Skip to content

Deal economics

One big license, three distorted metrics: data deal annualized revenue

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

Data deal annualized revenue overstates a business whenever one large license is recognized in a single quarter. In this article's illustrative model, annualizing the delivery quarter gives 1.96 times the next twelve months' revenue, growth reads 154.5 percent instead of 18.2 percent, and the new buyer holds 53.6 percent of the quarter. Report the license on its own line, annualize only recurring revenue and test repeats over 24 months.

Value drivers

Data deal annualized revenue

This page covers one big license, three distorted metrics: data deal annualized revenue.Illustrative

Key takeaways

  • In the illustrative model, annualizing the quarter that holds one static license overstated next-year revenue 1.96 times; annualizing the delivery month overstated it 4.07 times.
  • The same license lifts growth from 18.2 to 154.5 percent, then shows a 46.4 percent fall a year later while the base grew 15.4 percent.
  • Concentration moves with it: the licensee held 53.6 percent of the delivery quarter against 15.0 percent for the largest recurring buyer.
  • Recognition terms set the shape: Tempus recognizes record licenses on delivery and subscriptions ratably, and InterDigital expected static fixed-fee licenses to land in the quarter signed.
  • SEC rules bar labeling an item non-recurring in non-GAAP measures when a similar one is likely within two years or occurred in the prior two; a 24-month repeat test borrows that discipline.

What does data deal annualized revenue measure, and when does it mislead?

It measures what a business would book in a year if its latest month or quarter repeated, and it misleads whenever that period holds a license that will not repeat. Data licenses are prone to this. A buyer may pay once for a historical archive, the revenue may be recognized in the month the records arrive, and the next quarter starts again from the recurring base.

The investor's decision is how much of a reported run-rate, growth rate and buyer mix to believe when one deal dominates a period. Comparable evidence has three parts: a metric defined the same way in every period, a baseline that leaves the deal out, and an observation window long enough to see whether a similar deal repeats. Without all three, one delivery can make a data company look as if it had more than doubled, and a year later as if it had nearly halved.

This memo works through the mechanism with public filings and an illustrative model computed by script. It states no figures for SourceX or any real company. Every model value is an index unit chosen to show how the distortion works, not how large it is anywhere in particular.

Why one data license can land in a single quarter

Because license accounting ties revenue to what the buyer receives and when. Tempus AI's annual report for 2025 treats each licensed record as a right-to-use license whose revenue is typically recognized when the data is delivered, and treats multi-year subscriptions as right-to-access licenses recognized ratably over the access period. The same contract value can produce a spike or a smooth line, depending on whether the buyer takes delivery of a fixed set or gains access over time.

Patent licensors show the same split. InterDigital's annual report for 2018, written as it adopted the new revenue standard, separates dynamic fixed-fee agreements, which carry rights to future technologies and are recognized on a straight-line basis over the term, from static agreements without those rights, for which it expected to recognize most or all revenue in the quarter the license was signed. A data archive with no refreshes is the static case; a feed of new records is the dynamic one.

Headline terms can also overstate what is committed. Tempus's 2024 annual report says its data contracts often span several years but usually let customers opt in or end early without penalty, so the company treats the enforceable contract period as shorter than the stated term. A run-rate built on the stated term counts years the buyer can walk away from.

The first diligence question follows directly: for each large agreement, what was delivered, when was revenue recognized, and what does the buyer still owe or receive after the measurement date?

The illustrative model: one archive license inside a growing base

The model company has a recurring base of licenses and feeds that grows by 3 index units a quarter across 12 buyers, the largest taking 15 percent. In month 20 a new buyer takes delivery of a static archive license worth 90 index units, recognized in full on delivery. Nothing else changes, and no second archive deal follows. Every value is illustrative and computed by script with exact fractions.

Read the dashboard row by row. The delivery quarter, year 2 quarter 3, more than doubles revenue; the next quarter falls back to the base; and the same quarter a year later shows a 46.4 percent decline while the business underneath grew 15.4 percent. The largest-buyer column jumps for one quarter only, which is why a concentration figure should always name its period:

The illustrative model: one archive license inside a growing base
Quarter (illustrative)Revenue: base plus licenseReported growth, year on yearGrowth without the licenseLargest buyer's share
Year 2, Q172 + 020.0%20.0%15.0%
Year 2, Q275 + 019.0%19.0%15.0%
Year 2, Q378 + 90154.5%18.2%53.6%
Year 2, Q481 + 017.4%17.4%15.0%
Year 3, Q184 + 016.7%16.7%15.0%
Year 3, Q287 + 016.0%16.0%15.0%
Year 3, Q390 + 0-46.4%15.4%15.0%
Stacked quarterly bars from year 2 quarter 1 to year 3 quarter 3: a recurring base rising from 72 to 90 index units, with a 90-unit license stacked on year 2 quarter 3; labels compare reported growth of 154.5 percent and then minus 46.4 percent with growth of 18.2 and 15.4 percent without the license.
Illustrative model: one 90-unit license recognized on delivery in month 20. Index units, not real figures.

Run-rate, growth and concentration: three readings of the same quarter

Each headline metric reads the license differently, and each overstates in its own way. Run-rate first. Annualizing month 20 alone gives 1,392 index units, 4.07 times the 342 the company books over the following twelve months. Annualizing the delivery quarter gives 672, or 1.96 times. Trailing twelve-month revenue of 384 overstates by a smaller 1.12 times, because it holds the license once rather than four or twelve times.

The cleanest figure is the recurring base annualized, 312, with the license reported beside it on its own line. It understates the next year slightly, at 0.91 times, because the base keeps growing. That is the safer direction to err in, and the gap is the base's own growth, which an investor can see and judge.

Growth second. Year-on-year growth in the delivery quarter reads 154.5 percent against 18.2 percent for the base. A year later the license laps and the same metric shows a 46.4 percent decline, although nothing went wrong. A company that puts the first figure in a fundraising deck has, in effect, pre-announced the second.

Concentration third. The new buyer accounts for 53.6 percent of delivery-quarter revenue and 23.4 percent of trailing twelve-month revenue, against 15.0 percent for the largest recurring buyer. Both readings are true: the first shows one quarter's dependence on one counterparty, the second spreads it across a year. Ask for the largest buyer's share with and without the license, and for the period each figure covers.

Five timing scenarios and the 24-month rule

Run the same license through five timings and one decision rule, and the rule keeps the adjusted run-rate at 0.91 to 0.92 times next-year revenue in four of the five cases. The script applies the rule at the end of each evaluation quarter: count a license in run-rate at its monthly amount times twelve if it is recognized evenly under a contract running past that date; otherwise count the trailing twelve months of that kind of license if another one was recognized with a different buyer in the 24 months before it; otherwise leave it out of run-rate and report it on its own line.

The fourth row is the edge case. When a second archive license to a different buyer lands twelve months after the first, the rule treats archive licensing as a repeating line and adds its trailing twelve months, which overstates the next year by 1.15 times if no third deal follows. The reported growth rate does worse: two one-off deals a year apart make year-on-year growth read 7.1 percent while the base grew 15.4 percent, hiding both the lumpiness and the trend.

Ratable recognition, in the second row, removes the spike altogether, which is why the contract terms behind a smooth line matter as much as the line. In every row, the evaluation quarter is the one that holds the delivery, or the second delivery for the fourth row:

Five timing scenarios and the 24-month rule
Scenario (illustrative, 90-unit license)Annualized quarter vs next 12 monthsGrowth: that quarter, then a year laterLargest buyer's share of the quarterRule treatment; adjusted run-rate vs next 12 months
Static archive delivered in month 201.96x154.5%, then -46.4%53.6%One-off; 0.91x
Same value as a 36-month feed from month 200.89x25.8%, then 17.5%14.1%Contracted; 0.92x
Delivery slips to month 231.93x147.8%, then -45.6%52.6%One-off; 0.92x
Second archive to another buyer in month 321.85x7.1%, then -43.3%50.0%Repeating line; 1.15x
Largest existing buyer takes the archive1.96x154.5%, then -46.4%60.5%One-off; 0.91x

How public filers keep one license out of the trend line

They split it out, label it, and face limits on how they adjust for it. InterDigital's annual reports have long separated recurring revenue from revenue tied to new agreements: the 2016 report presents recurring revenue beside past-sales revenue from newly signed licenses, and the 2022 report attributes changes in non-recurring revenue to agreements signed the year before. A reader of those reports sees the run-rate and the catch-up as separate lines.

SEC rules add discipline to the labels. As the SEC staff's non-GAAP interpretations restate it, Item 10(e) of Regulation S-K bars adjusting a non-GAAP performance measure to eliminate or smooth items identified as non-recurring, infrequent or unusual when a similar item is reasonably likely within two years or occurred within the prior two. The same interpretations warn that a measure which accelerates revenue recognized ratably under GAAP, as though it were earned when billed, can be misleading. Item 303 points management's discussion at known events and uncertainties that make reported results not necessarily indicative of future ones.

Private companies raising money are not bound by these disclosure rules, but the logic carries over. The 24-month window in the model's rule borrows the two-year look-back and look-ahead of Item 10(e), and refusing to annualize a delivered license mirrors the warning against accelerating ratable revenue. A data company that applies both unprompted sends investors the clearest signal it can about its own numbers.

The growth-claim evidence checklist

Apply the checklist to any growth claim from a data licensing company before comparing it with peers or applying a multiple. Each item names the evidence that settles it, and a claim that fails an item should be restated rather than discounted by feel:

At a glance

The growth-claim evidence checklist

Checklist · 9 items

  1. Revenue by agreement, by quarter, largest first, for at least eight quarters, so every spike can be traced to its source.
  2. For each agreement above a size threshold the company fixes in advance: what was delivered, the recognition basis (on delivery or over time), and amounts owed or receivable after the measurement date.
  3. Run-rate defined in writing: which revenue is annualized, from which period and with which multiplier. A run-rate built from one month deserves the most suspicion.
  4. Every growth rate shown twice, with and without the largest agreement in each period, with the year-ago quarter named.
  5. The largest buyer's share with and without the exceptional agreement, for the quarter and for the trailing twelve months.
  6. A repeat test: any license of the same kind with a different buyer in the prior 24 months, with dates. One deal is an event; two within 24 months make a line that still needs its own trailing figure.
  7. Contract terms behind smooth revenue: termination, opt-in and renewal rights that shorten the enforceable term.
  8. Cash against revenue for the large agreements, to catch upfront payments recognized later and revenue recognized before cash arrives.
  9. The lap: the company's own forecast of reported growth in the quarter one year after the exceptional deal, and the explanation it plans to give.

0 of 9 ticked. Tick each item as you go.

What could make this checklist mislead

Five things could make the restatement too harsh or too lenient. Weigh them before applying it:

Checklist · 5 items

  1. A first deal can start a real line. Archive licensing that repeats with new buyers is a business, and stripping out the first deal understates it. The 24-month test sees repeats only after they happen, so ask for evidence of a second buyer without counting it as revenue.
  2. Smooth revenue can hide weak commitments. Ratable recognition passes the rule as contracted, yet Tempus's 2024 report shows how early termination rights make the enforceable term shorter than the stated one. Read the exit rights before accepting a contracted label.
  3. The illustrative base grows steadily. Real recurring revenue churns, and a shrinking base makes the ex-deal run-rate overstate the next year instead of understating it.
  4. The two-year window is a convention borrowed from disclosure rules for public companies, not an accounting standard for private ones. A different window changes which deals count as repeating.
  5. Buyer concentration is a business risk as well as a measurement one. If a data company's demand comes from a few model developers, correctly reported recurring revenue can still vanish when one of them stops buying, and no restatement removes that exposure.

0 of 5 ticked. Tick each item as you go.

Diligence questions for any data licensor, SourceX included

Put the checklist to every company in this market, SourceX included, and ask for the documents rather than summaries of them. SourceX licenses business data directly from companies and licenses it on to AI developers under its own name, running the steps between the two: sourcing, rights review, contracting, preparation, delivery and payment. It publishes no revenue, deal or buyer figures and this memo states none; SourceX says it gives those numbers to investors directly rather than printing them.

SourceX is raising from accredited investors under Rule 506(c); materials can be requested at sourcex.si/investors. The questions below work for any data licensing company at any stage:

  • Which agreements made up the largest share of revenue in each of the last eight quarters, and how was each recognized?
  • Which run-rate figure do you use internally, and from which period is it annualized?
  • What did growth look like with the largest agreement removed from both periods?
  • Has any license of the same kind repeated with a different buyer within 24 months?
  • Which large agreements carry early termination or opt-in rights, and what is the enforceable term?
  • What will reported growth show in the quarter a year after your largest delivery?

Methods note: the model, the filings and the limits

The quarterly dashboard, the five scenarios, the 24-month rule and the checklist were built for this memo on 2026-10-28 as editorial tools. The model script (xs0114_model.py) uses exact fractions and rounds once, half up: percentages to one decimal, ratios to two. Inputs, all illustrative index units: recurring base revenue in month m equals 20 plus the number of whole quarters completed before that month, split across 12 buyers at fixed shares of 15, 12, 11, 10, 9, 8, 8, 7, 6, 5, 5 and 4 percent; one license worth 90 units, delivered in month 20 unless a scenario says otherwise. Next-twelve-month revenue counts the twelve months after the evaluation quarter and assumes no further deals. Reported growth compares a quarter with the same quarter a year earlier, and growth a year later compares the quarter four quarters on with the evaluation quarter. The largest buyer's share is the biggest single buyer's revenue in the quarter divided by the quarter's total. The adjusted run-rate is four times the quarter's recurring base plus whatever the rule adds.

Public evidence comes from Tempus AI's annual reports for 2024 and 2025, InterDigital's annual reports for 2016, 2018 and 2022, the SEC staff's non-GAAP compliance and disclosure interpretations, and Items 10 and 303 of Regulation S-K on the eCFR. Each was checked on its publisher's own domain through search excerpts on 2026-10-28, because direct fetches of sec.gov and ecfr.gov were blocked. The excerpts were paraphrases, so this memo paraphrases them and takes no figures from those documents.

Not measured: how often real data licensing companies annualize exceptional deals, the size or recognition basis of any private company's agreements, any SourceX figure, and whether 24 months is the right window for data licensing. The model shows the direction and mechanics of the distortion, not its typical size.

Frequently asked questions

What is annualized revenue for a data deal?

It is the revenue from a recent month or quarter multiplied to a full year, usually by twelve or four. For a data license recognized on delivery, that multiplies a single event as if it recurred. In this memo's illustrative model, annualizing the delivery quarter gave 1.96 times the revenue of the following twelve months.

Is trailing twelve-month revenue a safe substitute for run-rate?

It is safer but not clean. Trailing revenue counts an exceptional license once instead of four or twelve times, so it overstated next-year revenue by 1.12 times in the illustrative model rather than 1.96. Report the recurring base annualized, with the license on its own line, alongside the trailing figure.

When does a large data license count as recurring?

When an enforceable contract keeps paying or keeps being recognized after the measurement date, or when licenses of the same kind repeat with different buyers. This memo's rule uses a 24-month window, borrowed from the two-year test in SEC non-GAAP rules. Check termination rights before treating any multi-year contract as committed.

How should a data company disclose a one-time license to investors?

Show it on its own line by quarter, give every growth rate with and without it, state its recognition basis and what remains owed, and show the largest buyer's share both ways. Then say in advance what growth will look like when the license laps a year later.

Why does a one-time license distort customer concentration?

Because the licensee's revenue lands in one period. In the illustrative model, the new buyer held 53.6 percent of the delivery quarter but 23.4 percent of the trailing year, while the largest recurring buyer held 15.0 percent. A concentration figure without its period and an ex-deal version cannot be compared.

Sources

  • Tempus AI's 2025 annual report treats licensed records as right-to-use licenses recognized on delivery and multi-year subscriptions as right-to-access licenses recognized ratably. Source
  • Tempus AI's 2024 annual report says its data contracts often span multiple years but usually include opt-in or early termination rights without penalty, shortening the contract period it accounts for. Source
  • InterDigital's 2018 annual report recognizes dynamic fixed-fee licenses over the term and expects most or all revenue from static fixed-fee licenses in the quarter they are signed. Source
  • InterDigital's 2016 annual report presents recurring revenue separately from past-sales revenue under newly signed licenses. Source
  • InterDigital's 2022 annual report attributes changes in non-recurring revenue to agreements signed in the prior year. Source
  • The SEC staff's non-GAAP interpretations restate Item 10(e)'s two-year limit on labeling items non-recurring and warn that measures accelerating ratably recognized revenue can mislead. Source
  • Item 10(e) of Regulation S-K bars smoothing items labeled non-recurring, infrequent or unusual out of non-GAAP measures when similar items are reasonably likely within two years or occurred in the prior two. Source
  • Item 303 of Regulation S-K focuses management's discussion on known events and uncertainties that make reported results not necessarily indicative of future results. Source
  • SourceX's investor page says it licenses business data directly from companies and onward to AI developers in its own name, and 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