cleaner and more valuable by acting now. None of this requires a decision to sell.
- Keep records. Do not delete old tickets, threads and repositories unless a retention rule requires it.
- Keep work in systems that export cleanly, with timestamps, authors and thread structure.
- Sign intellectual property assignments with every employee and contractor from the first day.
- Review customer terms. Avoid clauses that forbid all secondary use of de-identified data unless a customer requires them.
- Keep each version of your privacy notice and the dates it applied.
- Give employees a clear monitoring and data use notice.
- Maintain a simple data inventory: system, record type, date range and owner.
- Track every license or data-sharing agreement you sign in one register.
Common questions
Why is the minimum 50 employees?
Can a small company still sell data to AI labs?
Does the two-year history requirement apply if we are large but new?
Can we use the free tools if we do not qualify?
Sources
- Scale AI, data partnership page2026-11-16
- micro1, data partnerships page2026-11-16
- Mode, data licensing page2026-11-16
- Troveo, business data page2026-11-16
- MIT Technology Review, AI labs buying data from closed startups2026-10-13