Skip to content

Deal economics

AI data licensing buyer acquisition cost: how repeat orders flatter CAC

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

AI data licensing buyer acquisition cost should be measured two ways: the full cost of winning a buyer's first paid order, lost pursuits included, and the far smaller cost of winning each repeat order. Blending them flatters efficiency: in this article's illustrative book, blended cost per order falls 60 percent while new-buyer cost rises by half. Judge repeat buyers by orders to payback and by concentration, not by a blended CAC.

Value drivers

AI data licensing buyer

This page covers AI data licensing buyer acquisition cost: how repeat orders flatter CAC.Illustrative

Key takeaways

  • Report new-buyer acquisition cost, the cost to win a repeat order and blended cost per order separately; any one of them hides the other two.
  • In the illustrative model, a new buyer costs 120 points to win and pays back on its fifth order when each repeat order costs 25 points to win.
  • If every repeat order needs a new scope review and acceptance test, payback in the same model stretches to the sixteenth order.
  • A rising repeat share can cut blended cost per order by 60 percent while new-buyer cost climbs by half, so a falling blend proves nothing alone.
  • Repeat buying concentrates revenue: Reddit ties substantially all licensing contract value to two partners, and Innodata's largest customer was about 58 percent of its 2025 revenue.
  • Capitalized selling costs can push reported acquisition cost below cash spent, especially when renewals are expected; ask for both.

What should buyer acquisition cost measure in AI data licensing?

Buyer acquisition cost is everything spent to win a buyer's first paid order, divided by the buyers won, with failed pursuits counted. In AI data licensing it has to sit beside a second number, the cost to win each repeat order, because the two can differ several-fold and a blend of them describes neither.

Data licensing differs from software subscriptions at exactly this point. A subscription renews unless someone cancels it; a data license usually ends with its scope, and the next order has to be won. Repeat buyers lower acquisition cost per order only while they keep ordering, so the evidence is order history by cohort, not a retention percentage borrowed from software.

This article builds the split, follows one illustrative buyer through six orders, shows how a rising repeat share flatters the blend and lists the records an investor should request. Every value is illustrative, in index points where one order's revenue is 100, and no SourceX figures are reported. The five metrics, with the denominators that keep them honest:

What should buyer acquisition cost measure in AI data licensing?
MetricNumeratorDenominatorCommon distortion
New-buyer acquisition costAll selling, solution and onboarding cost spent on a cohort of pursued buyers, lost pursuits includedBuyers in that cohort who placed a first paid orderCounting only hours logged against buyers who signed
Cost to win a repeat orderProposal, scoping, rights scope review and order paperwork for repeat proposals, lost ones includedRepeat orders wonBooking repeat selling effort to account management or delivery
Blended acquisition cost per orderBoth numerators aboveAll orders wonFalls as the repeat share rises, even while new buyers get dearer
Buyer payback in ordersNew-buyer cost not yet recoveredContribution per repeat order net of the cost to win itMeasured before rework, overhead and the cost of slow collection
Repeat order rate by cohortBuyers in a cohort placing a second paid order within 12 monthsBuyers in the cohort whose first order is at least 12 months oldMixing young cohorts with mature ones

Why repeat orders are cheaper to win, and where the savings stop

A repeat order reuses the most expensive parts of the first sale: the master agreement, the vendor security review, the acceptance test and the working relationship. It does not reuse the rights analysis for a new use, the supply for a new dataset or the buyer's patience on price.

At a glance

Why repeat orders are cheaper to win, and where the savings stop

Those limits are why a repeat order is cheaper but never free, and why a company that claims near-zero repeat acquisition cost has probably booked the work somewhere else. What carries over and what starts again:

  • Carries over: the signed master agreement, so a repeat order needs an order form rather than a negotiation.
  • Carries over: vendor onboarding and security review, unless the buyer's policy or the data's sensitivity changes.
  • Carries over in part: the acceptance test, when the new order covers the same task; a new task needs a new test.
  • Starts again: rights scope review whenever the buyer wants a new use, since the original license may not cover it.
  • Starts again: supply, because each new dataset still has to be sourced, qualified and prepared, a cost that belongs in a separate supplier acquisition ledger.
  • Starts again: price, because a buyer that has bought before knows the market and negotiates from that knowledge.

The buyer lifetime waterfall: one buyer, six orders

In the illustrative model, one buyer returns 5 points after six orders. The first order cannot carry its 120-point acquisition cost, and the next five recover it slowly because each still has to be won. That small final number is the honest reading of what repeat buyers do to data licensing CAC.

The inputs are editable. Each order brings 100 points of revenue and 45 points of contribution after the data owner's share, preparation and delivery. New-buyer cost assumes 10 buyers pursued and 3 won, at 30 points per pursuit plus 20 points of onboarding per buyer won, or 120 per new buyer. Repeat cost assumes three repeat proposals at 15 points each win two orders, plus 2.5 points of order paperwork, or 25 per repeat order. Line by line for one buyer:

The buyer lifetime waterfall: one buyer, six orders
LineFormulaIllustrative pointsRunning totalEvidence to request
First-order contributionRevenue less data owner's share, preparation and delivery+4545Deal-level contribution ledger
New-buyer acquisition costCohort pursuit and onboarding cost divided by buyers won-120-75Pursuit log by cohort, lost pursuits included
Contribution on orders 2 to 65 orders times 45+225150Order history for each buyer
Cost to win orders 2 to 65 orders times 25-12525Repeat proposal log, lost proposals included
Rework, overhead and slow collectionAllowance across the six orders-205Rework log, allocation rule, days to cash
Lifetime contribution after six ordersSum of the lines above55Reconciliation to the general ledger

How many repeat orders pay back a new buyer?

With repeat orders costing 25 points to win, the buyer pays back on its fifth order; if each repeat needs a new scope review, it takes sixteen. Cumulative contribution after n orders equals 45, minus 120, plus the repeat margin for each of the n minus 1 later orders, where the repeat margin is contribution per repeat order less the cost to win it: 20 points in the base case.

At a glance

How many repeat orders pay back a new buyer

That repeat margin is the number to watch. Modest changes on either side of it move payback by many orders, which is why a claim that repeat buyers are cheap should arrive with the cost per won repeat order, lost proposals included. Four cases from the same model:

  • Base: repeat orders cost 25 points to win, and cumulative contribution turns positive on the fifth order, at plus 5 points.
  • New use each time: every repeat order needs its own scope review and acceptance test, lifting its cost to 40 points, and payback moves to the sixteenth order.
  • Price concession: the buyer negotiates terms that cut contribution per repeat order from 45 to 37 points, and payback moves to the eighth order.
  • Dearer new buyers: new-buyer cost rises to 160 points, as in year 3 of the blended book, and payback moves to the seventh order.
Paired bars of one illustrative buyer's cumulative contribution after orders 1 to 6: with repeat orders costing 25 points the balance climbs from minus 75 to plus 5 at order 5 and plus 25 at order 6; at 40 points it only reaches minus 50 by order 6.
Orders to payback for one buyer under two repeat-order costs. Illustrative model in index points.

How blended CAC can fall while new buyers get dearer

Blended acquisition cost per order can fall 60 percent while each new buyer costs 50 percent more, simply because repeat orders make up more of the mix. A falling blend is therefore not, on its own, evidence of better selling.

In the illustrative book, repeat orders cost 25 points each throughout. As the repeat share rises from 20 to 90 percent over four years, blended cost per order falls from 101 to 40.5 points, while the cost of each new buyer climbs from 120 to 180. A company reporting only the blend would show improvement in the very years its ability to win new buyers deteriorated. Year by year:

How blended CAC can fall while new buyers get dearer
Illustrative yearOrders from new buyersOrders from repeat buyersNew-buyer acquisition costBlended cost per order
Year 182120101
Year 25514082.5
Year 32816052
Year 41918040.5

Repeat buyers or a concentrated book? What filings show

Repeat buying and concentration are two readings of the same order book, and public filings show how far concentration can go in AI data businesses.

Reddit's S-1, filed in February 2024, disclosed data licensing arrangements with an original expected duration of more than one year and an aggregate transaction price of $203.0 million, of which it expected to recognize at least $66.4 million in 2024. Those are multi-year commitments rather than repeat orders: the buyer decided once. Its FY2025 Form 10-K adds that, to date, "substantially all of the contract value associated with our licensing revenue is derived from two of our partners."

Concentration also shows up at Innodata. Its FY2025 Form 10-K reports that one customer in its DDS segment generated approximately 58 percent of total revenue in 2025 and 48 percent in 2024, and its 10-Q for the six months ended June 30, 2026 reports one customer at about 46 percent and another at about 26 percent. A book whose repeat orders come from one or two buyers has a low blended acquisition cost and a single point of failure.

Buyers with announced budgets change the arithmetic in both directions. In a 2026 blog post, micro1 announced it would commit $1 billion over the next 12 months to acquiring and licensing enterprise operational data, which it will use to build reinforcement learning environments; it is a stated plan, not reported spending. A buyer running a standing program can make repeat orders cheap to win for suppliers it has already qualified, and the same program can contract with data-holding companies directly, so ask how much repeat volume depends on buyers able to bypass the company.

Capitalized commissions: when reported acquisition cost drifts from cash

Reported sales expense can understate the cash spent to win buyers, because accounting rules let companies capitalize part of it. Ask for acquisition cost on a cash basis alongside whatever the income statement shows.

Under Subtopic 340-40, added by ASU 2014-09, an entity recognizes as an asset the incremental costs of obtaining a contract that it expects to recover, and may expense them as incurred only when the amortization period is one year or less. A FASB staff paper prepared for the revenue Transition Resource Group (Memo 23) adds that when a renewal carries no commission, or one not commensurate with the first, the amortization period can extend beyond the initial contract if renewals are anticipated, and the one-year expedient then stops applying. Expected repeat business can therefore lengthen the period over which acquisition cost reaches the income statement.

Disclosure guidance points the same way. The SEC's January 2020 interpretive release on key performance indicators in MD&A (Release 33-10751) reminds companies that metrics they present remain subject to existing MD&A requirements and may call for further material information. For an acquisition-cost metric, that means the definition, the cohort rule, and whether lost pursuits and capitalized costs are inside it.

Contribution per order deserves the same scrutiny. A repeat buyer that pays slowly, asks for rework or absorbs a large share of overhead can look cheap to win and still return little, which is why the waterfall carries its own line for rework, overhead and slow collection.

What could make the repeat-buyer thesis wrong

The case that repeat buyers make data licensing cheaper can fail in five ways, and most have a public marker an investor can track.

At a glance

What could make the repeat-buyer thesis wrong

  • Concentration: Reddit attributes substantially all of its licensing contract value to two partners, and Innodata's largest customer generated about 58 percent of its 2025 revenue. A book of repeat orders from one or two buyers is a renewal risk, not a moat.
  • Disintermediation: buyers with their own acquisition programs, such as the one micro1 announced in 2026, can contract with data holders directly, and repeat volume that depends on them can leave.
  • New supply every time: if each repeat order needs a new dataset, supplier acquisition cost, not buyer acquisition cost, dominates the economics.
  • Price concessions: experienced buyers negotiate, and in the illustrative model a cut of 8 points of contribution per repeat order adds three orders to payback.
  • Thin evidence: few companies disclose the cost of a won repeat order, so most public comparisons rely on blended figures that the definitions above show to be unreliable.

Questions on buyer acquisition for any data licensing company, SourceX included

Ask for the three numbers separately and for the order history behind them; a single CAC figure cannot answer what repeat buyers are worth.

SourceX has not published buyer acquisition or repeat-order metrics, and it should face these questions like any company in the market. It is raising capital from accredited investors under Rule 506(c); its investor materials, including company and commercial metrics, are shared directly on request through its investors page. Questions to ask:

  • What is new-buyer acquisition cost by cohort, with lost pursuits and onboarding included, on a cash basis?
  • What does it cost to win a repeat order, counting repeat proposals that did not convert?
  • What share of orders and of contribution comes from repeat buyers, and from the top two buyers?
  • How many orders does a typical buyer need to pay back its acquisition cost, and how has that changed across cohorts?
  • How many repeat orders required a new rights scope review or a new acceptance test?
  • Which acquisition costs are capitalized, over what period, and how does that period treat expected repeat business?

Methods note: the buyer acquisition model and its gaps

The model follows one illustrative buyer and one illustrative order book in index points, where one order's revenue is 100. Contribution per order is 45 points. New-buyer cost is 10 pursuits at 30 points plus 3 onboardings at 20 points, divided by 3 buyers won, or 120. The cost to win a repeat order is 3 proposals at 15 points divided by 2 orders won, plus 2.5 points of paperwork, or 25. Cumulative contribution after n orders is 20 times n minus 95 in the base case. The blended book holds repeat cost at 25 and raises new-buyer cost from 120 to 180 over four years. Rework, overhead and slow collection are a single 20-point allowance across six orders.

Public sources give context, not inputs: Reddit's S-1 and FY2025 Form 10-K, Innodata's FY2025 Form 10-K and June 2026 Form 10-Q, micro1's 2026 announcement of a stated plan, ASU 2014-09 with a FASB Transition Resource Group staff paper on contract costs, and the SEC's 2020 release on key performance indicators. Sources were checked on October 26, 2026.

Not measured: real acquisition costs or repeat rates at any company, how often repeat orders need new rights review, typical price concessions, or how much repeat volume comes from buyers with their own acquisition programs. No SourceX data is used.

Frequently asked questions

What is buyer acquisition cost in AI data licensing?

It is the full cost of winning a buyer's first paid order, including failed pursuits and onboarding such as security review and master agreement negotiation, divided by the buyers a cohort produced. Report it next to the cost of winning each repeat order.

Why can blended CAC mislead investors in data licensing?

Repeat orders are far cheaper to win than new buyers, so a rising repeat share pulls the blend down even when new buyers cost more. In the illustrative book, blended cost per order falls 60 percent while new-buyer cost rises by half.

How many repeat orders does a new data buyer need to pay back?

It depends on the repeat margin, contribution per repeat order minus the cost to win it. In the illustrative model, a 20-point repeat margin pays back a 120-point acquisition on the fifth order, while a 5-point margin takes sixteen orders.

Are multi-year data licenses the same as repeat orders?

No. A multi-year license is one buying decision with revenue spread over its term, like the data licensing arrangements of more than one year that Reddit disclosed in its 2024 S-1. A repeat order is a fresh decision the buyer can decline, so it carries its own cost to win.

Do capitalized commissions change reported acquisition cost?

They can. Under Subtopic 340-40, incremental costs of obtaining a contract that are expected to be recovered are capitalized and amortized, and anticipated renewals can lengthen that period, so a year's income statement may show less acquisition cost than was spent in cash.

Sources

  • Reddit's S-1 (February 2024) disclosed data licensing arrangements of more than one year with an aggregate transaction price of $203.0 million, at least $66.4 million of it expected to be recognized in 2024. Source
  • Reddit's FY2025 Form 10-K states that substantially all of the contract value associated with its licensing revenue is derived from two partners. Source
  • Innodata's FY2025 Form 10-K reports one DDS customer at approximately 58 percent of total revenue in 2025 and 48 percent in 2024. Source
  • Innodata's 10-Q for the six months ended June 30, 2026 reports one customer at about 46 percent and another at about 26 percent of total revenue. Source
  • micro1 announced in 2026 a stated plan to spend $1 billion over 12 months acquiring and licensing enterprise operational data for reinforcement learning environments. Source
  • Subtopic 340-40 capitalizes incremental costs of obtaining a contract that are expected to be recovered, with a practical expedient to expense them when the amortization period is one year or less. Source
  • A FASB Transition Resource Group staff paper notes the amortization period can extend beyond the initial contract when renewals are anticipated and renewal commissions are not commensurate, ending the one-year expedient. Source
  • SEC Release 33-10751 (January 2020) gives interpretive guidance on key performance indicators and metrics in MD&A, reminding companies of existing MD&A requirements and the need for further material information. 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