Skip to content

Deal economics

Data supplier support cost: why the help belongs in gross margin

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

Data supplier support cost is the staff time spent helping suppliers produce rights evidence, run exports, answer questions and fix deliveries. The part that scales with accepted datasets belongs in the cost of each dataset, not in overhead. In this article’s illustrative cohort, support averages 58.2 hours per accepted dataset, and doubling applicants at weaker fit raises it to 65.1 hours while accepted datasets rise only 12.5 percent.

Value drivers

Data supplier support cost

This page covers data supplier support cost: why the help belongs in gross margin.Illustrative

Key takeaways

  • Supplier support has two halves: help on datasets that reach acceptance is a production cost, help on suppliers who never do is acquisition cost.
  • Airbnb books operations and support as its own line, while DoorDash puts supply-side onboarding and support inside cost of revenue.
  • In the illustrative cohort, 39.8 percent of support hours go to suppliers who never produce an accepted dataset.
  • Doubling applicants at weaker fit raises support hours 25.9 percent for a 12.5 percent gain in accepted datasets.
  • Moving dataset-driven support into cost of revenue lowers an illustrative 70 percent gross margin to 64.0 percent.

What is data supplier support cost, and why does it belong in dataset cost?

Data supplier support cost is the staff time a data licensing company spends helping suppliers get a dataset over the line: finding evidence of rights, pulling records out of their systems, answering questions about scope and terms, and fixing problems at delivery. The part of it that rises and falls with accepted datasets is a cost of producing those datasets. Booking it as overhead flatters gross margin and hides whether the business gets cheaper to run as it grows.

Support splits two ways, and investors need both halves. Hours spent on suppliers whose datasets reach acceptance behave like cost of revenue: more datasets, more hours, unless tooling or repeat suppliers bend the curve. Hours spent on suppliers who never reach acceptance behave like acquisition cost: they rise with the number and quality of applicants, not with output. A single blended operations line mixes a production cost with a marketing cost.

This article shows how to measure the two halves: a ledger of assistance by type and driver, a cohort model of hours per accepted dataset, a test of the claim that more applicants means more supply, and the effect on gross margin of putting dataset-driven support where it belongs.

Four kinds of supplier assistance and the driver behind each

Four kinds of assistance cover most of the help a supplier needs, and each has its own driver. Evidence work grows with the number of contract families and notices a supplier must find; export help grows with source systems and failed extracts; clarifications grow with questions; delivery support grows with deliveries and the re-deliveries that follow a defect.

Costing them per driver rather than per supplier follows time-driven activity-based costing, the method Robert Kaplan and Steven Anderson set out in Harvard Business Review in 2004. It needs two estimates per resource group, the cost of each unit of time the group supplies and the time each activity takes, and it uses time equations to show how an order’s characteristics change that time, so idle capacity stays visible instead of being spread across products. The time equations below are illustrative inputs to edit, not measurements:

Four kinds of supplier assistance and the driver behind each
AssistanceTriggered byTime equation (illustrative)Scales withBelongs in
Rights evidenceAuthority, ownership and contract questions3 h per new supplier (1 h if returning) plus 2 h per contract familyContract families per datasetCost of the dataset
Export helpPulling records out of each source system2 h per source system plus 5 h per failed extractSystems and extract failuresCost of the dataset
ClarificationsQuestions on scope, fields, terms and process0.75 h per questionSupplier experienceCost of the dataset
Delivery supportTransfer, de-identification issues and acceptance defects3 h per delivery plus 6 h per re-deliveryDeliveries and defectsCost of the dataset
Help to suppliers never acceptedAny of the above for suppliers who drop out or failSame equations, stopped at the stage reachedApplicant volume and fitAcquisition cost

Where public marketplaces book the cost of helping their suppliers

Two large public marketplaces book similar supply-side help in different places, so reported gross margins in this kind of business may not compare cleanly. Airbnb reports operations and support as its own expense line, separate from cost of revenue in its statement of operations. Its 2025 annual report describes the line, in substance, as mainly personnel costs and third-party service provider charges for community support, along with customer relations costs.

DoorDash takes the other route. Its 2025 annual report puts platform costs inside cost of revenue and describes them as including merchant and Dasher onboarding, support for consumers, merchants and Dashers, and technology infrastructure, with personnel costs for local operations and support staff alongside them. Help given to the supply side of the marketplace sits above the gross margin line at one company and below it at the other.

Disclosure is about to improve for public companies, though not for private ones. The FASB’s ASU 2024-03, as clarified by ASU 2025-01 in January 2025, requires public business entities to disclose, from the first annual reporting period beginning after December 15, 2026, a table that breaks relevant expense captions into categories that include employee compensation. That will show how much labor sits inside cost of revenue at a public filer. A private data licensing company owes no such table, so an investor has to ask for the ledger directly.

The support ledger an investor should ask to see

Ask for support logged at the level of each assistance event, tied to a supplier cohort and to the dataset it served, so the hours can be split by type, driver and outcome. Totals by month cannot be split after the fact. The fields that make the split possible:

At a glance

The support ledger an investor should ask to see

  • Date, supplier identifier and cohort, meaning the quarter the supplier first applied.
  • Stage when the help was given: before qualification, between qualification and signing, between signing and delivery, or after delivery.
  • Assistance type (rights evidence, export help, clarification or delivery support) and the driver unit it served, such as a contract family, a source system or a failed extract.
  • Minutes spent, and whether the person was an employee, a contractor or a vendor, so outsourced help is not lost in a vendor line.
  • Dataset identifier and final outcome: accepted, declined at acceptance, withdrawn or still open on the measurement date.
  • A flag for help triggered by the company’s own error, such as an unclear request or a changed requirement, kept apart from help the supplier needed.

Support hours per accepted dataset, by supplier profile and cohort

In this article’s illustrative model, support costs 25.0 hours for an accepted dataset from a first-time supplier with one system, 57.0 hours for a first-time supplier with three systems, and 15.0 hours for a returning supplier on the same system. Clarifications are the largest line for every first-time profile, so in this model supplier experience moves support cost more than dataset size does.

Cohort economics change the picture. Take 100 applicants, of whom 16 reach acceptance: 11 simple and 5 complex first-time suppliers. Supporting the 16 takes 560 hours, but the 84 who drop out along the way also took 371 hours of help, from a single answered question for each of 60 who never qualified, through 16 who qualified but never signed and 4 who signed but never delivered, to a full delivery and re-delivery for 4 whose data failed acceptance. That is 931 hours, or 58.2 hours per accepted dataset, with 39.8 percent of support spent on suppliers who never produced one.

Capacity follows directly. At an illustrative 390 practical hours per support analyst a quarter, one analyst carries 6.7 accepted datasets a quarter at this cohort’s mix. A company whose support per accepted dataset is falling should be able to show which of the four lines fell, and whether the cause was tooling, experience or a shift toward easier suppliers. The model by profile:

Support hours per accepted dataset, by supplier profile and cohort
Supplier profile (illustrative)Evidence, exports, clarifications, delivery (hours)Support hours per accepted datasetWhat moves it
First-time supplier, one source system7.0, 4.5, 9.0, 4.525.0Clear requests and answered questions
First-time supplier, three source systems11.0, 16.0, 21.0, 9.057.0Failed extracts and contract families
Returning supplier, same system3.0, 3.0, 4.5, 4.515.0Reused evidence and a known export route
Cohort of 100 applicants, 16 accepted560 hours on accepted datasets, 371 on suppliers never accepted (39.8%)58.2Applicant fit and how early a no arrives
A grid of illustrative support hours per accepted dataset, with rows for rights evidence, export help, clarifications and delivery support and columns for a first-time supplier with one system, one with three systems and a returning supplier; totals are 25.0, 57.0 and 15.0 hours, and the largest cell, 21.0 hours of clarifications for three systems, is outlined.
Illustrative: hours from this article’s time equations, not measured data.

Do more applications mean more supply? The support test

Not necessarily, and support hours are where the difference shows. Double the applicants to 200 with weaker fit, so that 146 never qualify, 26 qualify but never sign, 5 sign but never deliver and 5 deliver but fail acceptance, and accepted datasets rise from 16 to 18. Support hours rise from 931 to 1,172, up 25.9 percent, against a 12.5 percent rise in accepted datasets.

Support per accepted dataset climbs from 58.2 to 65.1 hours, the share spent on suppliers who never produce a dataset rises from 39.8 to 48.0 percent, and the analyst who carried 6.7 accepted datasets a quarter now carries 6.0. An application chart would show the business doubling; the support ledger shows each accepted dataset costing more.

The remedy is an earlier and cheaper no, not more support staff. Each stage a weak applicant passes before being declined adds hours from the equations above, so the order of checks matters as much as their speed.

Moving dataset-driven support into cost of revenue: the gross margin effect

Put support on accepted datasets into cost of revenue and leave support on suppliers who never reach acceptance with acquisition cost. In the baseline cohort, 60.2 percent of support hours went to datasets that reached acceptance.

Take an illustrative data company that reports a 70 percent gross margin and books all supplier support in an operations line equal to 10 percent of revenue. Moving the dataset-driven 60.2 percent of that line into cost of revenue lowers gross margin by 6.0 points, to 64.0 percent, and leaves 4.0 points with acquisition cost. Neither figure is wrong on its own terms; only the restated one is comparable with a company that books support the way DoorDash books platform costs.

The restatement also changes the operating leverage story. If the 6.0 points fall as a share of revenue while datasets grow, the business is learning; if they hold steady or rise, growth is buying volume with labor.

What could make a low support cost misleading

A low support cost can come from efficiency or from moving the work somewhere it does not get counted. Five patterns deserve a direct question, because each makes the support line look better without making datasets cheaper to produce.

  • Work pushed to suppliers. Asking suppliers to do their own evidence gathering and exports lowers logged hours but can raise drop-off, which shows up later as fewer accepted datasets per applicant.
  • Work hidden in other teams. Rights counsel, engineers who fix exports and account managers who answer questions may not log time as support at all.
  • Work outsourced. Help bought from contractors or vendors can sit in a vendor cost line rather than in support hours.
  • Mix mistaken for learning. A falling average can come from a shift toward returning or single-system suppliers, which need far less help in this model, rather than from better tools.
  • Buyer rework labeled as supplier help. Re-deliveries caused by a changed buyer requirement belong with the deal, not with the supplier’s readiness.

Support questions for any data company, SourceX included

Ask every data licensing company where supplier support sits in its accounts and how it is logged; SourceX should get the same questions, and its published process says plainly that it carries much of the assistance itself.

SourceX’s how-it-works page says that after a short application SourceX itself finishes the system inventory alongside the company, carries out the rights check and chooses how the data will be processed, and that after its go-ahead the records are extracted, transferred and de-identified on approved processing routes. Its application page says no files, passwords or API tokens are needed to apply. Its investor page lists rights and privacy tooling, with ownership checks and removal of personal details built into the workflow rather than handled ad hoc. In a model like that, support is part of what each dataset costs SourceX to produce, which is the article’s point.

SourceX has published no support hours, costs or margins, and none are stated here; its investor page reserves company and commercial metrics for investors. Its current raise, under Rule 506(c), is open only to accredited investors, who can ask for materials on that page. Questions to bring:

  • Where does supplier support sit in your income statement, and which part is in cost of revenue?
  • Can you split support hours by assistance type, stage and outcome for each supplier cohort?
  • What share of support went to suppliers who never reached acceptance, by quarter?
  • How do support hours per accepted dataset differ between first-time and returning suppliers on matched work?
  • How much assistance is done by contractors, vendors or other teams and not logged as support?
  • When applications rose, what happened to support hours per accepted dataset?

Methods note: the time equations, the cohort and what was not measured

The ledger fields, time equations and cohort model were built for this article on 28 October 2026 as an editorial model and computed by xs0183_support_model.py with exact fractions, rounding half up once at print. Every input is illustrative and none is a SourceX figure. Profiles: a first-time supplier with one system has 2 contract families, 0.5 failed extracts, 12 questions, one delivery and 0.25 re-deliveries; with three systems, 4 contract families, 2 failed extracts, 28 questions, one delivery and one re-delivery; a returning supplier on the same system has the 1-hour evidence refresh, 1 new contract family, 0.2 failed extracts, 6 questions, one delivery and 0.25 re-deliveries.

Suppliers never accepted: 0.75 hours for each who never qualified, 7.5 hours for each who qualified but never signed (3 hours of evidence and 6 questions), 20.5 hours for each who signed but never delivered (3 hours of supplier evidence plus 2 contract families, one system, half a failed extract and 12 questions), and 31 hours for each whose delivery failed acceptance (a full one-system profile plus one re-delivery). Accepted datasets in the doubled cohort are 13 simple and 5 complex. Analyst capacity is 390 practical hours a quarter. The gross margin example takes 70 percent and a 10 percent support line as inputs.

Public evidence: Airbnb’s 2025 annual report and its 2025 second-quarter results exhibit, DoorDash’s 2025 annual report, the FASB’s ASU 2025-01 and its In Focus summary of ASU 2024-03, and Harvard Business Review’s 2004 article on time-driven activity-based costing, all checked on 28 October 2026 through search excerpts on the publishers’ own domains, so the article paraphrases them and takes no figures from them. SourceX’s pages were fetched directly that day. Not measured: real support hours at SourceX or any data company, how support divides between first-time and returning suppliers in practice, and how companies classify support in private accounts.

Frequently asked questions

What counts as data supplier support cost?

Staff, contractor and vendor time spent helping suppliers gather rights evidence, export records, answer questions and fix deliveries. It excludes the company’s own data preparation and buyer work. Time spent on suppliers who never reach acceptance is still support cost, but it behaves like acquisition cost and should be reported separately.

Should supplier support be in cost of revenue?

The part that serves datasets reaching acceptance behaves like a cost of producing them, so it belongs in cost of revenue for comparison purposes. Public marketplaces differ: Airbnb reports operations and support as a separate line, while DoorDash includes supply-side onboarding and support in cost of revenue.

How do you measure support per accepted dataset?

Log each assistance event with its supplier cohort, stage, type, driver, minutes and the dataset’s final outcome. Divide all support hours in a cohort, including hours on suppliers who dropped out, by the datasets that cohort got accepted, and keep open cases visible rather than dropping them.

Why can more applications raise cost without raising supply?

Weak applicants still ask questions, gather evidence and sometimes deliver before failing. In this article’s illustrative model, doubling applicants at lower fit raises support hours 25.9 percent while accepted datasets rise 12.5 percent, so each accepted dataset carries more support.

Does SourceX publish its supplier support costs?

No. Its live pages describe the assistance it provides, such as completing the inventory with the company and running the rights check, but give no hours, costs or margins. Its investor page says company and commercial metrics are shared with investors directly.

Sources

  • Airbnb’s 2025 annual report describes operations and support expense as mainly personnel costs and third-party service provider charges for community support, along with customer relations costs. Source
  • Airbnb’s 2025 second-quarter results exhibit lists cost of revenue and operations and support as separate lines in its statement of operations. Source
  • DoorDash’s 2025 annual report says cost of revenue consists primarily of order management, platform and personnel costs, with platform costs including merchant and Dasher onboarding and support for consumers, merchants and Dashers. Source
  • ASU 2025-01 clarifies that public business entities first apply the ASU 2024-03 expense disaggregation disclosures in the first annual reporting period beginning after December 15, 2026. Source
  • The FASB’s In Focus summary says ASU 2024-03 requires a tabular disclosure disaggregating relevant expense captions into categories including employee compensation, purchases of inventory, depreciation and intangible asset amortization. Source
  • Kaplan and Anderson’s 2004 Harvard Business Review article on time-driven activity-based costing has managers account for idle or unused time and build time equations showing how characteristics change processing times. Source
  • Harvard Business Review’s summary of the article says the method requires, for each group of resources, estimates of only two parameters: the cost per time unit of capacity and the time each activity takes. Source
  • SourceX’s how-it-works page says SourceX finishes the inventory alongside the company, runs the rights check and sets the processing route, and that records are extracted, transferred and de-identified on approved processing routes. Source
  • SourceX’s application page says no files and no passwords or API tokens are needed to apply. Source
  • SourceX’s investor page lists rights and privacy tooling, with ownership checks and removal of personal details built into the workflow, and says company and commercial metrics are shared 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