Skip to content

Deal economics

The trust economics of data acquisition: test where the no happens

By SourceX Editorial · Reviewed by Noah Loul ·

9 sections

Short answer

Data acquisition trust economics comes down to one test: trust lowers cost only when clearer disclosures make unsuitable suppliers leave earlier or stop suitable ones withdrawing late, while review stays as strict. Completion rates mislead: in this article's illustrative trial, a reassuring page that draws 50 extra applicants raises hours per accepted dataset by 15.6 percent, and a thinner rights review looks cheapest until its defects are counted.

Value drivers

Data acquisition trust

This page covers the trust economics of data acquisition: test where the no happens.Illustrative

Key takeaways

  • Trust lowers acquisition cost through two measurable routes: unsuitable suppliers leave earlier, and suitable ones stop withdrawing after contracting starts.
  • Controlled studies of clearer disclosures show small effects, frequent nulls and occasional backfire, so a large completion gain is a reason for suspicion.
  • Completion and rejection rates can both improve while cost per accepted dataset rises; a disclosure trial needs blind re-review of passed rights files as its guardrail.
  • In the illustrative model, moving the no earlier cuts hours per clean accepted dataset by 12.8 percent, while a volume surge raises them by 15.6 percent.
  • Register the trial, set the minimum effect in advance and publish null and adverse results; SourceX has published no such trial or funnel figures.

Data acquisition trust economics: does trust lower what a dataset costs?

Sometimes, and only through two routes an investor can measure: suppliers who will never fit leave before anyone spends review hours on them, and suppliers who do fit stop walking away late, after contracting has begun. Data acquisition trust economics is the accounting of those two routes. A clearer disclosure that produces neither may leave suppliers better informed without making supply any cheaper.

The decision this page supports is narrow. A data company says that being open with suppliers lowers its cost of acquiring datasets; should an investor give it credit? A useful answer needs comparable supplier cohorts, one shown the disclosure and one not, followed from application to accepted dataset, with every staff hour logged, the stage at which each candidate left recorded, and a check that reviewers did not relax. Without those, a claim about trust is a claim about completion rates.

The idea deserves a fair hearing. Dyer and Chu studied supplier and automaker relationships in the United States, Japan and Korea (Organization Science, 2003) and found that perceived trustworthiness lowered transaction costs and went with more information sharing; the least trusted automaker spent significantly more of its face-to-face time with suppliers on contracting and haggling. The catch is that automotive suppliers deal with the same buyer for years. A business licensing its records usually decides once, with no shared history, so whatever trust exists at the start has to come from what the company is told.

Three ways a clearer disclosure moves the bill, and one way it hides a cost

A disclosure, in this article, is a short statement of the facts a supplier needs before committing time: who approves what before any data moves, which material is usually out of scope, when payment happens and when it does not, and what becomes of data in a deal that falls through. Placed before the first form, it can change acquisition cost through three mechanisms, and it can also mask a fourth effect.

At a glance

Three ways a clearer disclosure moves the bill, and one way it hides a cost

Only the first two lower cost per accepted dataset. The third can raise it while every completion chart improves, and the fourth looks like the best result of all until someone re-reviews the files. The four effects, in the order a trial should look for them:

  • Earlier no: a supplier whose records belong to its clients, or who cannot accept exclusivity for AI training, reads the terms and leaves at intake instead of after a rights review has been paid for.
  • Fewer surprises: a supplier who fits stops withdrawing at the signature stage, and contracting needs fewer clarification rounds, which is the Dyer and Chu mechanism at small scale.
  • More volume: a reassuring page draws more applicants, many of whom fail, so intake and early review hours rise before any extra supply appears.
  • Hidden cost: staff read a calmer, better-informed applicant pool as permission to shorten checks, and rights problems surface after acceptance, where they are most expensive to unwind.

What controlled studies say about clearer disclosures

They show small effects, frequent nulls and at least one way for reassurance to backfire, which is the right prior for an investor. None of these studies involves data licensing; they are the closest controlled evidence on what clearer disclosures do to understanding and participation.

The cleanest test comes from cancer research. In a trial run through the Eastern Cooperative Oncology Group and published in the Journal of Clinical Oncology in 2003, institutions were randomly assigned to give patients either an easy-to-read consent statement or the standard one. Patients who received the easier version reported significantly lower consent anxiety and higher satisfaction, yet their comprehension did not change, and accrual into the parent trials did not differ significantly between the two groups. A disclosure people prefer can leave both understanding and completion where they were.

Survey research adds the backfire. A quantitative review of experiments by Singer, von Thurn and Miller (Public Opinion Quarterly, 1995) found no support for the general claim that confidentiality assurances raise response, but did find a small, significant benefit when the questions were sensitive. An earlier study by Singer, Hippler and Schwarz (1992) reported that such assurances are not always read as reassuring and do not necessarily increase willingness to take part. Business records sit in between: an owner may not see an order history as sensitive until a long privacy paragraph suggests otherwise.

Attention is the third limit. Tracking 48,154 monthly visitors to the websites of 90 online software companies, Bakos, Marotta-Wurgler and Trossen (Journal of Legal Studies, 2014) found that only one or two of every 1,000 retail software shoppers opened the license agreement, and concluded that reading and comprehending terms, not reaching them, is what keeps people uninformed. Human-subjects regulation already builds on that lesson: 45 CFR 46.116 requires an informed consent to open with a concise, focused presentation of the key information most likely to help a prospective subject decide. A supplier disclosure that is long, or that sits behind a link, should be expected to change nothing.

The disclosure trial model: five arms and one rule for giving credit

The model below compares a control cohort with four ways a disclosure trial could turn out, and gives credit only when cost per clean accepted dataset falls without a loss of supply or a rise in defects. Every number in it is illustrative, invented to show the arithmetic; none is SourceX data or describes any company.

Each arm starts with 100 applicants, except the volume arm, which draws 150. Candidates pass through five stages, each with an illustrative cost per candidate who enters it: an intake read (1 hour), a fit call and inventory (2.5 hours), a rights review (5 hours), terms and signature (6 hours) and preparation and acceptance (18 hours). The fewer-surprises arm needs 4.5 hours at terms because suppliers arrive informed; the thinner-review arm spends 3 hours on rights. An accepted dataset with a rights problem that a blind second review would catch counts as a latent defect, costs 40 hours to unwind and is withdrawn, so it never counts as clean supply.

The decision rule is set before any result: credit the disclosure only if hours per clean accepted dataset fall by at least 10 percent against control, clean accepted datasets do not fall, and latent defects found by blind re-review do not exceed control's. Two arms pass. Moving the no earlier (arm A) cuts cost by 12.8 percent with the same ten datasets; fewer late withdrawals (arm B) cut it by 18.8 percent and add three. The volume arm raises cost by 15.6 percent, and the thinner review looks cheapest, at 59.0 hours per accepted dataset, until its two latent defects are unwound and it lands 3.5 percent above control.

Sunk effort tells the same story from the other side. In the control arm 57.3 percent of all hours go to candidates who never become clean accepted supply; earlier exits bring that to 51.0 percent, and the volume arm pushes it to 63.0 percent. The figure plots hours per clean accepted dataset for each arm. The arms, where their candidates left and the verdict each earns:

The disclosure trial model: five arms and one rule for giving credit
Arm (illustrative)Exits after intake, fit, rights, terms; acceptance failuresClean accepted datasetsHours per clean accepted (vs control)Verdict under the rule
Control: current disclosures, 100 applicants40, 25, 15, 8; 2 fail acceptance1076.1Baseline
A. Earlier no: misfits leave at intake49, 25, 11, 3; 2 fail acceptance1066.4 (-12.8%)Credit the disclosure
B. Fewer surprises: fewer late withdrawals40, 25, 15, 4; 3 fail acceptance1361.8 (-18.8%)Credit the disclosure
C. Volume illusion: 150 applicants75, 35, 19, 9; 2 fail acceptance1088.0 (+15.6%)Do not credit: cost rose
D. Thinner review: 3-hour rights checks49, 25, 7, 3; 4 fail acceptance; 2 latent defects1078.8 (+3.5%); 59.0 per accepted dataset before unwindingDo not credit: cost and defects rose
A dot chart of hours per clean accepted dataset for five illustrative trial arms on a 0 to 100 hour scale, with a dashed line at the control value: control at 76.1, earlier no at 66.4 and fewer surprises at 61.8, both credited, volume illusion at 88.0, and thinner review at 59.0 hours per accepted dataset before unwinding, with an arrow to 78.8 per clean accepted dataset once two latent defects are unwound, both not credited.
Illustrative disclosure trial model; values computed by script for this article, not SourceX data.

Why completion and rejection rates mislead in a disclosure test

Because both can improve while acquisition gets more expensive. The strongest alternative explanation for any trust claim is that more applications increase expense without improving accepted supply, and arm C shows how: completed applications rise by half, clean accepted datasets stay at ten, and hours per clean accepted dataset climb from 76.1 to 88.0. A company reporting its application growth after a disclosure change has reported the input, not the result.

Rejection rates fail differently. The share of rights reviews that end in rejection is 42.9 percent in the control arm, 42.3 percent in arm A and 26.9 percent in arm D. Arm D's fall comes from looser review. In arm A the rate barely moves only because misfits of both kinds left early; had the disclosure screened out mainly rights misfits, the rate would have fallen for a good reason. A falling rejection rate is consistent with both stories, so it cannot serve as the guardrail.

The guardrails that can tell the stories apart are blind and downstream. A second reviewer re-reviews a fixed random share of passed rights files in both arms without knowing the arm; the trial counts the problems that reviewer finds, the buyer acceptance failures and any rights issue raised after acceptance. If those hold steady while cost falls, the disclosure earned its credit. If they rise, the saving was borrowed from review.

A trial protocol that still counts when the answer is no

Register the design before the first applicant arrives, fix the outcome and the minimum effect in advance, and publish null and adverse results with the same prominence as gains. A disclosure trial that can only report success measures the company's hopes, not its costs.

At a glance

A trial protocol that still counts when the answer is no

Sample size is the constraint most companies underestimate. With roughly ten accepted datasets per hundred applicants, as in the model, a single cohort per arm cannot separate these arms statistically; the table shows mechanisms, not detectable effects. Pool several cohorts, report intervals around the cost difference, and say how many candidates were still open when each cohort closed. The protocol in full:

  • Treatment: one short key-terms page shown before the first assessment question, with content frozen for the trial; the control group sees the current pages unchanged.
  • Assignment: randomize applicants at first visit, stratified by inbound or outbound source and by data category, or alternate calendar weeks if the same staff speak with both groups; the oncology trial randomized whole institutions for the same reason.
  • Registration: record the design in a public registry before launch; the AEA RCT Registry flags trials registered before the intervention starts and asks for intervention dates, outcomes and whether treatment was clustered.
  • Primary outcome: fully loaded staff hours per clean accepted dataset in each arm at a fixed cohort close, with candidates still open censored rather than dropped or counted as failures.
  • Secondary outcomes: assessment completion, the stage at which each candidate left, contracting rounds per signed supplier and a comprehension check of three to five factual questions answered after reading.
  • Guardrails: blind re-review of a fixed share of passed rights files, latent defects, buyer acceptance failures and rights issues raised after acceptance.
  • Reporting: the 10 percent minimum effect, every outcome named in the registration, intervals rather than single numbers, and adverse effects, such as lower completion among suppliers who would have qualified.

What could make the trust thesis wrong

Five things could make transparency irrelevant to acquisition cost, or turn it into a liability. An investor should hear which of them a company has tested and which it has assumed away.

  • No effect: if suppliers read disclosures the way software buyers read license agreements, one or two in a thousand, the treatment changes nothing, and the oncology trial shows that even a preferred disclosure can leave participation flat.
  • Backfire: reassurance can plant a worry, as Singer, Hippler and Schwarz reported for confidentiality assurances, lowering completion among suppliers who would have qualified.
  • Persuasion posing as disclosure: the FTC staff report Bringing Dark Patterns to Light (September 2022) lists hard-to-find-and-read disclosures among the older tricks it describes; a page tuned for completion rather than understanding invites the same reading.
  • Volume and mix: a disclosure change that coincides with a new channel or a different data category changes who applies, and the cost difference then belongs to the mix, not the message.
  • Review drift and missing hours: reviewers who relax, or founders whose hours go unlogged, make any arm look cheaper than it is; only blind re-review and complete time records catch either.

SourceX's published supplier disclosures as a test case

SourceX's live pages already state several facts a disclosure treatment would contain, which makes it a convenient worked example rather than evidence. Its application page tells companies that nothing is licensed or transferred by applying, that no passwords or API tokens are requested and that nothing is signed by applying. Its how-it-works page says that pricing stays indicative until a Dataset Deal Confirmation is signed, that data that does not reach Closing earns no payment, and that if a deal does not close the supplier receives a no-deal notice and declined data is deleted.

Whether statements like these move the stage at which candidates leave, shorten contracting or leave review intact is precisely what the trial measures, and SourceX has not published such a trial or any funnel, cost or acceptance figures; its investor page reserves company and commercial figures for direct conversations with investors. Its current raise is a Rule 506(c) offering limited to accredited investors, who can ask for materials at sourcex.si/investors. Questions worth bringing to that conversation, or to any data company's:

  • When did your supplier-facing disclosures last change, and how did the stage at which candidates left shift in the cohorts before and after?
  • What are fully loaded hours per accepted dataset, and do they include founders' and contractors' time?
  • How many contracting rounds does a signed supplier need now, compared with before the change?
  • What share of passed rights files is re-reviewed blind, and how often does the second reviewer disagree?
  • Which disclosure or intake change produced a null or adverse result, and what did you do with it?

Methods note: inputs, rules and limits of the disclosure model

The disclosure trial model was built for this article on 2026-10-28 and computed by a short script, trust_disclosure_model.py, kept in the writer's working folder for review. All inputs are illustrative: 100 applicants per arm (150 in arm C), the exits after each stage and the acceptance failures shown in the table, stage costs of 1.0, 2.5, 5.0, 6.0 and 18.0 hours per entering candidate (4.5 hours at terms in arm B, 3.0 at rights review in arm D), and 40 hours to unwind each latent defect. No randomness is used.

The script computes total hours as entrants times hours per candidate at each stage plus unwind hours, clean accepted datasets as accepted minus latent defects, hours per clean accepted dataset as total hours divided by clean accepted, sunk share as the hours not spent on the end-to-end path of a clean accepted dataset divided by total hours, the rights rejection rate as exits after rights review divided by entrants to it, and the change against control. It applies the three-part decision rule stated above. Values are rounded half up, once, at the end.

Sources: Dyer and Chu (2003), Singer, von Thurn and Miller (1995), Singer, Hippler and Schwarz (1992), the Eastern Cooperative Oncology Group consent trial (2003), Bakos, Marotta-Wurgler and Trossen (2014), 45 CFR 46.116, the FTC's 2022 dark patterns report, the AEA RCT Registry and SourceX's live application, how-it-works and investor pages. Not measured: any real disclosure effect, at SourceX or elsewhere; comprehension, which the model does not simulate; and the statistical power of any particular trial size. No SourceX results are reported.

Frequently asked questions

Does more transparency always increase supplier completion?

No. A review of survey experiments found confidentiality assurances helped response only when questions were sensitive, and an oncology trial found that an easier consent statement raised satisfaction without changing comprehension or accrual. Completion can rise, fall or stay flat, so a trial has to measure it rather than assume it.

What is the best single measure of data acquisition trust economics?

Fully loaded staff hours per clean accepted dataset, compared between a disclosure arm and a control arm. It needs guardrails beside it: blind re-review of passed rights files, buyer acceptance failures and rights issues raised later, so that a saving borrowed from review is not counted as a gain.

Why not judge a disclosure by how many applications it brings?

Because more applications can raise total effort without adding accepted supply. In this article's illustrative volume arm, applications rise by half, accepted datasets stay at ten and hours per clean accepted dataset rise by 15.6 percent.

How large does a disclosure trial need to be?

Larger than one cohort per arm. When only about ten in a hundred applicants become accepted datasets, differences of the size this model shows cannot be separated from noise in a single cohort, so companies should pool cohorts, report intervals and say how many candidates were still open at the close.

Has SourceX published results from a disclosure trial?

No. SourceX's live pages state several supplier-facing facts, such as indicative pricing until a Dataset Deal Confirmation is signed and no payment for data that does not reach Closing, but it has published no trial, funnel or cost figures. Company and commercial figures are shared with investors directly.

Sources

  • Dyer and Chu (Organization Science, 2003): in supplier and automaker relationships in the United States, Japan and Korea, perceived trustworthiness lowered transaction costs and went with more information sharing; the least trusted automaker spent more face-to-face time on contracting and haggling. Source
  • Singer, von Thurn and Miller (Public Opinion Quarterly, 1995): a quantitative review of experiments found no general effect of confidentiality assurances on response, but a small significant benefit when questions were sensitive. Source
  • Singer, Hippler and Schwarz (1992): confidentiality assurances are not always perceived as reassuring and do not necessarily increase willingness to respond. Source
  • Coyne and colleagues (Journal of Clinical Oncology, 2003): institutions randomized to an easy-to-read or standard consent statement; the easier version lowered consent anxiety and raised satisfaction, comprehension was unaffected and accrual did not differ significantly. Source
  • Bakos, Marotta-Wurgler and Trossen (Journal of Legal Studies, 2014): 48,154 monthly visitors to 90 software company sites; only one or two of every 1,000 shoppers accessed the license agreement; reading and comprehension, not access, limit how informed buyers become. Source
  • 45 CFR 46.116(a)(5)(i) requires informed consent to begin with a concise and focused presentation of the key information most likely to help a prospective subject decide. Source
  • The FTC staff report Bringing Dark Patterns to Light (September 2022) describes dark patterns across industries and names hard-to-find-and-read disclosures among older tricks. Source
  • The AEA RCT Registry registers randomized trials in economics and other social sciences, flags trials registered before the intervention starts and records intervention dates, outcomes and clustering. Source
  • SourceX's live application page says nothing is licensed or transferred by applying, no passwords or API tokens are requested and nothing is signed by applying. Source
  • SourceX's live how-it-works page says pricing stays indicative until a Dataset Deal Confirmation is signed, data that does not reach Closing earns no payment, and declined data is deleted after a no-deal notice. Source
  • SourceX's live investor page says it 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