Build a data product or license data to AI developers?

Build a data product or license the records: which should a mid-sized company choose?

Most companies with 50+ full-time employees at peak (contractors excluded) should license historical records rather than build a data product. A product needs a team, a customer base, support and a roadmap. A license is a one-time payment for a defined snapshot of records the company already holds, agreed on terms the company approves and signs.

The exception is a company whose core business already is data: a benchmarking firm, a data vendor, a platform whose customers pay for aggregated insight. For everyone else the build route competes with the core business for the same scarce people.

How do the two routes compare side by side?

The table compares shapes, not outcomes. No price or result is implied for either route.

When does building a product win?

Building makes sense under three conditions, and all three should be true.

• The data is already packaged for outsiders, with documentation, access controls and a named owner.
• There is a repeatable customer who pays for current data, not only historical archives.
• Leadership is willing to run a second business line, with its own support and compliance load.

A CFO can test this quickly by asking who would own the product P&L next year. If nobody volunteers, licensing is probably the better frame.

When does licensing win?

Licensing wins when the value sits in history rather than currency. Years of tickets and resolutions, deal histories with outcomes, approvals and exceptions have value as a record of how work was done. Researchers at Epoch AI have projected that language models could fully use the stock of public human-written text between 2026 and 2032, a forecast with wide uncertainty, which is one reason permissioned business records attract interest. For a view on how buyers think about budgets, see the page on how much AI labs spend on data.

Licensing also suits companies that cannot absorb distraction. A wholesale distributor with a lean finance team has little use for a product roadmap; the wholesale distribution page explains which operating records tend to matter there.

The 3-gate test for fractional CFOs

Use three gates in order. Stop at the first no.

1. Gate one, rights. Did the company create the records, and do customer contracts and employee notices allow licensing? If not, neither route works.
2. Gate two, capacity. Could an internal owner spend a few hours a month on a license, or a full team on a product? Be honest about the second.
3. Gate three, timing. Is there a window, such as a system migration, add-on integration or exit preparation, when records may be lost or locked up?

• Rights confirmed in writing
• An authorized sponsor named: owner, CEO, CFO or authorized representative
• Exports possible from the main systems
• No existing AI-training license over the same data
• Insurance questions routed to the broker; see cyber and D&O coverage

How should the revenue be planned?

A license is typically a one-time payment, so plan it as non-recurring and keep it out of run-rate assumptions. How to label it in reporting is covered in whether one-time data license income counts in ARR. For the year-end conversation where one-time items get discussed, use the year-end tax planning checklist.

Government work changes the answer: where a company holds controlled information, much of the archive may be off the table, as the government contractor page explains.

How does SourceX fit?

SourceX manages data licensing for companies, from sourcing and rights review to delivery and payment, between businesses that hold proprietary data and the AI developers who license it. SourceX does not train AI models. The company keeps ownership, approves scope and price, and signs only if the terms work. Large deliveries can stay in the seller's own storage.

You do not run any of that. Partners make introductions and give basic fit information only, and never export, upload or describe confidential records. Partners earn 25% of the eligible platform fees SourceX actually collects from the referred company's licensing deals, capped at $100,000 per referred company. The reward is paid only after the buyer pays and SourceX receives its fee; an introduction, meeting or signed agreement alone does not trigger payment, and no reward is guaranteed. The reward is a share of SourceX's fee and is never deducted from what the company receives. If you are a licensed professional, check your own rules on referral fees and disclosure first, and read the program terms.

What mistakes do advisers make when comparing the two?

The common error is comparing a finished product with a hypothetical license, or the reverse. Compare like with like: the full cost of a team, a customer base and ongoing support on one side, against a single prepared snapshot and an approval process on the other. A second error is treating data volume as value. Buyers care about connected, outcome-labeled records, so a smaller, well-structured archive can matter more than a very large messy one. A third is skipping the rights check because the business "obviously" owns its own records; customer contracts and employee notices often say otherwise.

Next step

Run the three gates on one client this week; the company fit checker gives a preliminary, non-binding screen and who qualifies lists the baseline. If the client passes, register as a partner and make the introduction.