Deal economics
Data supplier offer scope: the four lines behind a headline number
By SourceX Editorial · Reviewed by Noah Loul ·
9 sections
Short answer
A headline number in a data offer buys exactly what the terms describe and nothing more. Before comparing offers, reconcile it with four lines of data supplier offer scope: the records covered, the uses granted, the deadlines that apply and the conditions that must be met before money moves. A generous figure on a narrow slice with broad rights can be worth less than a modest one on a wider scope.
Value drivers
Data supplier offer scope
Key takeaways
- Read the number against the slice it buys: the systems, years, fields and record types the terms name.
- Uses are part of the price: sublicensing, distribution, exclusivity and term can give away more than the records suggest.
- Deadlines and conditions decide when, and whether, the number turns into money.
- Compare offers per unit of scope and per unit of rights granted, never by headline alone.
- An indicative number stays provisional until scope and terms are signed, so reconcile it while scope is still negotiable.
What does a data offer's headline number actually buy?
It buys the records, uses, time window and conditions written into the terms, and nothing outside them. A large figure attached to two systems, four years and a short list of fields is a price for that slice; the rest of the archive is not part of the deal, whatever the conversation suggested.
The owner's decision is whether the offer, read at its actual scope, is worth the rights and work it asks for. A useful answer needs four things in writing: the exact records in scope, the uses the licensee may make of them, every deadline that applies to either side, and the conditions that must be met before compensation is due. Missing any one of them makes the headline impossible to compare with anything.
This guide reconciles the headline against those four lines of data supplier offer scope, gives a reconciliation sheet to fill in from the draft terms, works one illustrative example in counts rather than money, and tests the strongest objection: that a narrow scope can be the better deal.
Why a generous figure can sit on a thin slice of the archive
Because buyers price the slice they need, and several ordinary forces shrink that slice after the first conversation. Each is legitimate on its own; together they explain how an offer that sounded like a price for the archive ends up covering a small part of it.
Buyer preference narrows scope first. OpenAI's data partnerships page, for example, asks for large-scale data that is not already easily accessible online and singles out data that expresses human intention, such as long-form writing or conversations rather than disconnected snippets. An archive of short, public or fragmentary records will find only part of itself in play.
Preparation narrows it again. Under the HIPAA Safe Harbor method described in HHS guidance from 2012, de-identifying health information means removing 18 listed types of identifiers, such as names, geographic units smaller than a state and dates tied to an individual. Fields that carried those identifiers leave the deliverable or change form. Any redaction rule works the same way: what is paid for is the processed dataset, not the raw export.
Headline arithmetic can mislead as well. In its 2024 S-1, Reddit reported that the data licensing arrangements it signed in January 2024 carry a combined contract value of $203.0 million, spread over terms of two to three years. That is a total across several agreements and several years, and reading it as one deal or one year would overstate it. A supplier should apply the same discipline to its own offer: is the figure per year or for the whole term, for one dataset or several, fixed or capped?
Records: which systems, years, fields and record types are in scope?
Reconcile the records line by listing what the company assumed was on offer beside what the terms name, at four levels: systems, date ranges, fields and record types. The comparison needs only the draft terms and the company's own inventory, and it is the check most likely to change how the offer reads.
Write coverage as counts, not impressions, for example: two of five systems, 2021 to 2024 of a 2016 to 2024 history, 14 of 60 fields, closed cases only. Then compute a coverage ratio for each level, records in scope divided by records the company assumed. A ratio well below one is not a problem in itself; it is the denominator the headline should be judged against.
Watch for quiet exclusions. Attachments, free-text notes, internal comments and audit history can be the richest part of a business record and the costliest to prepare, and a single clause can include or drop them. Depth also matters to buyers: the general guide is that they want history that stays consistent for years (two or more, typically) and at least tens of thousands of records sitting in one system, so a scope that keeps one system deep may suit a buyer better than one that skims five.
Uses: what the number pays for, and what it gives away
The uses line is where a narrow records scope can still carry a wide grant. Read which purposes are permitted, who else may receive the data, for how long and on what exclusivity, because each is part of what the number pays for.
Separate four questions. Purpose: training, evaluation, retrieval, or all three. Onward rights: whether the licensee may sublicense or distribute, and to whom. Term and exclusivity: how long the grant lasts and whether the company may license the same records elsewhere in the meantime. Survival: what happens to delivered copies and trained models when the term ends.
Know the structure of the deal in front of you. A SourceX deal puts the company's signature, on a scope and terms it has approved, ahead of any delivery; once signed, SourceX may use the de-identified dataset, sublicense it and distribute it within whatever the agreement permits, and the signed agreement for each deal sets exclusivity and term. The signed scope is therefore the boundary of everything downstream, which is why it deserves line-by-line reading.
Check upstream limits too, because a company can grant only what its own vendors and contracts allow. Slack's API terms bar apps offered outside a customer's organization from using API data to train a large language model; an offer that assumes a route the company cannot use covers records that cannot be delivered that way. And when the Data Provenance Initiative went through more than 1,800 text datasets in 2024, it found the correct licenses were often more restrictive than the labels on hosting sites, a reminder that a summary of rights is not the rights.
Deadlines and conditions: when does the number become money?
Only when every condition in the terms is met, in the order the terms set and within the deadlines they state. List each condition, who controls it and what happens if it is missed; anything the company does not control is a discount on the headline, not a detail.
Deadlines run on both sides: how long the offer stays open, when the company must deliver, how long the licensee has to review, and when payment falls due after its triggering event. Write each as a date or as a rule from a defined event, and mark the ones that start only after a step the company cannot schedule.
Conditions decide whether money moves at all: acceptance of the delivered dataset against a stated test, completion of de-identification, a closing step, a minimum volume or a third party's approval. Buyers under documentation rules have reasons to add conditions. Since 2 August 2025, providers of general-purpose AI models in the EU must maintain a copyright policy and publish a summary of their training content, so provenance evidence can become a condition rather than a courtesy.
Read capped and contingent figures for what they are. A number described as up to an amount, or tied to the volume that passes review, is a ceiling until its conditions resolve. Note the share of the headline that depends on events outside the company's control, and compare offers on the part that does not.
The data supplier offer scope reconciliation sheet
The sheet sets the headline beside the signed terms one line at a time, with one test per line. Fill the middle columns from the draft terms, not from emails or calls; where the terms are silent, the line fails until they are not.
Work through these eight lines, in order, before comparing any two offers:
| Line | What the headline implies | What the signed terms must state | Reconciliation test | Red flag |
|---|---|---|---|---|
| Systems | The archive as discussed | Each named source system | Systems in scope against systems offered | A system named in talks but missing from the terms |
| Date range | The full history | Start and end dates for each system | Years in scope against years held | A range defined by an event the company cannot date |
| Fields and record types | Complete records | Included and excluded fields, attachments and notes | Fields in scope against fields held | Free text or attachments dropped without discussion |
| Permitted uses | Training on the data | Each purpose: training, evaluation, retrieval | Purposes granted against purposes priced | A catch-all purpose clause |
| Onward rights | Use by one licensee | Sublicensing and distribution, and to whom | Recipients allowed against recipients expected | Unlimited onward distribution at a single-use price |
| Term and exclusivity | A one-time license | Duration, exclusivity scope and what survives the term | Records locked up against records paid for | Exclusivity wider than the records in scope |
| Deadlines | Money soon | Each deadline as a date or a rule from a defined event | Deadlines the company controls against those it does not | A delivery clock that starts before scope is final |
| Conditions and triggers | A fixed amount | Every condition that must be met before compensation is due | Share of the headline that is conditional | Payment tied to a test the company cannot see |
Illustrative reconciliation: a large number on a small slice
Illustrative: a 400-person logistics company holds nine years of records, 2016 to 2024, across five systems: a transportation management system, a warehouse system, a CRM, a ticketing system and email. It receives an indicative offer with an attractive headline. Every count below is invented to show the method.
Records line: the terms name two systems, transportation and ticketing, for 2021 to 2024, closed cases only, 14 of roughly 60 fields, with no attachments or internal notes. Coverage: two of five systems, four of nine years and about a quarter of the fields. Uses line: training and evaluation, sublicensing permitted, a three-year term, and exclusivity drafted over all shipment records rather than only the four years in scope. Conditions: compensation is due after acceptance of the delivered dataset and a closing step, and part of the figure depends on how many records pass review.
Read this way, the headline is a price for a thin slice with a broad lock-up. The best response is not necessarily rejection: narrowing exclusivity to the records actually in scope, or pricing the excluded years separately, may fix the mismatch. What the reconciliation prevents is comparing this offer with another on headline alone.
Is a narrow scope ever the better deal?
Yes, and sometimes clearly so, which is why the reconciliation measures fit rather than size. A narrow scope means less internal review, fewer exports and a smaller privacy surface, and it leaves the rest of the archive free to license elsewhere when exclusivity is drafted to match.
The objection that matters is different: an indicative number, like an eligibility result, is not a commitment. A dataset can be assessed favorably and still not be selected, not close, or close on a different scope and amount. That cuts both ways. It argues against investing heavy work on the strength of a headline, and for writing the reconciliation early, while scope is still negotiable.
The narrow-scope case fails in three situations: when exclusivity or onward rights reach beyond the records paid for; when the slice strips out the context that made the records valuable, such as notes and outcomes; and when the remaining archive cannot be licensed separately because the slice took its most distinctive part.
The decision memo for a scope-limited offer
End with a short memo that turns the sheet into a decision the owner can sign. Four outcomes cover most cases: accept as scoped; accept if uses or exclusivity are narrowed to match the records; ask to widen the records at a separately stated price; or decline.
At a glance
The decision memo for a scope-limited offer
Where SourceX is involved, the order of events helps. Its first step is a short company assessment that asks for no uploads of data, and the company's signature on approved scope and terms comes before anything is delivered. An initial fit check through the estimator assesses the dataset; it is not an offer and implies no amount. On geography and size, SourceX works with U.S., Canadian and EU companies that peaked at 50 or more full-time employees, and with companies in several other markets whose offered data is at least 90 percent English.
Record the memo in these fields:
- Headline as stated, and whether it is per year or per term, fixed or capped.
- Coverage ratios for systems, years and fields.
- Uses granted, onward rights, term and exclusivity, one line each.
- Conditions and deadlines the company does not control.
- Decision, and the change in terms that would reverse it.
- Next evidence needed: the clause, count or confirmation still missing from the terms.
Frequently asked questions
What does scope mean in a data licensing offer?
Scope is the exact set of records an offer covers, defined by source systems, date ranges, fields and record types, together with the uses the licensee may make of them. Anything not named in the signed terms is outside the deal, however the offer was described in conversation.
Should we compare data offers by the headline number?
No. Compare them per unit of scope and per unit of rights granted: which systems, years and fields each covers, which uses and exclusivity each takes, and how much of each figure depends on conditions. Two equal headlines can buy very different things.
What is the difference between records scope and use scope?
Records scope says which data is licensed; use scope says what may be done with it, by whom and for how long. A narrow records scope with broad sublicensing, distribution or exclusivity can give away more than a wide records scope with tight uses.
Which conditions should be in writing before we rely on an offer?
Every event that must happen before compensation is due: acceptance against a stated test, de-identification, closing steps, volume thresholds and any third-party approvals, each with its deadline or trigger. If a condition exists only in conversation, treat the related part of the number as uncommitted.
Is an indicative offer binding on the buyer?
Usually it is not a commitment to select the dataset, close or pay a particular amount; the signed terms govern. Counsel should confirm how a specific document is worded, because the clauses matter more than labels such as indicative or non-binding.
Sources
- OpenAI's data partnerships page seeks large-scale data not already easily accessible online, including data that expresses human intention such as long-form writing or conversations rather than disconnected snippets. Source
- HHS guidance on the HIPAA de-identification standard describes Safe Harbor, which removes 18 types of identifiers such as names, small geographic units and dates tied to an individual, and Expert Determination. Source
- Reddit's S-1 disclosed data licensing arrangements entered in January 2024 with an aggregate contract value of $203.0 million and terms of two to three years. Source
- Slack's API terms bar apps offered outside a customer's organization from using API data to train a large language model and from bulk exporting message and file data without an additional agreement. Source
- The Data Provenance Initiative's audit of more than 1,800 text datasets found that correct licenses were often more restrictive than those shown on hosting sites. Source
- EU AI Act obligations for general-purpose AI model providers, including a copyright policy and a public summary of training content, apply from 2 August 2025. Source
See if your company qualifies
A short company assessment. No data uploads are needed.