Deal economics
Where a data licensing SaaS valuation comparison breaks, metric by metric
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
A data licensing SaaS valuation comparison breaks wherever a SaaS metric assumes the same customer renews the same service. Annualizing one quarter, counting expiring contracts as renewed, computing net retention on project revenue that SaaS filers exclude, and comparing gross margins built on different principal choices all overstate data licensing revenue quality. Restate each metric on contracted, repeat and per-dataset terms before choosing comparables.
Value drivers
Data licensing SaaS
Key takeaways
- Annualizing one quarter has the clearest public warning: one Wiley AI project, annualized, comes to 1.7 times the AI license revenue Wiley reported for the full year.
- ARR assumes expiring contracts renew on existing terms; a delivered data license usually has no terms to renew.
- SaaS filers such as CrowdStrike keep services revenue out of net retention; a licensor of delivered datasets that computes retention on all revenue puts exactly that kind of revenue in.
- Gross margin swings with whether the supplier's share is booked gross or net, so compare contribution per dataset instead.
- In the illustrative downside model, a correlated pause by the top buyer cuts next year to 41 of 100 index units against 110 under a SaaS lens.
Which SaaS metrics mislead in a data licensing SaaS valuation comparison?
Every metric that assumes the same customer will renew the same service: annualized run-rate and ARR, net and gross retention, and the payback and lifetime-value figures built on them. Gross margin misleads for a different reason, because data intermediaries choose between gross and net revenue presentation in a way a software vendor selling its own product does not face. Restate all four on contract evidence before picking comparables or a multiple.
The decision at stake is which valuation method a data licensing business has earned. Software revenue multiples price a stream of renewals. A licensor whose revenue is mostly delivered licenses has a stream of separate sales, each needing new supply, rights work and preparation. Using the first method on the second business is not wrong by a fixed amount; the error depends on the contract mix, which is why the mix has to be measured first.
This memo takes the four metric families through public filings, then sets out a restated metric set, an illustrative downside model, the risks that cut against this critique and the questions to put to any company, SourceX included. It quotes no valuation multiple, because the inputs need fixing before anyone multiplies them.
What one AI licensing project does to a run-rate
It can push the run-rate far past what the full year delivers. John Wiley and Sons provides a clean public example. Its quarterly report for the three months ended July 31, 2024 says it executed a $21 million content rights project for training generative AI large language models, of which $17 million was recognized in that quarter. Its annual report for the year ended April 30, 2025, which contains that quarter, puts AI license revenue at $40 million for the year, against $23 million the year before.
Apply the SaaS habit of annualizing the latest quarter to that single project and the answer is $68 million, 1.7 times the AI license revenue Wiley reported for the whole year, a total that also included other AI licensing revenue. The arithmetic is trivial; the trap is that the quarter looked like momentum. Wiley's own report took the project out: excluding it, adjusted revenue grew 2 percent at constant currency.
The procedural lesson is simple. Ask for revenue by quarter with each project shown separately, use trailing 12-month revenue rather than an annualized quarter, and treat growth that rests on one large delivery as a one-time event until a second buyer repeats it.
Why ARR's renewal assumption does not transfer
Because ARR counts renewals before they happen, and a delivered data license usually has nothing to renew. CrowdStrike's annual report for fiscal 2026 calculates ARR as the annualized value of customer subscription contracts on the measurement date, assuming any contract that expires during the next 12 months is renewed on its existing terms. A lapsed subscription stays in ARR while the company is in discussion about a renewal or a new subscription, or until the customer says it is not renewing.
That definition suits a subscription business with years of renewal history. Carried over to data licensing, it imports two assumptions the contracts may not support: that an expiring agreement has existing terms to renew on, and that renewal costs the seller little. A one-time delivery has neither. Even a refresh agreement renews only if the supplier still grants the rights and new records get prepared, so its renewal carries supply cost that a software renewal does not.
Public filers such as CrowdStrike print their metric definitions, and the SEC issued interpretive guidance on key performance indicators and metrics in MD&A in January 2020 (Release 33-10751). Private company decks carry no such discipline unless an investor asks. Replace ARR with two numbers: contracted forward revenue, meaning amounts due under signed agreements over the next 12 months by buyer and expiry date, and renewal-option revenue, meaning agreements with a priced option or automatic renewal, each shown with the supplier rights it depends on. Everything else is pipeline, however often the buyer has bought before.
Net revenue retention assumes a base that delivered licenses do not create
Net retention assumes a recurring base to retain, and delivered-license revenue is the kind SaaS filers keep out of that base. CrowdStrike's fiscal 2026 report computes dollar-based net retention by taking ARR from customers who were subscription customers 12 months earlier and dividing their current ARR by that earlier figure, so expansion adds, contraction and churn subtract, new customers are left out, and revenue from its incident response and proactive services is excluded.
Those services are project work that does not renew on a schedule, so the software company removes them. A data licensor that computes net retention on all of its revenue is putting into the calculation exactly what CrowdStrike takes out. A buyer who returns to license a different dataset is a repeat customer, but the second sale needed its own supply and preparation; scoring it as retention makes a project business look like a subscription.
Measure buyer behavior directly. Repeat purchase rate: of the buyers who signed in a cohort period, the share who signed another agreement within 12 months, with the denominator stated. Repeat revenue at stable delivery cost: second-sale revenue counted only where the cost to deliver did not rise, because a repeat that needs a new supplier is a new sale. Even in software the metric moves: CrowdStrike's report for the quarter ended October 31, 2025 cautions that large customer contracts and incentives can make the rate fluctuate.
Gross margin depends on who books the supplier's share
A data licensor's gross margin is partly an accounting judgment about whether it acts as principal, so it cannot sit beside software margins without restating. Shutterstock's annual report for 2025 says it recognizes revenue gross of contributor royalties because it is the principal in the transaction, and that royalties are the largest component of its operating expenses, reported within cost of revenue.
Any business that pays the owners of the data it licenses faces the same judgment. Presented gross, the supplier's share sits in cost of revenue: revenue is larger and gross margin lower. Presented net, revenue shrinks to the amount the seller keeps and margin looks like software. The economics are identical, yet a revenue multiple applied to the gross figure prices the supplier's money as the company's. Preparation labor adds a second judgment: de-identification, record linking and review can land in cost of revenue or in operating expense, and that choice moves gross margin without changing a single deal.
Ask for both presentations and for contribution per accepted dataset: the buyer's price less the supplier's share, preparation, rework and delivery. That figure compares across companies whatever their accounting policy, and it shows whether the second sale of a cleared dataset really costs less than the first.
The restated metric set: what to measure instead
Restate each SaaS metric before it enters a comparison. Each row names what the metric assumes, how it misleads when the revenue comes from data licenses, the substitute measure and the document that proves it; the methods note at the end gives sources and limits:
| SaaS metric | What it assumes | How it misleads for data licensing | Measure instead | Evidence to request |
|---|---|---|---|---|
| Annualized run-rate | The latest quarter repeats | One delivery inflates a whole year; one Wiley project annualized came to 1.7 times its reported full-year AI license revenue | Trailing 12-month revenue with each project shown separately | Quarterly revenue by agreement |
| ARR | Expiring contracts renew on existing terms | Delivered licenses have no terms to renew, and refreshes need new supply | Contracted forward revenue by buyer and expiry date | Signed agreements and payment schedules |
| Net revenue retention | Revenue comes from a renewing base | New datasets sold to old buyers count as retention, though SaaS filers exclude comparable project revenue | Repeat purchase rate by cohort, and repeat revenue at stable delivery cost | Buyer-level agreement history with delivery costs |
| Gross retention and logo churn | Losing a customer is a failure | A completed one-time license ends by design, so churn reads as failure where none occurred | Share of buyers with an open agreement or an active request | Dated agreement status by buyer |
| Gross margin | Cost of revenue means hosting and support | Gross or net booking of the supplier's share, and where preparation labor sits, move it | Contribution per accepted dataset after supplier share, preparation and rework | Revenue under both presentations and a per-dataset cost ledger |
| CAC payback and lifetime value | Customers keep paying after acquisition | Lifetime value built on assumed renewals multiplies the ARR error | Observed contribution from first and second agreements per buyer | Cohort table of agreements and contribution by buyer |
| Revenue growth rate | Growth compounds from a recurring base | A single large delivery reads as momentum | Growth without the largest agreement, and the count of buyers above a size threshold | Revenue by agreement, largest first |
Downside sensitivity: how far a SaaS lens can overshoot
On an illustrative mix, a SaaS lens overstates next year's revenue by 23.5 to 69 index points out of 100, and the gap is driven by the share of one-time revenue and the repeat rate, not by growth. Every input is an assumption chosen to show the mechanics; none describes a real company or SourceX.
The model company has trailing revenue of 100 index units: 60 from delivered one-time licenses, 25 from multi-year term access with more than a year to run, and 15 from annual refresh feeds. The SaaS lens multiplies all 100 by an assumed net retention of 110 percent, giving 110. The contract lens adds the 25 contracted units, refresh revenue times a renewal rate, and one-time revenue times a repeat rate, meaning the share of last year's one-time spend that the same buyers spend again.
The correlated row matters most to an investment committee. If the largest buyer accounts for half the one-time revenue and two-thirds of the refresh revenue, its pause removes both at once, and next year falls to 41 even at base-case rates for everyone else. Forecasts that treat repeat rate, renewal and concentration as independent inputs hide exactly this case. Values come from a script using exact fractions:
| Illustrative scenario | Repeat rate on one-time revenue | Refresh renewal rate | Contract-lens next year (index) | Gap to SaaS lens (points) |
|---|---|---|---|---|
| Base case | 40% | 80% | 61.0 | 49.0 |
| Repeat buying slows | 20% | 80% | 49.0 | 61.0 |
| Refresh feeds churn | 40% | 50% | 56.5 | 53.5 |
| Top buyer pauses (correlated) | 40% of the remaining buyers | 80% of the remaining feeds | 41.0 | 69.0 |
| Strong repeat buying | 80% | 90% | 86.5 | 23.5 |
Risk register: what could make this critique wrong
Six developments would weaken the case against SaaS logic or change what the restated metrics show. Track each with its signpost:
- Data licensing turns recurring. If buyers move from one-time deliveries to multi-year refresh agreements with renewal history across several buyers, ARR and retention become fair measures for that slice. Signpost: the share of revenue under term agreements with renewal options, buyer by buyer.
- Concentration overwhelms the metric debate. When one buyer drives most revenue, every metric, SaaS or restated, mostly measures that relationship, as the correlated scenario shows. Signpost: the largest buyer's revenue share and the date its agreement ends.
- Rights and regulation raise supply costs. Wiley's annual report for fiscal 2025 warns that AI technologies could significantly disrupt the markets in which it operates and subject it to increased competition, legal and regulatory risks. A licensor whose supply rights narrow pays more per dataset, which no revenue metric shows. Signpost: changes in supplier consent terms and new obligations on training data.
- Synthetic or public data passes buyers' acceptance tests. Repeat rates would then fall for reasons unrelated to the seller's execution. Signpost: whether buyers' acceptance criteria still require real business records.
- Pricing stays opaque. Data licensing prices are rarely published, so any comparable set may rest on a handful of disclosed deals. Signpost: the number of priced, dated agreements behind a comparison.
- Accounting judgments shift the inputs. Principal versus agent and point-in-time versus over-time recognition are decided contract by contract with auditors, and a change can move revenue and margin without moving cash. Signpost: revenue recognition policy notes and any restatement.
Questions for any data licensing company, SourceX included
Put the same diligence questions to every company in this market, SourceX included, and ask for documents rather than summaries.
SourceX licenses business data directly from companies for AI training and evaluation. Its role spans the chain: it finds suitable records, checks the rights, writes the contract, prepares the data, delivers it and pays the company that licensed it, and each deal's signed agreement sets the term and exclusivity. Its revenue, contract mix and cost figures are not public, and this memo states none; SourceX shares those numbers with investors itself. The current raise is a Rule 506(c) offering limited to accredited investors, who can request materials at sourcex.si/investors. The questions:
- What share of trailing revenue came from the largest single agreement, and what was growth without it?
- Which agreements create contracted revenue for the next 12 months, and when does each expire?
- Of the buyers in each cohort, how many signed a second agreement, and did the second cost less to deliver than the first?
- Is revenue presented gross or net of supplier payments, and what would gross margin be under the other presentation?
- Where is preparation labor booked, and what is contribution per accepted dataset after rework?
- Which ARR or retention figures appear in your materials, and what exact definitions produce them?
Methods note: sources, model and limits
The restated metric set, the downside model and the risk register were built for this memo on 2026-10-27 as editorial tools. Public evidence comes from CrowdStrike's fiscal 2026 annual report and its report for the quarter ended October 31, 2025, Wiley's report for the quarter ended July 31, 2024 and its fiscal 2025 annual report, Shutterstock's 2025 annual report, and the SEC's 2020 interpretive release on key performance indicators and metrics, each verified on sec.gov through search excerpts because direct fetches were blocked.
The run-rate figure is the writer's arithmetic: $17 million times four is $68 million, which divided by Wiley's reported $40 million gives 1.7. The downside model uses exact fractions, rounds printed values once, and labels every input illustrative; both scripts are kept with the draft's working files for recount.
Not measured: valuation multiples for any company, renewal or repeat rates at any data licensor, contribution per dataset anywhere, and how often data licensing revenue is presented gross. Those need contract-level data that no filing provides, so the model shows direction and mechanism, not magnitude.
Frequently asked questions
Should a data licensing startup report ARR in its pitch materials?
Only for revenue that is contracted and renewable, and only with the definition attached. Everything else is better shown as trailing 12-month revenue and contracted forward revenue by buyer. If a deck shows ARR, ask whether expiring agreements are assumed to renew and whether one-time deliveries are annualized.
Can you value a data licensing company on an ARR multiple?
Only the part of revenue that behaves like ARR, meaning contracted, renewable access with renewal evidence across several buyers. Applying a software ARR multiple to delivered licenses prices renewals the contracts do not provide. Restate revenue by contract type first; this article quotes no multiple.
What should replace net revenue retention when diligencing a data licensor?
A buyer-level repeat purchase rate with a stated cohort and denominator, plus repeat revenue counted only where delivery cost did not rise. Net revenue retention as software companies define it assumes a renewing base and treats a new dataset sold to an old buyer as expansion.
Why do gross margins of data companies vary so much?
Partly because of accounting judgments. A company acting as principal books revenue gross of the supplier's share and puts that share in cost of revenue, as Shutterstock does with contributor royalties, while a net presentation shows smaller revenue and a higher margin. Where preparation labor is booked moves margin too.
How should an investor annualize lumpy data licensing revenue?
Use trailing 12-month revenue with each large agreement shown separately, and report growth with the largest agreement excluded. Annualizing a quarter that holds one big delivery can overshoot badly: one Wiley AI project, annualized, gives 1.7 times the AI license revenue Wiley reported for that full year.
Sources
- Wiley executed a $21 million AI content rights project, $17 million of it recognized in the quarter ended July 31, 2024; excluding it, adjusted revenue grew 2 percent at constant currency. Source
- Wiley reported AI license revenue of $40 million for the year ended April 30, 2025 versus $23 million the prior year, and lists AI disruption among its risk factors. Source
- CrowdStrike defines ARR with expiring contracts assumed renewed on existing terms, and computes net retention on prior-period subscription customers with services revenue excluded. Source
- CrowdStrike cautions that large customer contracts and incentives can make its net retention rate fluctuate. Source
- Shutterstock recognizes revenue gross of contributor royalties as principal and reports royalties, its largest operating expense, within cost of revenue. Source
- The SEC issued interpretive guidance on key performance indicators and metrics in MD&A in January 2020. Source
Request SourceX investor materials
SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.