Skip to content

Leadership and readiness

Most data business concentration risk never shows in a revenue table

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

Data business concentration risk is the chance that one party or one market decides a company's results, and a revenue share captures only the buyer side of it. Measure five lenses, each on its own denominator: independent buying decisions, source companies supplying records, data categories, processing or distribution routes, and the parties whose permission the licenses rest on. Then look for a party sitting in several lenses, which a revenue table never shows.

Value drivers

Data business concentration

This page covers most data business concentration risk never shows in a revenue table.Illustrative

Key takeaways

  • A revenue percentage measures one lens; IFRS 8 asks only about customers at 10% or more of revenue, and lets the customer stay unnamed.
  • A pre-revenue data company has no revenue share to report, yet it can be fully concentrated in supply, rights or a single processing route.
  • Report three numbers per lens on its own denominator: the largest party's share, the effective number of parties (10,000 divided by HHI) and the months a replacement would take.
  • Filings show the pattern: FICO's three bureau partners hold the data, sell most of its scores and were 20% of revenue in fiscal 2017; four licensors supplied about 72% of Spotify's label-delivered audio streams in 2025.
  • In this memo's illustrative register, the buyer lens counts 8.3 effective parties while every other lens counts fewer than three, and one stacked party puts 35.6% of annual revenue at risk from an 8% revenue share.

What does a revenue percentage miss about data business concentration risk?

It misses every dependency that never sends the company an invoice. A revenue share counts who pays. Data business concentration risk also sits with whoever supplies the records, whichever category carries the demand, whatever route every delivery passes through and whoever granted the permission the licenses rest on. A company can show ten balanced buyers and still stop shipping when one of those other parties changes its mind.

The committee question is therefore narrower than whether a company is concentrated. It is which single party or market, if it changed terms or walked away, would cut deliverable revenue for longer than the company can fund, and whether documents show how long a replacement would take. That needs a comparable number for each kind of dependency, checked against contracts and delivery logs, not a pie chart of customers.

Data licensing businesses are unusually exposed because they are chains. Source companies hand over records, a processor prepares them, rights trace back to whoever authorized the use, and buyers pay at the far end. A weak link anywhere in that chain is a concentration of the whole business, whichever link the revenue table happens to show.

The disclosure investors usually get measures one lens

The standard disclosure is a customer threshold, so it reports one lens and stops. IFRS 8, paragraph 34, says: "If revenues from transactions with a single external customer amount to 10 per cent or more of an entity's revenues, the entity shall disclose that fact, the total amount of revenues from each such customer, and the identity of the segment or segments reporting the revenues." The same paragraph adds that the entity "need not disclose the identity of a major customer".

Two things follow for diligence. A filing will show customers at or above the line, often unnamed, and nothing about suppliers, channels or rights holders. And the paragraph contains an instinct worth extending: "a group of entities known to a reporting entity to be under common control shall be considered a single customer." Count independent decisions, not legal entities. Two subsidiaries spending from one parent budget are one buyer, and so are buyers whose purchases all wait on the same event.

For a pre-revenue data startup the customer test is empty by construction. There is no revenue to divide, so a memo that waits for a revenue table records no concentration at the very stage when one source company, one rights holder or one processor may carry the entire plan. Pre-revenue diligence has to start from the other four lenses.

Five lenses, five denominators: the concentration register

Give each kind of concentration the denominator that matches what could be lost, then report three numbers in every row so the lenses compare: the largest party's share, the effective number of parties, and the months a replacement would take. The effective number is 10,000 divided by the Herfindahl-Hirschman Index (HHI), the sum of squared percentage shares; ten equal parties give 10, one party gives 1.

For scale, the 2023 Merger Guidelines published by the Justice Department and the FTC describe HHI as the sum of squared market shares, reaching 10,000 when one firm holds the market, and treat an HHI above 1,800 as highly concentrated. That line was drawn for antitrust markets, not for one company's exposure, so it is a reference point here and not a limit. An HHI of 1,800 equals about 5.6 effective parties.

The register below is the original asset of this memo. Each row names the measure, the reading that should worry a committee, the documents that prove the number and the signal to keep watching:

Five lenses, five denominators: the concentration register
LensMeasure and denominatorRed flagEvidence to requestSignal to monitor
BuyerShare of contracted value and open pipeline by independent buying decision; one parent or one budget counts onceOne decision unit could stall more contracted value than the runway coversSigned orders and pipeline by counterparty, with parent company and budget owner namedScope changes, renewal notices and a buyer's announcements about sourcing data itself
SupplyShare of accepted records, and of next-quarter deliveries, by source companyOne source company behind most deliveries with no qualified second sourceSupply agreements, delivery logs by source and the list of alternates already qualifiedDelivery slips by source and supplier agreement end dates
CategoryShare of revenue and pipeline by data category, with the demand driver behind eachMost revenue in one category whose buyers all react to the same eventRevenue by category over several periods and the written requirements behind each orderRequests in new categories and demand shifts after major model releases
Channel or intermediaryShare of deliveries through each processing route, marketplace, reseller or payment collectorEvery delivery passes through one route the company does not controlProcessor and platform agreements, termination clauses and a tested alternate routeCapacity, pricing and terms changes at the route
RightsShare of licensed records by the party whose permission they rest on: the supplier, a parent, a platform's export terms or a customer contractMost records rest on terms one outside party can change aloneA chain-of-title file per dataset naming who granted what, under which documentTerms updates by that party and rights challenges against similar data
Stacked partyEach party's share in every lens, side by sideOne party appears in two or more lensesA party-by-lens matrix built from the documents aboveAny change at that party, reviewed across all lenses at once

Four filings where the revenue table understates the dependency

Public companies outside AI data show each hidden lens plainly, which makes them good calibration cases. None of the four below sells AI training data; they show how a dependency surfaces in documents when the revenue table looks calm.

FICO is the clearest stacked case. Its Form 10-K for fiscal 2024 says "The majority of our scores are marketed and sold through consumer reporting agencies" and describes its scoring algorithms as running on credit data collected and maintained by Experian, TransUnion and Equifax. The latest revenue share for those three that this memo could verify comes from the fiscal 2017 Form 10-K: agreements with them "collectively accounted for 20%, 19% and 16% of our total revenues" in fiscal 2017, 2016 and 2015. A revenue lens shows three counterparties at about a fifth of revenue; the supply and channel lenses show the same three firms under the product and most of its distribution.

Spotify shows rights concentration behind a diverse customer base. Its Form 20-F for 2025 says content licensed from Universal Music Group, Sony Music Entertainment, Warner Music Group and Merlin accounted for approximately 72% of streams of audio content delivered by record labels that year, and warns that "even one entity, or a small number of entities working together, may unilaterally affect our access to music and other content."

Duolingo shows the single-intermediary lens. Its Form 10-K for 2024 describes relying mainly on a small number of third-party platforms, chiefly the Apple App Store and Google Play, both to distribute its apps and to collect payment (paraphrased from the filing). Verisk shows supply overlapping with buyers: its Form 10-K for 2025 carries a risk factor titled "We could lose our access to data from external sources" and describes relying on data supplied by its own customers and by other sources, under agreements that are often short-term, with some data suppliers also competing with it (paraphrased). The table sets the four side by side:

Four filings where the revenue table understates the dependency
Company and filingWhat a revenue table showsLens the filing revealsDated evidence
FICO, Forms 10-K for fiscal 2017 and 2024Three bureau counterparties at a combined 20% of revenue (fiscal 2017)Supply and channel: the same three bureaus hold the data and sell most scoresFiscal 2024: majority of scores sold through consumer reporting agencies
Spotify, Form 20-F for 2025Millions of subscribers and advertisersRights: four licensors behind most label-delivered streamsAbout 72% of streams of audio content delivered by record labels in 2025
Duolingo, Form 10-K for 2024Millions of individual learnersChannel and payment collection through two app storesRisk factors on reliance on Apple App Store and Google Play (paraphrase)
Verisk, Form 10-K for 2025A broad customer baseSupply: data from customers and other sources, often under short-term agreementsRisk factor: "We could lose our access to data from external sources"

What losing the largest party costs in each lens (illustrative)

In this memo's illustrative register, losing the largest buyer costs 9.0% of a year's revenue, while losing the largest source company costs 41.3%; the revenue lens is the smallest exposure of the five. The register describes no real company. Every share, replacement time and recovery assumption is an editable input chosen to show the mechanics.

Annual revenue lost is the largest share in a lens, times the share that cannot be recovered, times the replacement months divided by 12. The category row models demand for the largest category halving for a year rather than a party leaving, and the rights row assumes half the affected records can be re-papered or replaced within the year. Each result was computed with exact fractions and rounded once.

The same register gives the effective number of parties per lens: 8.3 buyers, 2.7 source companies, 2.2 rights holders, 1.9 categories and 1.5 processing routes. Only the buyer lens clears the 5.6-party line that an HHI of 1,800 implies. Change any cell to fit a real company; the method matters more than these numbers:

What losing the largest party costs in each lens (illustrative)
Lens (illustrative)Largest shareEffective partiesReplacement assumptionAnnual revenue lost
Buyer18% of contracted value8.36 months, nothing recovered meanwhile9.0%
Supply55% of accepted records2.79 months to qualify a new source company41.3%
Category70% of revenue1.9Demand in that category halves for 12 months35.0%
Channel or intermediary80% of deliveries1.53 months to move to another processing route20.0%
Rights60% of licensed records2.2Half the records re-papered or replaced within 12 months30.0%
Stacked party8% of revenue plus the largest rights sliceNot one lensIts own purchases plus the rights loss, overlap removed35.6%
A dot plot of effective independent parties per concentration lens in an illustrative data company: buyers 8.3, supply 2.7, rights 2.2, category 1.9 and channel or intermediary 1.5, with a dashed line at 5.6 parties, the equivalent of an HHI of 1,800.
Effective parties equal 10,000 divided by HHI. Illustrative register, not measured data; the 1,800 line comes from the 2023 Merger Guidelines.

One party, several lenses: how a small account becomes the largest exposure

A party that appears in two lenses can carry more risk than the largest buyer even when its revenue share is small. In the register, a software platform buys 8% of revenue and its export terms also govern the 60% rights slice. If it withdraws both, the loss is 8 plus 30, less the 2.4 points where the rights loss falls on its own purchases: 35.6% of annual revenue from an account the revenue table ranks seventh.

The public cases follow the same shape. FICO's bureaus sit in the supply, channel and revenue lenses at once; Verisk's contributing customers sit in supply and buyers. When one party occupies several lenses, a single decision there moves all of them together, which is exactly the correlated risk a forecast built from independent buyers leaves out.

Finding stacked parties takes one matrix. List every party above a chosen share in any lens, write its share in every other lens beside it, and ask what that party would gain by changing terms. A platform that also buys data, a supplier that also resells or a processor that also competes deserves a scenario of its own.

Is concentration a warning or a sign of fit?

Often it is both. Early concentration usually means a company found one buyer, source or category worth serving deeply, and a first large licensor can validate a dataset nobody else would test. The risk lies in replacement time and contract protection, not in the share itself.

At a glance

Is concentration a warning or a sign of fit

The strongest alternative explanation for a calm forecast is that the lenses are independent, so losses would arrive one at a time and be absorbed. Test it directly: ask for the forecast rerun with the largest party in each lens removed, then with the stacked party removed. If the company cannot produce those runs, the forecast is counting contingent revenue as if each dependency were separate.

Decision rules for the memo follow from that test:

  • Pass the lens when the contract protects the exposure for longer than the replacement time and a second party is already qualified.
  • Wait when replacement time is unmeasured; ask for a dated substitute pilot or a signed alternate route before the next tranche.
  • Discount when one party can change terms alone, as Spotify says its largest licensors could, and the replacement time exceeds the runway.
  • Decline or restructure when a stacked party could remove more than the cash runway covers in a single decision.

What could make this register mislead a committee

The register is a frame, and frames mislead in predictable ways. Name the risks before trusting any row:

  • Wrong denominator. Records, revenue and contract value give different shares; a lens measured on the convenient one can look diversified.
  • Hidden groups. Affiliates, shared budgets and a common parent turn several counterparties into one decision, which is why IFRS 8 treats entities under common control as a single customer.
  • Guessed replacement times. Months to replace are estimates unless a substitute was actually qualified; ask for the dated pilot.
  • Lenses that move together. A category shock can hit buyers and source companies at the same moment, so summing lens losses can understate the worst case.
  • Invisible rights concentration. Chain-of-title gaps often surface only when someone challenges the data, long after the share was calculated.
  • Demand that shrinks regardless. Synthetic data, buyers sourcing in-house or falling prices can cut a category's value even when no single party leaves.
  • Key people outside the lenses. A rights reviewer or engineer who alone knows how a source system exports its records is a concentration the five lenses do not show; record it separately.
  • A borrowed threshold. The 1,800 HHI line comes from antitrust review of markets and is not a calibrated risk limit for a company.

Lens by lens: what to ask SourceX or any data licensor

SourceX sits in the middle of all five lenses by design, so it is a fair place to apply the register. It finds the companies, checks their rights, writes the contracts, prepares the records and pays the suppliers before datasets reach AI developers. Buyers ask for data through a request form and datasets are sourced to order from U.S. businesses rather than kept in inventory, so the supply lens is rebuilt with each request.

Its published terms also name the other lenses. Suppliers approve scope and terms and sign before any data is delivered, so each dataset's rights rest on that supplier's signed agreement, and de-identification is done by SourceX itself or through its processing partner, which makes the processing route a lens worth examining. SourceX has not published concentration figures for any lens; it shares company and commercial metrics with investors directly.

Investors weighing its current raise, which is limited to accredited investors under Rule 506(c), can request materials at sourcex.si/investors and put the questions below in writing. They apply equally to any data licensing company:

Checklist · 6 items

  1. For each lens, what is the largest party's share, on which denominator and as of which date?
  2. Which parties appear in more than one lens, and what would each lose by changing terms?
  3. How many processing routes can carry a delivery today, and how long would a switch take?
  4. For the largest source company and the largest rights holder, which document grants the permission, and can that party revoke or change it alone?
  5. After grouping by parent and budget, how many independent buying decisions sit behind contracted value and pipeline?
  6. What does the forecast show with the largest party in each lens removed?

0 of 6 ticked. Tick each item as you go.

Methods note: how the register and its figures were built

The concentration register, the stacked-party test and the illustrative figures are editorial frameworks built for this memo on 2026-10-27. Shares are whole percentages that sum to 100 within each lens; HHI is the sum of squared percentage shares; effective parties equal 10,000 divided by HHI; annual revenue lost equals the largest share times the unrecoverable share times replacement months over 12. Arithmetic used exact fractions, with each result rounded once, half up, to one decimal. No figure in the register describes SourceX or any other real company.

Public evidence comes from the IFRS Foundation's text of IFRS 8, the 2023 Merger Guidelines, and SEC filings by FICO (Forms 10-K for fiscal 2017 and 2024), Spotify (Form 20-F for 2025), Duolingo (Form 10-K for 2024) and Verisk (Form 10-K for 2025). Each was checked on its publisher's domain through domain-limited search excerpts on 2026-10-27; passages marked as paraphrased were not quoted because the excerpt was not verbatim.

Not measured: actual concentration in any lens at SourceX or other AI data companies, real replacement times, how strongly the lenses move together, and the newest FICO revenue share for the three bureaus, which this memo could not confirm beyond fiscal 2017.

Frequently asked questions

What counts as concentration risk in a data licensing business?

Any single party or market whose departure or change of terms would cut deliverable revenue for longer than the company can fund. That includes buyers, the source companies that supply records, the categories that carry demand, the processing or distribution route and the parties whose permission the licenses rest on. A revenue share captures only the first.

How should a pre-revenue data startup show concentration risk?

With the lenses that exist before revenue: supply by source company, rights by the party granting permission, processing routes and pipeline by independent buying decision. Each needs its own denominator and a date. A pre-revenue company can be fully concentrated in one supplier or one rights holder while having no revenue table at all.

What is the effective number of parties, and why use it?

It is 10,000 divided by the HHI, the sum of squared percentage shares. Ten equal parties give 10 and one party gives 1, so the number reads as how many equal-sized parties the real spread is worth. It lets an investor compare lenses measured on different denominators.

Do two subsidiaries of the same parent count as two buyers?

For concentration purposes, treat them as one when they buy under common control or from one budget. IFRS 8 already counts entities known to be under common control as a single customer for its disclosure. Extending that logic to shared budgets and shared buying triggers gives a truer count of independent decisions.

Is a highly concentrated data company a reason to pass?

Not by itself. Concentration early can reflect a strong fit with one buyer or source. It becomes a reason to discount or restructure when the replacement time exceeds the runway, when one party can change terms alone, or when a single party sits in several lenses at once.

Sources

  • IFRS 8 paragraph 34 requires disclosure when one external customer provides 10 per cent or more of revenues, need not name the customer, and treats entities under common control as a single customer. Source
  • The 2023 Merger Guidelines describe HHI as the sum of squared market shares, reaching 10,000 with one firm, and treat an HHI above 1,800 as highly concentrated. Source
  • FICO's fiscal 2024 Form 10-K says the majority of its scores are marketed and sold through consumer reporting agencies and that its algorithms run on credit data maintained by Experian, TransUnion and Equifax. Source
  • FICO's fiscal 2017 Form 10-K says revenues from agreements with Experian, TransUnion and Equifax collectively accounted for 20%, 19% and 16% of total revenues in fiscal 2017, 2016 and 2015. Source
  • Spotify's 2025 Form 20-F says content from four licensors accounted for approximately 72% of streams of audio content delivered by record labels in 2025, and that one entity or a few acting together may unilaterally affect its access to content. Source
  • Duolingo's 2024 Form 10-K describes relying primarily on a limited number of third-party platforms, chiefly the Apple App Store and Google Play, to distribute its apps and collect payments. Source
  • Verisk's 2025 Form 10-K carries the risk factor 'We could lose our access to data from external sources', describing data from customers and other sources under short-term agreements, some with competing suppliers. 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