Skip to content

AI data market

Put AI data supplier eligibility in the TAM, not in a conversion rate

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

AI data supplier eligibility belongs inside the TAM as named lines, each measured on the companies left by the line before, not folded into one conversion rate. Borrowed rates compound: Eurostat's 2025 ERP and CRM rates for large EU enterprises multiply to 4.38 times the all-enterprise product. Then divide any capture claim by the eligible share; with illustrative filters, signing 2% of companies means 9.7% to 95.3% of eligible ones.

Value drivers

AI data supplier eligibility

This page covers put AI data supplier eligibility in the TAM, not in a conversion rate.Illustrative

Key takeaways

  • A single conversion rate hides which exclusions were applied, on which population, and whether any were applied at all; an exclusion ledger names each one.
  • Rates move with the population: in Eurostat's 2025 survey, 88.71% of large EU enterprises used ERP software against 46.45% of all enterprises with 10 or more employees, 1.91 times as many.
  • Borrowed rates compound: the all-enterprise ERP and CRM rates multiply to 13.24%, the large-enterprise rates to 58.04%, 4.38 times higher.
  • Some eligibility rules add companies an official table leaves out, such as firms that peaked above 50 staff and shrank, or wound-down companies, and need their own add-back lines.
  • The implied-penetration test divides a capture claim by the eligible share: with this page's illustrative filters, a 2% claim needs 9.7% to 95.3% of eligible companies.

Where does AI data supplier eligibility belong in a TAM model?

Inside it, as named lines between the company count and any revenue arithmetic, each measured on the companies that survive the line before. A supplier-side TAM that starts from every business in a region and applies one optimistic conversion rate at the end has not removed anyone; it has hidden its exclusions in a number nobody can check.

The decision this page supports is narrow. When a company in AI data licensing presents an addressable market of suppliers, can an investor see who is ruled out, by which rule and on what evidence? A useful answer needs three things: the eligibility rules written as the company applies them, a rate for each rule measured on the right population, and a test that shows whether the claimed share of the market is plausible once the rules have run.

Eligibility and conversion are different questions. Eligibility asks whether a company could license records at all under the rules a buyer and an intermediary apply; conversion asks whether an eligible company chooses to sign. Merging the two is how a 2% conversion rate ends up carrying a size rule, a rights review, a language test and a sales cycle at once, with no way to tell which of them did the work.

Three exclusions a single conversion rate hides

A blended rate hides three distinct errors, and each needs a different fix. None is visible in a deck that shows a total market and one percentage; all three become visible, and arguable, once each rule has its own row:

  • Population mismatch: a filter rate measured on all businesses is applied to an eligible frame of larger ones, or the reverse. System-of-record use, AI use and record depth all rise with company size, so a borrowed rate is wrong by a factor the section on published rates measures.
  • Silent compounding: when two or more borrowed rates are multiplied, their errors multiply as well. A model can be off by a factor of four before anyone questions an assumption.
  • Rules that add as well as remove: a rule written on peak headcount, or one that accepts acquired and wound-down companies, admits firms that a register of active businesses does not count. A conversion rate can only shrink a number, so it cannot represent an add-back.

The exclusion ledger: rules, rates and add-backs in order

The ledger lists every eligibility line in the order the model applies it, states what kind of line it is, and names the population its value must be measured on and the evidence that sets it. Rules define the frame and are applied to counts. Rates are applied to whatever frame the rules leave. Add-backs enter as separate positive lines. Conversion stays out of the ledger and is tested on its own.

The rows use SourceX's published eligibility as a labeled application, because a real rule set shows where each kind of line comes from. Its how-it-works page says SourceX focuses on companies with 50+ full-time employees at peak and that qualifying companies list at least 5 systems. Its FAQ calls U.S. companies the priority, accepts Canada and the EU, considers companies in nine further markets when at least 90% of their data is in English, says operating, acquired and wound-down companies can all qualify, and adds that a smaller company with truly one-of-a-kind data can still be a strong fit. The ledger, line by line:

The exclusion ledger: rules, rates and add-backs in order
Ledger lineKindMeasured onEvidence that sets itWhere it enters
Region and language: U.S., Canada, EU; other listed markets at 90% EnglishRule, partly checked on the datasetBusiness counts by countryOfficial registers for the frame; English share measured on the records offeredDefines the frame, before any rate
Size: 50 or more full-time employees at peak, contractors excludedRule with a known mismatchSize-class counts of current headcountOfficial size-class tables; peak size only from company recordsFrame; current counts miss firms that have shrunk
Peak-size, acquired and wound-down companies outside current countsAdd-backCompanies absent from an active-business registerCompany histories, acquisition and closure recordsSeparate positive line after the frame
One-of-a-kind data below the size lineAdd-back, case by caseFirms under 50 staffAssessed deal by deal; no public rateSmall separate line, never folded into the size rule
Systems: at least 5 systems listedRateThe 50-or-more frame, not all firmsERP and CRM use by size class as proxies; the count of 5 needs assessment dataFirst rate applied to the frame
Rights clear, and records not licensed for AI training beforeRatesCompanies that pass the systems lineSampled rights reviews and supplier confirmationsAfter systems
Fit with documented buyer demandRateCompanies that pass rightsDated buyer requests matched to record familiesLast eligibility line
Willingness to signConversion, not eligibilityEligible companies onlyA funnel with stated denominatorsOutside the ledger, tested separately

How far does a filter rate move with the population it is measured on?

Far enough to swamp most assumptions. Eurostat's 2025 survey of EU enterprises with 10 or more employees publishes the same technology-use rates by size class, which turns the error of borrowing an all-enterprise rate from a hypothesis into a measurement. Eurostat defines small enterprises as 10 to 49 employees and self-employed persons, medium as 50 to 249 and large as 250 or more.

ERP software, the closest published proxy for a system of record, was used by 46.45% of all enterprises in the survey and by 88.71% of large ones, 1.91 times as many; for medium-sized enterprises (69.93%) the ratio is 1.51. A frame of companies with 50 or more staff, the size line in the ledger, therefore carries an ERP rate between 1.51 and 1.91 times the all-enterprise figure. CRM software shows a wider gap, 28.51% against 65.43% for large enterprises, a ratio of 2.29. AI use, the demand-side analogue, runs from 19.95% of all enterprises to 55.03% of large ones, 2.76 times. Small enterprises sit below the average on all three, at 0.85 to 0.88 times.

Multiplying two borrowed rates multiplies their errors. Treating ERP and CRM use as independent filters, the all-enterprise rates give 46.45% x 28.51% = 13.24%, while the large-enterprise rates give 88.71% x 65.43% = 58.04%, 4.38 times higher. The within-class product still assumes the two are independent, which these tables cannot test, so 4.38 is the error from the population choice alone.

The U.S. shows the same tilt from another angle. A Census Bureau working paper on the 2026 AI supplement to the Business Trends and Outlook Survey reports that 18% of firms used AI in at least one business function between November 2025 and January 2026, against 32% on an employment-weighted basis. One survey, two answers 14 points apart, depending on whether firms or workers are the unit: the same choice a TAM makes when it picks its frame. The ratios from Eurostat's published rates:

How far does a filter rate move with the population it is measured on?
Filter, EU 2025All enterprises (10+)Medium (50-249)Large (250+)Large divided by all
ERP software46.45%69.93%88.71%1.91
CRM software28.51%Not read65.43%2.29
AI technologies19.95%30.36%55.03%2.76
ERP x CRM, multiplied as independent13.24%Not computable58.04%4.38

The implied-penetration test for any capture claim

Divide the share of the whole frame a plan says it will win by the share of the frame that is eligible. The result is the share of eligible companies the plan needs, which is the number a conversion rate was hiding. Above 100% the claim is impossible; well below it, the claim is merely unproven.

Capture claims are usually small, and that is what makes them persuasive. A plan to sign 2% of companies sounds modest until the ledger shows that only 5% are eligible, at which point it needs 40% of eligible companies to sign. Each cell below is the claim divided by the eligible share, rounded half up to one decimal:

The implied-penetration test for any capture claim
Eligible share of the frameClaim of 0.5%Claim of 1%Claim of 2%Claim of 5%
40%1.3%2.5%5.0%12.5%
20%2.5%5.0%10.0%25.0%
10%5.0%10.0%20.0%50.0%
5%10.0%20.0%40.0%100.0%
2%25.0%50.0%100.0%250.0%, impossible

The eligibility workbook and its sensitivity chart

With one anchored line and three illustrative ones, the eligible share of a 50-or-more frame spans 2.10% to 20.65%, and a claim to sign 2% of that frame needs between 9.7% and 95.3% of eligible companies. Every input is labeled; none is a SourceX figure, and the frame is left as a count the reader supplies, so no market size is stated.

The systems line takes Eurostat's 2025 ERP rates for medium (69.93%) and large (88.71%) enterprises as its low and high values, with 75% as an illustrative central value between them. ERP is a proxy: a company can hold qualifying records in a help desk or a CRM without an ERP, and listing five systems is a stricter test than running one. The rights, prior-licensing and buyer-fit lines have no public anchor at all.

The arithmetic multiplies each line on the population the previous line left. Low: 69.93% x 25% x 80% x 15% = 2.10%. Central: 75% x 40% x 90% x 25% = 6.75%. High: 88.71% x 60% x 97% x 40% = 20.65%. A 1% claim then needs 47.7%, 14.8% or 4.8% of eligible companies; a 2% claim needs 95.3%, 29.6% or 9.7%. Buyer fit (2.67 times from low to high) and rights (2.40 times) move the eligible share most; the anchored systems line moves it 1.27 times and the prior-licensing line 1.21 times.

The chart plots the same relationship for any eligible share from 1% to 25%, so a reader can test a deck quickly: take the eligible share from the company's own ledger, find the claim's curve, and read how much of the eligible pool it needs. The inputs:

The eligibility workbook and its sensitivity chart
Ledger lineLowCentralHighLabel
S. Systems of record, 50-or-more frame69.93%75%88.71%Low and high anchored to Eurostat 2025 ERP use, medium and large; central illustrative
R. Rights clear for at least one record family25%40%60%Illustrative
L. Records not licensed for AI training before80%90%97%Illustrative
B. Fit with documented buyer demand15%25%40%Illustrative
Eligible share E = S x R x L x B2.10%6.75%20.65%Illustrative result
Share of eligible companies a 2% claim needs95.3%29.6%9.7%2% divided by E in the same column
Two curves showing the share of eligible companies a capture claim needs as the eligible share of the frame rises from 1% to 25%, for claims of 1% and 2%, with the region above 100% marked impossible and three markers on the 2% curve at 95.3%, 29.6% and 9.7%.
Illustrative: implied penetration equals the claim divided by the eligible share; markers use the workbook's low, central and high eligible shares of 2.10%, 6.75% and 20.65%.

Why AI spending, global data and venture rounds cannot stand in for the ledger

Because none of them counts a company that could sign. AI spending measures buyers' budgets for compute, software and staff; global data volume measures bytes stored, most of which nobody may license; venture funding measures investor appetite for the category. Each is real and large, and each sits outside the ledger: it can inform the demand side of a model or the timing of a thesis, but it cannot say how many suppliers pass a size rule, a rights review and a buyer-fit test.

Top-down numbers also flatter the implied-penetration test in a specific way. A larger frame makes any capture share look smaller without adding a single eligible company, which is why the test divides by the eligible share and never by the starting frame. Adoption figures carry the same trap from inside one survey: Eurostat counts 55.03% of large EU enterprises using AI in 2025, a statement about deployment that says nothing about whether any of them may license their records.

What could make an eligibility-filtered TAM wrong?

Six weaknesses could move the true eligible share outside any range built this way, and each pushes in a direction a reader can anticipate:

At a glance

What could make an eligibility-filtered TAM wrong

  • Correlated rates: rights, record depth and buyer fit may move together within the frame, so the true joint share can fall outside the product of independent ranges in either direction; only an end-to-end screen of sampled companies measures it.
  • Proxy drift: ERP use stands in for a five-system rule it does not measure. Where qualifying systems are mostly help desks and CRMs, the systems line is set too high or too low depending on the sector.
  • Frame mismatch: official tables count current headcount and active businesses, so peak-size and wound-down companies are missing, and an enterprise group can appear as several entries.
  • Rule changes: an intermediary can widen or narrow its rules, for example by adding markets at a 90% English threshold, which changes the frame with no change in the economy.
  • Prior licensing: SourceX's terms ask suppliers to confirm the data has not been licensed for AI training before and make licenses exclusive for AI training, so every exclusive license signed anywhere lowers the prior-licensing line for the next model.
  • Selection bias: rates measured on applicants overstate the frame's rates, because companies that apply have already screened themselves.

Questions for any company that filters its TAM, SourceX included

Ask for the ledger before the total. A company that can show each eligibility line, the population it was measured on and its denominator has done the part of market sizing that matters most when supply, not demand, is the constraint.

SourceX makes a convenient test case for these questions. Its investor page describes licensing business data directly from companies and supplying it onward to AI developers in its own name, handling sourcing, rights review, contracting, preparation, delivery and payment, and its how-it-works page names ten fit signals, from volume and historical depth to clear ownership and Acquisition Partner demand. SourceX publishes no supplier counts, eligible shares, conversion rates or market-size estimate; company and commercial metrics go to investors who request materials. Its current raise relies on Rule 506(c), so only verified accredited investors can take part, and requests start on the investor page.

Six questions test any eligibility-filtered TAM, SourceX's included:

  • Which eligibility rules do you apply, in what order, and which are rules rather than rates?
  • On which population was each rate measured, and how many companies sit in its denominator?
  • Which lines add companies an official table misses, and how large are they?
  • What eligible share does your ledger produce, and what share of eligible companies does your plan need?
  • Which rates were measured on applicants rather than on a sample drawn from the frame?
  • How do you measure willingness to sign, separately from eligibility?

Methods note: published rates, ledger arithmetic and gaps

Built on 2026-10-28. Size-class rates come from two Eurostat Statistics Explained articles, Use of artificial intelligence in enterprises (AI use) and E-business integration (the ERP and CRM size splits), both from the 2025 survey of EU enterprises with 10 or more employees and self-employed persons; size-class definitions come from Eurostat's 2026 publication on AI technologies in the EU; the U.S. comparison comes from the Census Bureau's working paper CES-WP-26-25. SourceX's rules come from its live how-it-works, FAQ and investor pages, read the same day. The Eurostat and Census figures were read as search excerpts on the publishers' own domains, so a reviewer opens each page before release.

Arithmetic: each ratio is a size-class rate divided by the all-enterprise rate; the compounded figure multiplies the ERP and CRM rates as if independent, once for all enterprises and once for large ones; the eligible share multiplies four ledger lines; implied penetration divides a claim by the eligible share. Every computation uses exact fractions and rounds once, half up: to two decimals for ratios and eligible shares, one decimal for implied penetration. The script xs0370_ledger.py prints every input and result, including the chart points.

Not measured: the medium-enterprise CRM rate, which the page read did not give; joint ERP and CRM use within any size class; any rights, prior-licensing or buyer-fit rate; the number of companies in any frame; willingness to license; and anything about SourceX's applicants, suppliers or pipeline. No market size, revenue or growth figure is stated or implied.

Frequently asked questions

What is the difference between eligibility and conversion in an AI data TAM?

Eligibility asks whether a company could license records at all under the rules a buyer and an intermediary apply, such as size, region, rights and buyer fit. Conversion asks whether an eligible company chooses to sign. Keep them apart: eligibility lines are measured on samples of the frame, conversion on a funnel with stated denominators.

Why not apply one conversion rate to the whole market?

Because a single rate cannot show which exclusions it contains or which population it was measured on, and it can only shrink a number, never add back companies a register misses. Separate lines make each assumption visible and replaceable when better evidence arrives.

How large is the error from borrowing an all-enterprise rate?

In Eurostat's 2025 data, large EU enterprises used ERP software at 1.91 times and CRM software at 2.29 times the all-enterprise rate. Multiplying the two as filters, the large-enterprise product is 4.38 times the all-enterprise product, before any other assumption is touched.

What does the implied-penetration test tell an investor?

It converts a modest-sounding capture claim into the share of eligible companies the plan actually needs, by dividing the claim by the eligible share. A result near or above 100% means the claim is implausible or impossible; a small result means the claim is unproven rather than wrong.

Does SourceX publish its eligible supplier pool?

No. SourceX publishes its eligibility rules on its how-it-works and FAQ pages but no supplier counts, eligible shares, conversion rates or market-size estimate. Investors who request materials through its investor page receive company and commercial metrics directly.

Sources

  • In 2025, 19.95% of EU enterprises with 10 or more employees used AI technologies, including 30.36% of medium and 55.03% of large enterprises. Source
  • Eurostat's Statistics Explained article E-business integration: in 2025, ERP software was used by 46.45% of EU enterprises (41.08% small, 69.93% medium, 88.71% large) and CRM by 28.51% (24.69% small, 65.43% large). Source
  • Eurostat defines small enterprises as 10-49 employees and self-employed persons, medium as 50-249 and large as 250 or more; 17.0% of small enterprises used AI in 2025. Source
  • CES-WP-26-25 reports that 18% of U.S. firms used AI in at least one business function in the November 2025 to January 2026 reference period, 32% on an employment-weighted basis. Source
  • SourceX focuses on companies with 50+ full-time employees at peak, qualifying companies list at least 5 systems, it evaluates ten fit signals, suppliers confirm the data has not been licensed for AI training before, and licenses are exclusive for AI training. Source
  • SourceX prioritizes U.S. companies, works with Canada and the EU, considers nine further markets at 90% English, accepts operating, acquired and wound-down companies, and may accept smaller companies with one-of-a-kind data. Source
  • SourceX's investor page describes licensing business data directly from companies and onward to AI developers in its own name and offers company and commercial metrics to 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