Deal economics
Dataset evaluation cash conversion: 4 tests before faster review pays
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
Faster buyer evaluation improves cash conversion only when the evaluation sits on the path to the payment trigger. Four tests decide it: payment follows acceptance, money is still at risk during the review, no fixed payment date or budget release overrides that clock, and speed does not raise disputes. Measure contingent spend-days, the share of cost at risk times the days until payment is certain.
Evaluation split
Dataset evaluation cash
Key takeaways
- Evaluation speed shortens the certainty gap; it shortens the collection gap only when the contract starts the payment clock at acceptance.
- The value of a faster review is the share of cost at risk during it multiplied by the days saved, so delivery reviews matter far more than sample reviews.
- In the illustrative timelines, staging work behind a sample cut exposure by about a third without any speed-up at all.
- Under a fixed payment date or budget release, a faster review buys certainty but no cash.
- Recompute evaluation speed without the largest buyer; if the gain vanishes, the company has a fast customer, not a fast process.
When does faster buyer evaluation improve cash conversion?
Faster buyer evaluation improves cash conversion when the end of the evaluation starts the payment clock and money spent before it is still at risk. If payment is tied to a calendar date, a budget release or a signed order instead, a quicker review brings certainty sooner but leaves the cash date where it was.
Three terms keep the analysis honest. Buyer evaluation is the buyer's review of a sample or a delivery against agreed criteria. Payment certainty is the moment the buyer's duty to pay stops depending on its own judgment, usually acceptance or a signed order with no open conditions. Cash conversion here means two things: the days from first spend to cash, and how much of the deal's cost is exposed along the way.
The evidence that settles the question for one company is dated, contract by contract: first spend, evaluation start and end, acceptance, invoice and cash, plus spending by date. An average of days sales outstanding cannot answer it, because the exposure that matters sits before any invoice exists.
The certainty gap and the collection gap: stages from first spend to cash
Between first spend and cash, a data deal passes through up to eight stages, and at each one the money already spent is either still contingent on the buyer's judgment or simply waiting for a clock.
Steps
The certainty gap and the collection gap: stages from first spend to cash
The useful split is between the certainty gap, from first spend to payment certainty, and the collection gap, from payment certainty to cash. Evaluation speed works on the certainty gap. It reaches the collection gap only when the contract starts the payment clock at acceptance.
A quick evaluation does not have to be a loose one. Anthropic's guide to agent evaluations says teams often wait because they assume they need hundreds of tasks, when 20 to 50 simple tasks drawn from real failures is a great start. A buyer and seller who agree a task set of that size before the sample ships can shorten the sample review without lowering the bar.
The stages, with what is contingent at each:
- Scoping: spending on specification and fit checks, contingent because the buyer has committed to nothing.
- Sample preparation: a small share of total cost, contingent on the sample review.
- Sample evaluation: the buyer tests the sample on its own tasks; only the sample's cost is at risk.
- Order or commitment: the buyer signs for a defined scope, and certainty rises, but delivery acceptance usually remains open.
- Full preparation: most of the cost is spent here, contingent on delivery acceptance.
- Delivery evaluation: the acceptance window, during which everything spent so far is at risk.
- Invoice and payment term: payment is certain but not yet received, so only credit risk remains.
- Cash: the collection gap closes.
Contingent spend-days: what each day of evaluation is worth
Contingent spend-days add up, for every day before payment certainty, the share of the deal's total cost already spent. The value of a faster evaluation is the share at risk during the review multiplied by the days saved.
Express spending as a percentage of total deal cost so the measure works across deal sizes, and call the unit percent-days. A deal with all of its cost spent and a 45-day delivery review accumulates 4,500 percent-days during that review alone. Cut the review to 15 days and the saving is 3,000. Shorten a sample review by the same 30 days when only 15% of cost has been spent and the saving is 450.
Track spend-days to cash beside it, which adds the collection gap. The pair separates two benefits that are easy to blur: faster certainty, which lowers the chance of spending more on a deal that will fail, and faster cash, which frees capital for the next deal.
Neither measure needs price or revenue, so a company can report both per contract without disclosing what any buyer paid.
The evaluation-to-cash matrix: where the review sits decides what speed buys
Where an evaluation sits in the contract decides what a faster one buys. The matrix maps six evaluation positions to what becomes certain when the review ends, whether cash moves, how much spend is exposed during the review and what the position says about buyer commitment and concentration.
The last column ties speed to concentration. A company whose quick evaluations all come from one buyer has learned that buyer's process; it has not built a faster cycle.
| Evaluation position | Certain when it ends | Does faster review move cash? | Spend exposed during review | Commitment and concentration read |
|---|---|---|---|---|
| Sample review before any order | The buyer's interest in a scope | Indirectly, by bringing the order date forward | Low: the sample's cost only | Quick reviews from many buyers signal broad demand; from one buyer, a single program |
| Paid evaluation license | The buyer's willingness to pay for testing | Yes for the next order, if the license converts | Low to moderate | A paid test commits the buyer to testing, not to buying at scale |
| Delivery acceptance, invoiced on acceptance | The payment obligation | Yes, day for day | High: most cost already spent | The cash cycle follows the slowest large buyer's review queue |
| Delivery acceptance, invoiced on delivery | The absence of a rejection | Little, unless a dispute halts payment | High | Suggests seller leverage; check for holdbacks tied to acceptance |
| Payment on a fixed date or budget release | The obligation, but cash waits for the date | No | High until acceptance, then waiting | Buyers sharing one budget calendar move cash together |
| Multi-year contract paid over the term | Each tranche as its conditions are met | Only for the first tranche | Varies by phase | Contract value overstates near-term revenue (Reddit: at least $66.4 million of $203.0 million expected in 2024) |
Five illustrative timelines: speed, staging and a fixed payment date
In five illustrative timelines, cutting a delivery review from 45 to 15 days brings cash forward 30 days and removes a third of spend-days to cash when the invoice follows acceptance. When payment waits for a fixed date, the same cut leaves the cash day and spend-days to cash unchanged.
Illustrative assumptions: full preparation takes 30 days and spends the deal's cost evenly; payment is due 30 days after acceptance unless stated. In the staged timelines, a sample costing 15% of the total takes 10 days to prepare, full preparation then takes 20 days and the delivery review 15 days. Percent-days are percentage points of total cost multiplied by days. No timeline describes a real deal.
A and B isolate speed: the same deal with a 45-day or a 15-day delivery review. C keeps the 15-day review but pays on day 120 regardless; with a 45-day review it would show the same cash day and the same 10,500 percent-days to cash, so the faster review buys certainty only. D stages the work behind a 30-day sample review and reaches the same cash day as A with about a third less exposure. E shortens D's sample review to 10 days: cash arrives 20 days sooner, but exposure falls by only 300 percent-days, because little had been spent while the sample was under review.
Two lessons follow. Speed pays most where exposure is highest, at delivery acceptance under an acceptance-triggered invoice. Staging cuts exposure more than speed does, and it works even with a slow buyer.
| Timeline (Illustrative) | Review being shortened | Cash day | Contingent spend-days | Spend-days to cash |
|---|---|---|---|---|
| A. Prepare, then 45-day delivery review | None (baseline) | 105 | 6,000 | 9,000 |
| B. Prepare, then 15-day delivery review | Delivery review, 45 to 15 days | 75 | 3,000 | 6,000 |
| C. As B, but paid on day 120 | Delivery review, 45 to 15 days | 120 | 3,000 | 10,500 |
| D. 30-day sample review, then prepare | None (staged baseline) | 105 | 3,175 | 6,175 |
| E. 10-day sample review, then prepare | Sample review, 30 to 10 days | 85 | 2,875 | 5,875 |
Which rules already bound acceptance and payment clocks?
Some procurement law already treats acceptance time as part of payment time, which is why a contract's acceptance clause is the first place to look for cash timing.
In the EU, Directive 2011/7/EU on late payment says a contractual payment period between businesses should not exceed 60 calendar days unless expressly agreed and not grossly unfair to the creditor. Its recitals set 30 calendar days as the general maximum for a procedure of acceptance or verification, with longer periods allowed in particularly complex contracts when expressly agreed.
For U.S. federal contracts, FAR 32.904 makes an invoice payment due on the later of the 30th day after the billing office receives a proper invoice and the 30th day after government acceptance. For computing interest penalties only, it deems acceptance to occur on the 7th day after delivery unless there is a disagreement over quantity, quality or compliance, and it states that this does not compel officials to accept or pay before fulfilling their responsibilities.
Most AI data deals are private commercial contracts, so these rules work less as constraints than as reference points. They show that acceptance windows can be bounded and that deemed acceptance is a recognized drafting tool. Whether either applies to a given deal is a question for counsel.
Does one fast buyer make the cash cycle look healthier than it is?
Often it does. When a few buyers dominate revenue, a company's evaluation speed is really those buyers' review capacity, and better cash conversion can be dependence that happens to be quick.
This is the strongest alternative explanation for any improvement. A prestigious buyer or a large mandate may review fast because its program is a priority, then stop buying when priorities move. Appen's ASX update of 12 February 2024 shows how abrupt that can be: Google was ending its contract after a strategic review, with all projects ceasing by 19 March 2024, after $82.8 million of FY2023 revenue.
Concentration is common enough to check every time. Reddit's Q1 2026 10-Q says that "substantially all of the contract value associated with our licensing revenue is derived from two of our partners," and Innodata's FY2025 10-K reports one customer at about 58% of 2025 revenue. In a base like that, a faster review by the top buyer improves the average cash cycle and the concentration risk at once.
The test is straightforward: report evaluation days and spend-days by buyer, then recompute the averages without the largest buyer. If the improvement disappears, the company has a fast customer, not a fast process. Correlated timing counts as well, since buyers that share a budget calendar or a research cycle move cash together however many logos there are.
What could make the evaluation-speed case wrong
The case weakens if speed comes from looser checks, if payment terms rather than reviews set the cash date, or if the simple timelines hide where money actually goes.
- Speed bought with weaker tests can turn rejections into disputes after payment, so the days saved come back as refunds or rework.
- Payment terms can stretch to absorb a faster review; the Late Payment Directive itself allows longer periods when expressly agreed and not grossly unfair to the creditor.
- Projects can end after evaluation whatever the result: Gartner predicted in July 2024 that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025, citing data quality, risk controls, escalating costs or unclear business value.
- The timelines assume even spending, no credit losses and a single buyer; real deals spend in lumps and pay data owners on their own schedule.
- Revenue and cash differ: a contract can be recognized over its term while cash follows invoices, so neither alone shows conversion.
- Evaluation speed is partly outside the seller's control; the seller controls how ready its evidence is, not the buyer's queue.
Evaluation-timing questions for a data company, SourceX included
Ask for dated evidence per contract rather than averages, and ask which clock each contract starts at acceptance. The same questions apply to SourceX.
SourceX is one application. When a business licenses operational data to AI developers, SourceX covers the work from sourcing and qualification through rights review, contracting and preparation to delivery and payment. Nothing is prepared for stock: a dataset is sourced only after a buyer requests it, so preparation trails a stated need instead of running ahead of one, and SourceX receives no data until the supplier has said yes to the scope and terms and signed the deal.
SourceX has not published evaluation, cash or contract timing metrics; company and commercial metrics are something it provides to investors directly. Its capital raise runs under Rule 506(c) and is open to accredited investors only, so an investor who requests materials can put the dated-evidence questions below to the company itself.
- For each contract in the last four quarters, what were the dates of first spend, evaluation start and end, acceptance, invoice and cash?
- Which contracts start the payment clock at acceptance, and which at delivery, signature or a fixed date?
- What share of total cost is typically spent before the first buyer evaluation, and before delivery acceptance?
- What are contingent spend-days and spend-days to cash by buyer, and how do the averages change without the largest buyer?
- How many acceptances were later disputed, and did quicker reviews produce more disputes?
- Which buyers share a budget calendar or research cycle, and how much revenue moves with each?
Methods note: how the timelines and matrix were built
The evaluation-to-cash matrix, the five timelines and the contingent spend-days measure are editorial tools built for this article on 2026-10-26. Public anchors: Directive 2011/7/EU, FAR 32.904, Reddit's 2024 S-1 and Q1 2026 10-Q, Innodata's FY2025 10-K, Appen's ASX update of 12 February 2024, Gartner's July 2024 press release and Anthropic's agent evaluation guide. All timeline inputs are illustrative and expressed in days and percentages of cost.
Not measured: actual evaluation durations, spend profiles, payment triggers or dispute rates at any company, SourceX included, and the share of AI data contracts that invoice on acceptance. The timelines leave out credit losses, payments to data owners and financing costs.
Frequently asked questions
Does faster dataset evaluation always speed up cash collection?
No. It speeds cash only when the payment clock starts at acceptance. When payment follows delivery, a signed order, a fixed date or a budget release, a quicker review raises certainty sooner but leaves the cash date in place.
What are contingent spend-days?
They add up, for each day before payment becomes certain, the share of a deal's total cost already spent, expressed in percent-days. The measure shows how much spending sits at risk during each review and therefore what shortening that review is worth.
Why is days sales outstanding not enough for AI data deals?
Days sales outstanding starts counting at the invoice. In data deals much of the cost is spent and at risk before any invoice exists, during sample and delivery reviews, so a company can show a short DSO while carrying long, exposed evaluation periods.
Is there a legal limit on how long a buyer can take to accept a delivery?
Sometimes. The EU Late Payment Directive's recitals set 30 calendar days as the general maximum for acceptance or verification unless expressly agreed otherwise, and U.S. federal rules deem acceptance on the 7th day after delivery for interest purposes. Most private data contracts set their own windows, so counsel should check the specific terms.
How can an investor tell a fast process from a fast customer?
Ask for evaluation days and spend-days by buyer, then recompute the averages without the largest buyer. If the improvement disappears, the speed belongs to one customer's priorities rather than to the company's process.
Sources
- Anthropic's agent evaluation guide says 20 to 50 simple tasks drawn from real failures is a great start for an evaluation set. Source
- Reddit's 2024 S-1 reported January 2024 data licensing arrangements with $203.0 million of aggregate contract value over two-to-three-year terms, with at least $66.4 million expected to be recognized in 2024. Source
- Directive 2011/7/EU limits contractual business-to-business payment periods to 60 calendar days unless expressly agreed and not grossly unfair, and its recitals set 30 days as the general maximum for acceptance or verification. Source
- FAR 32.904 sets federal invoice payment at 30 days after the later of a proper invoice or government acceptance and deems acceptance on the 7th day after delivery for interest penalty purposes. Source
- Appen told the ASX on 12 February 2024 that Google would end its contract after a strategic review, with projects ceasing by 19 March 2024; FY2023 revenue from Google was $82.8 million. Source
- Reddit's Q1 2026 10-Q states that substantially all of the contract value associated with its licensing revenue is derived from two partners. Source
- Innodata's FY2025 10-K reports one customer at about 58% of total revenue in 2025. Source
- Gartner predicted in July 2024 that at least 30% of generative AI projects would be abandoned after proof of concept by the end of 2025. Source
Request SourceX investor materials
SourceX shares company and commercial metrics with investors directly. Any offering is made only to verified accredited investors.