Data valuation: how companies put a price on data

Data valuation is the process of estimating what a dataset is worth. Valuers use three standard approaches: cost (what it would take to recreate the data), market (what comparable data sells for) and income (the cash the data is expected to produce). For data licensed to AI buyers, public market evidence is thin, so the approaches work best together.

Valuers often apply more than one approach and reconcile the results, so no single number settles the question. This page shows how each approach applies to company records, which factors AI buyers weigh, and how the terms of a license move the price. We write for company owners and operators considering data licensing. Our disclosure page explains how buyer referrals fund this site.

Data valuation methods: the three approaches

Accounting and valuation practice group valuation techniques into the same three approaches. KPMG's fair value measurement handbook describes the market approach as using prices and multiples from comparable assets, the income approach as converting future cash flows into a single present amount, and the cost approach as the amount needed to replace an asset's service capacity.

Cost approach: what it would take to recreate the data

The cost approach rests on a simple idea: a buyer will not pay more for an asset than it costs to obtain a substitute of comparable usefulness, as the KPMG handbook puts it. Applied to data, you add up what it would take to rebuild the records: staff time spent creating and documenting them, the software that stores them, and the effort to clean and export them.

Two things limit the approach. First, substitutes may not exist: ten years of support tickets reflect ten years of real customers, and no budget today can buy that history.

Second, cost tells you what the data took to build and says little about demand. A company can build a useful dataset cheaply as a byproduct of work it does anyway.

Example: A 60-person services company wants to value eight years of project records with the cost approach. It adds the staff time spent documenting each project, the software that stores and tracks the work, and the effort to clean the records for export. The total is a useful reference point, though it leaves out the eight years of history, which cannot be bought.

Market approach: what comparable data sells for

The market approach compares your data with similar data that has changed hands. For operational company records the evidence is thin. Licensing deals are private, the parties rarely publish terms, and the figures that do appear come from press coverage or from buyers describing their own programs.

One independent report shows that such deals happen. In July 2026, The Information reported that data labeling startups were paying HVAC companies $150,000 for data. It is a single reported figure for one kind of business, and it does not tell you what another company's records would fetch.

When a buyer publishes a range for its own program, the range describes what the buyer says it pays, without the deals behind it. Treat any figure you find as a claim to test, and ask three questions of it:

1. Who published it? A buyer's own range, a press report and a commercially funded guide each carry different weight.
2. What does it describe? A headline range, a single deal and an average are different measures.
3. When was it published? Prices in a young market move, so a figure from a year ago may no longer hold.

In practice, the most useful market evidence you will get is an offer for your own records. Where you can, ask more than one buyer, and compare the offers on the same scope and terms.

Income approach: what the data will earn

The income approach values data by the cash it is expected to produce, discounted to a present amount because money received later is worth less than money received today. Company data can earn in two ways:

1. Licensing income. The fees you expect over the life of a license, minus the cost of preparing the data, legal review and any taxes.
2. Internal value. The savings or revenue you get from using the data yourself, such as a knowledge base that shortens support handling time or project history that sharpens bids.

The second stream is real but hard to separate from everything else your company does, so any estimate rests on assumptions you should write down.

Example: A company expects one license fee, paid in two installments, and estimates the preparation and legal costs of earning it. The income approach subtracts those costs from the fees and discounts the later installment to its present value. The result is the data's licensing value under those assumptions. Change the assumptions and the result changes.

What AI buyers weigh

No buyer publishes a pricing formula. The factors below follow from standard valuation reasoning and from what a buyer can do with the records.

Most of these describe the data. Exclusivity is a term of the deal, so ask each buyer to price exclusive and non-exclusive versions.

How the terms move the price

Two deals for the same records can price differently because the terms differ. The directions below are what we would expect from what each term gives the buyer. No buyer has published the size of the differences, so ask for them.

How to value your own data

You can build a working range from your own inventory:

1. Inventory your systems: contents, years of history, volume and export options.
2. Set aside what you cannot license, mainly data covered by client contracts and NDAs.
3. Rate what is left on the factors above: depth, structure, cleanliness and personal-data load.
4. Estimate a cost floor and an income case: what the records would take to rebuild, and what a license would bring after preparation and legal costs, discounted to today.
5. Adjust for terms: exclusivity, duration and who does the personal-data removal.
6. Write a low case and a high case, with the assumptions behind each, and test both against any offer you receive.

Example: A 90-person engineering firm has twelve years of project records in a consistent format, a documented knowledge base and a helpdesk history. It sets the helpdesk aside because it is mixed with client material, and rates the project records and knowledge base as deep and well structured. Its low case is a non-exclusive license for the project records alone. Its high case adds the knowledge base and a longer term. It writes down the assumptions behind each and checks both against the first offer it receives.

Questions to ask a buyer about price

An offer is easier to judge when you know what it covers. Ask each buyer:

1. Which systems, record types and years does the price cover?
2. Is the price for an exclusive or non-exclusive license, and for what term?
3. Who removes personal data, and who pays for it?
4. When is payment due, and what triggers it?
5. How would the price change if you added systems or years later?
6. If the buyer publishes a range, where in it would records like yours fall, and why?

When you need a formal valuation

Most owners weighing a license do not need a formal valuation. The buyer proposes terms, and you compare them with your own inventory and estimates. A formal valuation prepared by a valuation professional is the route when a financial reporting, tax or transaction requirement calls for one, for example when the buyer of your company allocates the purchase price among the assets it acquires.

A formal report sets out the approach, the assumptions and the resulting range. Having your inventory and contracts ready shortens the work. Our page on data as an asset explains how accounting treats company data.

Common valuation mistakes

• Anchoring on the highest figure you have seen. A headline figure is the top of someone's range or a single reported deal, so a company with standard records should not plan around it.
• Counting data you cannot license. Client data and personal data come out first.
• Treating cost as value. What the records cost to build says little about demand.
• Counting the same benefit twice. Savings from using the data internally and fees from licensing it can both be real, but a license that limits your own use may reduce the first.
• Ignoring preparation costs. Removal, review and legal work come out of the fee.
• Comparing offers with different scope. Two prices only compare when they cover the same systems, years and terms.

What a valuation leaves out

• Rights. Data you do not own or cannot license has no value to this group of buyers, however good it is.
• Costs. Preparation, legal review and staff time reduce what you keep.
• Timing. Approval and payment can take months, which matters if you discount future cash.
• Source quality. Our research found most published figures are self-reported or come from commercially funded guides, including this site, which is why we treat them as claims to test.

Next steps

Write down your inventory, then read the selling handbook for the full process and the data licensing guide for the terms that move price. If you want to see which records buyers look for, our page on what company data AI labs buy goes type by type.