Data Valuation Methods: Cost, Market, Income

Data is valued with the three approaches used for other intangible assets. The cost approach estimates what it would take to recreate the data, the market approach looks at what comparable data has sold for, and the income approach estimates the cash the data can generate. Most credible valuations use all three and compare them.

What are the three approaches to valuing data?

The cost, market and income approaches each answer a different question, and each gives a different number for the same dataset. The cost approach asks what you would have to spend to build it again. The market approach asks what a willing buyer has paid for something similar. The income approach asks what the data lets someone earn.

None of them is the "right" one; they are lenses. A dataset that is expensive to build but useless to buyers scores high on cost and low on income. A dataset that is a by-product of operations might have almost no creation cost yet be worth a great deal to a model builder. Looking at all three shows you where the real value sits and where your own assumptions are weak.

If you want a quick first estimate before doing this work properly, the free ten-question valuation at /evaluate gives a starting view, and how much is my company's data worth explains the non-method factors, such as uniqueness and cleanliness, that move any of these numbers.

How does the cost approach work?

The cost approach adds up what it would take to collect, clean, label and document the data again from scratch. It is the easiest method to calculate because the inputs are your own spending, and it gives a defensible floor for a negotiation, though rarely a ceiling.

A reproduction cost estimate normally includes:

Collection cost: the infrastructure, staff time and third-party fees needed to gather the raw records.
Preparation cost: cleaning, deduplication, de-identification, formatting and quality checks.
Annotation cost: any labels, categories or structured fields added by people or tools.
Documentation cost: schema notes, data dictionaries and provenance records.
Time cost: the delay a buyer avoids by licensing instead of rebuilding.

Where the cost approach works

It suits datasets where recreation is genuinely possible: a labelled set built by paid annotators, a scraped and curated collection, a survey. Anyone could in principle repeat the work, so the cost of doing so caps what a rational buyer would pay.

Where it fails

It breaks down for data that cannot be recreated at any price.

Ten years of transaction history is not reproducible by spending money today, because the events already happened. For such data, cost understates value badly. It also overstates value when a company spent heavily on something nobody wants. Cost tells you what was spent, not what the data is worth.

How does the market approach work?

The market approach values data by reference to what comparable datasets have been sold or licensed for, adjusted for differences. It is the most persuasive method when real comparables exist, because it reflects what buyers actually pay, but data markets are young and good comparables are often thin.

To apply it, you identify comparable transactions, adjust for size, freshness, exclusivity, license scope and quality, and derive a range. The adjustment step is where judgment enters. A comparable with exclusive commercial rights is not comparable to your non-exclusive evaluation license unless you correct for that difference. How to structure data license tiers and scope shows how much the rights alone change a price.

The comparable problem

Data deals are often private, and published prices, where they exist, rarely describe the license terms. A headline number with no scope behind it is close to useless. Be wary of anyone who quotes a price per record or per gigabyte as if it were a standard rate. Unit prices vary enormously with content, rights and buyer need.

Practical sources of market evidence

Within your own business, past quotes and offers are evidence. So are the bids you receive in negotiation: on LogitAtlas a buyer's bid is a live market signal, and the accepted bid becomes the checkout price. Asking several buyers what they would pay for different scopes teaches you more than any desk estimate.

How does the income approach work?

The income approach estimates the cash flows the data can generate and discounts them to a present value. It is the most relevant method for data that feeds a specific revenue-producing use, and the most sensitive to assumptions, because small changes in forecast revenue or discount rate swing the answer.

There are two common versions. In the first, you estimate the income the owner earns from the data, for example through better pricing or fewer losses. In the second, you estimate the income a buyer could earn from it, and then decide what share of that income a buyer would pay for the data. The buyer will not pay all of it, because they must also fund the models, engineering and sales that turn the data into revenue.

A hypothetical worked example

This is a hypothetical example with invented numbers, not a benchmark.

Suppose a buyer expects a model trained on your dataset to add 200 units of annual gross profit for three years. Suppose they would only pay for around a fifth of that, because other inputs earn the rest. That implies about 40 units a year, or 120 over three years before discounting. Apply a discount rate to reflect risk and timing, and you might land near 100 units.

Change the share from a fifth to a tenth and the figure roughly halves.

That sensitivity is the point. The income approach is useful for understanding which assumptions drive value, and for explaining a price to a buyer, but it should never be presented as precise.

How do you choose between the three approaches?

Choose by what kind of data you have and why you need the number. Use cost when data is replicable and you need a floor, market when comparables or live bids exist, and income when the data supports a clear revenue use. In most cases the honest answer is to run two or three and reconcile the gaps.

A simple decision guide:

If the data could be rebuilt with money and time, start with cost.
If you have received offers or can find well-described comparable licenses, add market.
If a buyer can point to a specific product or saving the data enables, add income.
If the three results diverge widely, investigate before choosing: the gap usually reveals a hidden assumption.
If you only need a rough price to open a conversation, use the lower-middle of your range and let bids refine it.

What factors adjust a data valuation, whichever method you use?

Whatever method you choose, a handful of attributes raise or lower the figure. They matter more than the arithmetic, because they change the inputs to every method.

Uniqueness: data that nobody else can produce supports a higher figure than data available from many sources.
Freshness and update cadence: stale data loses value; a regularly refreshed feed keeps it.
Coverage and size: breadth across customers, regions or time periods increases usefulness.
Cleanliness and documentation: well-documented data reduces the buyer's cost to use it. Poor documentation cuts value sharply.
Provenance and rights: if you cannot show where the data came from and that you may license it, buyers discount heavily or walk away. How to verify data provenance describes what buyers look for.
Privacy exposure: personal information reduces what you can license and raises preparation cost.
License scope: exclusive, long-term, broad-use rights command more than narrow ones.

What are the common mistakes in valuing data?

The biggest mistakes are anchoring on cost, treating a single price per record as a rule, and ignoring the cost of making the data saleable. Another common error is valuing data you have no right to license.

Anchoring on sunk cost. What you spent on a data warehouse says little about what a buyer will pay for the contents.
Counting data twice. If you value the data by the model revenue it enables, you cannot also add the model's own value.
Ignoring preparation. De-identifying, documenting and packaging the data costs money. Net the cost out of what you expect to receive.
Assuming value is permanent. Data can lose relevance as formats, products and markets change.
Valuing data you cannot sell. Some data is restricted by contract, consent or law. A figure for that data is not a price, it is a fantasy. See what types of company data you can sell.
Mistaking an accounting number for a price. A figure built for internal reporting or negotiation is a different thing from a transaction price.

Do accounting and tax rules change how data is valued?

They can, because accounting treatment, tax treatment and a negotiated price are three separate questions that may use different rules. This section is general information and not legal, tax or accounting advice; speak to a qualified lawyer and a tax or accounting professional about your own situation.

For a negotiation, none of those rules bind you. The price is whatever a willing buyer and seller agree. For financial reporting or tax, specific standards may apply to how internally generated data is recognised, and the rules differ by country. Do not assume a number that works in a negotiation will also work on a balance sheet, or the reverse.

If you want an outside view, a valuation specialist can document assumptions in a way an auditor or counterparty will accept. For most first-time sellers, a documented range built from the three approaches and tested against real bids is enough to start. When you are ready to test it, list your dataset on LogitAtlas: listing is free and the commission is 10%, deducted only when an order completes.

Frequently asked questions

Which data valuation method is the most accurate?

None is universally the most accurate. The market approach is the most persuasive when strong comparables exist, the income approach fits data tied to a clear revenue use, and the cost approach suits data that can be rebuilt. Reliable valuations compare at least two methods and explain the difference between them.

Can I value data by price per record?

Only as a rough sense check. Price per record hides differences in content, freshness, rights and buyer need, so it is not a standard rate. Two datasets of equal size can differ enormously in value. Use per-record figures to compare your own scenarios, not to quote a market price.

Is the cost of collecting data its value?

No. Cost is a floor for data that can be recreated and an irrelevant figure for data that cannot. A dataset cheaply produced as a by-product of operations may be worth a lot to a model builder, while expensive data nobody wants is worth little. Buyers pay for usefulness, not your spending.

How do I get market evidence if I have never sold data?

Use buyer bids and conversations. Ask potential buyers what they would pay for different scopes, and treat each bid as a data point. Offering a small evaluation license is another way to see real prices. Early evidence will be thin, so widen your range and update it as offers arrive.

Do I need a professional valuation to sell data?

Not to sell. A documented range from the three approaches, tested against real bids, is usually enough for a first sale. A professional valuation can help for tax, financial reporting, investment or disputes, where an outside party must rely on your figure.