Skip to content

Deal economics

Dataset valuation vs data company valuation: separate the two numbers

By SourceX Editorial · Reviewed by Noah Loul ·

9 sections

Short answer

Dataset valuation prices one asset: the cash specific records can earn under normalized rights, after exclusions, preparation and a charge for the team and systems that help earn it. Data company valuation adds the engine that produces future datasets, then subtracts overhead and corporate risk. The two can diverge in either direction, so value them separately and never infer one from the other.

Value drivers

Dataset valuation

This page covers dataset valuation vs data company valuation: separate the two numbers.Illustrative

Key takeaways

  • A dataset's value is its records' cash under one normalized grant, less rights exclusions, preparation and a contributory charge for the team and systems.
  • A company's value adds the expected value of future datasets, then subtracts overhead and corporate risk; only the held datasets are asset value.
  • Tax and accounting rules already separate the information base from workforce, goodwill and going concern value.
  • In the illustration the same 57-unit archive sits inside companies worth about 18.5 times, 1.5 times and 0.1 times its value.
  • Inventory times an asking price overstates a data company because it treats future datasets as certain and free, and may count records the company does not control.

Why do a dataset's value and its owner's value diverge?

Because they price different things. Dataset valuation prices one asset: the cash that specific records can earn under a specific license, after the costs and charges that belong to those records. Company valuation prices everything the owner controls, including the people and process that produce the next dataset, minus the overhead and risk of running a business. A dataset can be worth a sliver of its owner, most of it, or more than all of it.

The decision here is the one an investor or strategic buyer faces when a pitch slides between the two numbers: the size of an archive offered as support for a company price, or a company price offered as proof that its archive is valuable. A useful answer keeps two ledgers. The asset ledger needs executed license terms or a buyer's acceptance test, the rights documents behind the records and the cost of preparing them. The company ledger needs evidence that the work repeats: deliveries accepted more than once, overhead that is known, exposures that are named.

This article builds the bridge between those ledgers line by line, fills it with illustrative value units (U) so every step can be checked, and runs the same archive through four kinds of owner. A value unit is an editorial device, not a currency. No figure on this page describes SourceX or any real company.

Tax and accounting already split the asset from the business

They do, and their categories line up with the split an investor needs. The U.S. Treasury regulation on amortizing intangibles, 26 CFR 1.197-2 (read on eCFR on 2026-10-28), treats goodwill, going concern value, workforce in place and information base as separate kinds of intangible. Goodwill is "the value of a trade or business attributable to the expectancy of continued customer patronage." An information base covers "business books and records, operating systems" and other information bases, and the regulation lists "data files, and accounting or inventory control systems" among the items to which part of an acquired business's price can be attributed. Workforce in place is a different entry again: "the composition of a workforce (for example, the experience, education, or training of a workforce)" and its terms of employment.

The IRS instructions for Form 8594, the asset acquisition statement (November 2021 revision), show which category takes the remainder. Under the residual method they describe, the price is allocated through the asset classes in order, and goodwill and going concern value, Class VII, receive whatever is left once the identifiable assets, including the Class VI intangibles, have been allocated (paraphrased). A dataset is an identifiable asset. The premium a buyer pays for a business that keeps producing datasets lands in the residual.

Financial reporting draws the same line around people. IFRS 3, the international standard on business combinations, describes an assembled workforce as an existing collection of employees that lets an acquirer keep operating the business it bought, and says that because it is not an identifiable asset, "any value attributed to it is subsumed into goodwill." Whether data reaches a balance sheet at all is a separate question, answered in this site's explainer on company data and the balance sheet. The point here is narrower: the team that turns records into licensed datasets is valued with the business, not with the records.

Worksheet part 1: what is left once the dataset pays its helpers?

A dataset's own value is the cash its records can earn under one normalized grant, less the records the rights review removes, the preparation and delivery that this dataset alone needs, and a charge for the team, systems and working capital that help earn the cash. What survives those four lines belongs to the asset and to nothing else.

Normalize the rights first, because the same records are worth different amounts under different grants. Fix the comparison at one grant, here an exclusive license for AI training, and price it for a named buyer purpose. Then strip out what the grant cannot carry: customer text a contract keeps out, personal data with no lawful route, anything already licensed for AI training. Usable scope, preparation and buyer purpose are judged against that single grant, which is the only footing on which two datasets, or two deals, can be compared.

The fourth line comes from valuation practice. IVS 210, the intangible asset standard of the International Valuation Standards Council, describes an excess earnings method that values an intangible from the cash flows left after removing the share earned by other assets, and its contributory asset charges cover working capital, fixed assets and an assembled workforce among other contributors, while ruling out a charge on goodwill as a whole or on going concern value (paraphrased from the standard). For a dataset, the charge stops the records from claiming cash the team earned. That cash is not lost; it is counted once, in the company layer.

In the worked example the archive's gross cash potential under the normalized grant is 100 U. Records carrying 20 percent of it fail the rights review, preparation and delivery cost 15 U, and the contributory charge is 8 U, leaving a dataset value of 57 U. The dataset half of the worksheet, with the evidence each line needs (Illustrative):

Worksheet part 1: what is left once the dataset pays its helpers?
LineWhat it measuresEvidence that settles itIllustrative U
1. Gross cash potentialCash the records can earn under one grant (exclusive AI training) for a named buyer purposeExecuted terms for comparable rights or a buyer's acceptance test; never an asking price100
2. Rights haircutShare of that potential sitting in records the rights review removesThe review log: each excluded record set, its reason and the document that decided it-20 (20% of line 1)
3. Preparation and deliveryExport, de-identification, documentation and delivery this dataset alone requiresTime records, vendor quotes, the processing route-15
4. Contributory chargeReturn owed to the team, systems and working capital that help earn the cashA schedule of contributory assets and the return each requires-8
Dataset valueWhat the asset is worth on its own: line 1 less lines 2 to 4The four rows above, each documented57

Worksheet part 2: engine, overhead and corporate risk

A company is worth the datasets it already holds rights in, plus the expected value of the datasets its engine will source and close, minus overhead no single dataset carries and a haircut for corporate risk, plus any other net assets. Only the first term is dataset value. Everything after it is execution, and execution is what the residual in a purchase price pays for.

The engine is where most data company valuations stand or fall. In the illustration the company expects six more datasets, each worth 40 U at the dataset layer, and judges that half will close, so the engine is worth 120 U. That probability needs a track record behind it: deliveries accepted more than once, by unrelated buyers, at a cost close to the first. Without one, the closing probability is a guess and the engine is a story.

Overhead is the cost of being a company: leadership, finance, legal, systems and selling time that no dataset carries. Here it runs 20 U a year over the same four-year horizon, 80 U in total, undiscounted to keep the arithmetic visible. Corporate risk is a 20 percent haircut on the engine for concentration, financing and key-person exposure, 24 U; it applies to the engine only, because the held dataset is already valued after its own rights haircut. Add 10 U of other net assets and the company comes to 83 U, about 1.5 times the one dataset it holds.

Change one input and the order flips. Cut the closing probability from one in two to one in ten, as can happen when a company runs short of cash or loses the people who source its datasets, and the engine falls to 24 U, the risk haircut to 4.8 U and the company to 6.2 U. The archive is still worth 57 U, about 9.2 times the company that holds it. This site's guide to valuing data assets in liquidation covers what happens when that gap reaches a sale. The company half of the worksheet, side by side for both cases (Illustrative):

Worksheet part 2: engine, overhead and corporate risk
LineWhat it measuresEvidence that settles itData producer (U)Distressed producer (U)
Held dataset valueDataset value of the archive the company already holds rights inWorksheet part 1 and the signed grant5757
5. EngineFuture datasets x net value each x closing probabilityAccepted repeat deliveries by unrelated buyers and their cost+120 (6 x 40 x 0.5)+24 (6 x 40 x 0.1)
6. OverheadCorporate cost no dataset carries, over the same four yearsBudgets against actuals-80 (20 a year)-80 (20 a year)
7. Corporate risk20% haircut on the engine for concentration, financing and key-person exposureLargest buyer's and supplier's shares; months of runway-24-4.8
8. Other net assetsCash, contracts and property outside the dataThe balance sheet+10+10
Company valueHeld dataset value plus lines 5 to 8All rows above836.2

Same archive, four owners: what the gap tells an investor

The ratio of company value to dataset value tells an investor what it is actually paying for. Run the same 57 U archive through four owners and the ratio ranges from about 18.5 to about 0.1, and each position calls for a different kind of diligence.

An operating business that licenses a by-product archive once is valued on its core business; in the illustration the archive is 5.4 percent of a 1,057 U company, and it adds one-off cash rather than a new multiple. A data producer is valued mostly on its engine. A distressed producer is an asset sale waiting to happen. An on-request licensor holds rights only in datasets that have already closed, so the records its suppliers have not signed over belong in nobody's valuation of the licensor. The four cases, with what each calls on an investor to underwrite (Illustrative):

Same archive, four owners: what the gap tells an investor
Owner (Illustrative)Dataset valueCompany valueCompany / datasetWhat to underwrite
Archive owner: an operating business licensing a by-product archive once57 U1,057 U (core business 1,000 U)About 18.5The core business; treat the license as one-off cash
Data producer: collects and licenses datasets as its business57 U83 UAbout 1.5The engine: repeat acceptances and their cost
Distressed producer: the same archive, closing probability cut to 0.157 U6.2 UAbout 0.1The asset and whether its rights transfer to a buyer
On-request licensor: sources records only after a buyer asksOnly datasets already closedClosed datasets plus engine, less overhead and riskNot computedThe process; never add suppliers' unsigned records
Two stacked columns for a data producer and a distressed producer holding the same 57-unit dataset: held dataset, engine and other assets stack above zero, overhead and risk below, with company values of 83 and 6.2 marked against a dashed line at 57.
Illustrative value units: the same 57-unit archive inside a data producer and a distressed producer.

Can inventory times an asking price stand in for company value?

No. The shortcut fails three ways at once: asking prices are not prices anyone paid, the multiplication counts future datasets as if they were already held and free to produce, and it often includes records the company does not control.

Start with the price. An asking price or a published range records what a seller hopes for, and data licensing terms are usually confidential, so the comparison set is thin and tilted toward the most optimistic quotes. Line 1 of the worksheet accepts executed terms for comparable rights or a buyer's acceptance test, and nothing else.

Then the multiplication. Treat the producer's six future datasets as already held, at full value, with no closing probability and no overhead, and the archive plus pipeline sums to 297 U, about 3.6 times the 83 U the bridge supports. The difference is not a matter of opinion: it is the closing probability, the overhead and the risk haircut that the inventory method silently sets to certainty and zero.

Last, control. A company that sources records only when a buyer asks cannot count a supplier's archive as its own inventory before the supplier signs, and a holding company cannot count records whose rights would not pass to it. Value that belongs to someone else is not discounted in the bridge; it is excluded.

What could make the bridge wrong

Six things can move value between the two ledgers or out of both, and each has a check an investor can run.

  • Rights that do not travel. IFRS 3's guidance treats a customer list as separable unless confidentiality terms or other agreements stop the entity from selling, leasing or otherwise exchanging customer information (paraphrased). A dataset that is valuable to its current owner can be worth far less inside an acquirer if a license cannot be assigned, a change of control ends it, or privacy promises limit transfer. Check assignment and change-of-control terms dataset by dataset.
  • Allocation is judgment. Contributory charges and overhead are allocations, and modest changes move value between the dataset and the company. Ask for the bridge under a second allocation and see how far the split moves.
  • Small samples behind the engine. A closing probability drawn from two or three deals is close to a guess. Treat it as a scenario range, not a point.
  • Buyer concentration. When a few buyers fund most licensing, the engine and the risk line fall together, so a single lost buyer cuts more than its own share. Ask for the largest buyer's share each quarter.
  • Substitutes. If synthetic or public data starts passing a buyer's acceptance test for a category, line 1 falls for every future dataset in it, and the engine with it.
  • The model is simplified. It ignores discounting, taxes and timing, and it treats overhead as flat. A real valuation needs all three; the bridge is a map of where value sits, not a price.

Questions for any data company's valuation, SourceX included

Ask any data company to split its value into the two ledgers, using its own documents, and apply the same questions to SourceX.

SourceX's request-data page says datasets are sourced on request, not held in stock. Before a deal closes, the records belong to the company supplying them and sit outside SourceX's asset ledger. Under its signed agreements, SourceX owns the deidentified dataset built from those records, subject to the supplier's underlying intellectual property rights, and may use, sublicense and distribute it as the agreement allows. On this page's terms it is an on-request licensor: rights in datasets that have closed, plus a process, less overhead and risk.

SourceX publishes no figures for any of those lines, and its investors page says it shares company and commercial metrics with investors directly. Its current raise is limited to accredited investors under Rule 506(c), and the investors page is where to ask for materials. Questions that work for any company in this market:

  • Which datasets do you hold rights in today, under which signed grants, and what is each worth at the dataset layer after exclusions, preparation and contributory charges?
  • What share of your valuation sits in the engine, and how many deliveries accepted by unrelated buyers support the closing probability?
  • How do you allocate overhead and contributory charges across datasets, and how far does the split move under a different allocation?
  • Which of your rights would not survive an assignment, a change of control or a sale of the company?
  • Do any figures in your materials count records that suppliers have not yet signed over?
  • Which of your figures rest on asking prices or indicative ranges rather than executed terms?

Methods note: the bridge, its inputs and what it leaves out

The bridge, both worksheet tables and the four-owner comparison were built for this article on 2026-10-28 by a short Python script (xs0215_bridge.py, kept with its output in the writer's working folder). It uses exact fractions, rounds each printed figure half up once at output to one decimal place, and uses no randomness. Every input is an editorial example in value units.

Rules as the script applies them. Rights haircut = 20 percent of gross cash potential. Dataset value = gross cash potential less the rights haircut, preparation and delivery (15 U) and the contributory charge (8 U). Engine = number of future datasets (6) x net value each (40 U) x closing probability (0.5, or 0.1 in the distressed case). Overhead = 20 U a year x 4 years, undiscounted. Corporate risk = 20 percent of the engine. Company value = held dataset value + engine - overhead - corporate risk + other net assets (10 U). The archive owner's company value is a core business value of 1,000 U plus the dataset value. The naive inventory sum is the held dataset plus six future datasets at 40 U each, with no probability, overhead or risk. Ratios and percentages divide the exact values (company by dataset, dataset by company for the 9.2 figure, naive sum by company value for 3.6) before the single rounding.

Public sources, each checked on 2026-10-28 through a search restricted to its publisher's domain: 26 CFR 1.197-2 on eCFR, the IRS instructions for Form 8594, IFRS 3 on the IFRS Foundation's site and IVS 210 as hosted by the IVSC; passages without quotation marks are paraphrases. SourceX's request-data, how-it-works and investors pages were fetched directly. Not measured: the value of any real dataset or company, actual contributory charge rates, overhead levels or closing probabilities, how often rights fail to transfer in acquisitions, and any SourceX figure.

Frequently asked questions

Is a data company worth the sum of its datasets?

Rarely. The sum of held datasets is only the asset ledger; the company also carries an engine that may produce more datasets, overhead no dataset pays for and corporate risk. In the illustration a producer holding one 57-unit dataset is worth 83 units, while the same company in distress is worth 6.2.

Can a dataset be worth more than the company that owns it?

Yes. When the engine stalls and overhead keeps running, the company can be worth less than the records it holds, which is why distressed sales price data as an asset. The records keep their value only if the rights to them can pass to a buyer.

Does licensing one dataset change an operating company's valuation?

It adds the dataset's net cash, not a new multiple, unless licensing becomes a repeatable line of business. A buyer of the company will also read the license terms, so exclusivity, assignment and deletion clauses matter as much as the amount.

Why charge a dataset for the team that prepares it?

Because otherwise the same cash is counted twice: once in the dataset and again in the value of the team and process that earned it. The contributory charge keeps the dataset's value to what the records themselves earn and leaves the team's contribution in the company layer.

Does SourceX publish dataset or company valuations?

No. Its public pages give no prices, ranges or company figures, and its investors page says company and commercial metrics are shared with investors directly. Any valuation question about SourceX itself goes through that page.

Sources

  • 26 CFR 1.197-2 treats goodwill, going concern value, workforce in place and information base as separate intangibles; an information base includes business books and records and data files. Source
  • Form 8594 instructions: the residual method allocates consideration through asset classes in order, with goodwill and going concern value (Class VII) taking what remains after Class VI intangibles. Source
  • IFRS 3: value attributed to an assembled workforce is subsumed into goodwill; a customer list is not separable where confidentiality terms prevent selling or exchanging customer information. Source
  • IVS 210: the excess earnings method values an intangible from cash flows after removing the share attributable to contributory assets such as working capital, fixed assets and an assembled workforce; no charge on goodwill as a whole. Source
  • SourceX's request-data page says datasets are sourced on request, not held in stock. Source
  • SourceX's how-it-works page: under the signed agreement SourceX owns the deidentified dataset, subject to the supplier's underlying IP rights. Source
  • SourceX's investors page says it shares company and commercial metrics with investors directly. Source

Request SourceX investor materials

SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.

See if your company qualifies