Skip to content

Deal economics

Partial archive licensing: a supplier's accept, counter or decline memo

By SourceX Editorial · Reviewed by Noah Loul ·

10 sections

Short answer

When a buyer wants only one part of an archive, the partial archive licensing supplier decision comes down to accept, counter or decline. Decide on three written items: the slice defined in system terms, the internal effort by owner, and the draft terms' conditions, exclusivity and payment triggers. Nothing binds until the company signs, so test the boundary and the effect on the rest of the archive first.

System export

Partial archive licensing

This page covers partial archive licensing: a supplier's accept, counter or decline memo.Illustrative

Key takeaways

  • A buyer's choice of one slice reflects its task and screening rules, not a grade on the rest of the archive.
  • Define the slice by system, date field, status, fields, links and exclusions; a category name describes several different record sets.
  • Zendesk's incremental export works by last change and archived tickets vanish from views, so date and status lines need exact wording.
  • A slice can cost nearly as many internal hours as the whole archive, because the hard tasks attach to the boundary.
  • Tie exclusivity to the slice definition word for word, and start expensive work only after the agreement is signed.

What is the partial archive licensing supplier decision?

It is the owner's choice among three answers when a buyer wants only part of the company's records: accept the slice as offered, counter with a redrawn slice or different terms, or decline. Make it on three written items, not on the conversation: a definition of the slice in system terms, an estimate of the internal effort by owner, and the conditions, exclusivity and payment triggers in the draft terms.

The company holds the deciding right throughout. Nothing binds until it approves the scope and terms and signs, and in SourceX deals no data is delivered before that signature. A counter or a decline costs nothing but time, so the real risk is the opposite one: saying yes to a slice nobody has defined, then discovering the work, the exclusivity or the payment conditions after effort has been spent.

Partial requests are normal. Buyers build datasets for a specific model task, and a task rarely needs every system and year a company holds. The decision is not whether a partial deal is acceptable in principle; it is whether this slice, with these terms, is worth the work it takes and the options it closes.

Why a buyer's pick of one slice says little about the rest

Very little: a buyer selects for its task and its own rules, not to grade the whole archive. Reading the unselected records as worthless, or the selected ones as the archive's price, both lead to bad decisions.

Buyers screen out material for reasons unrelated to quality. OpenAI's Data Partnerships page, for example, asks for large-scale data that is not already easy to find online and says personal or sensitive details, along with information a third party owns, fall outside what it wants. A buyer working to rules like these may drop a well-kept customer chat archive because it is full of personal details and keep a smaller set of engineering tickets that carry none.

Selection also follows systems. SourceX's general guide is that buyers favor a single system holding tens of thousands of records or more, with years of consistent history behind it, usually two or more, though a smaller set can still make the cut if the work behind it is rare. A slice that follows one system's boundary is often the buyer saying where the consistent history is, not what the rest is worth. Ask the buyer which rule drove the selection and write the answer into the memo.

Draw the slice in system terms, not category names

Define the slice by system, object, date field, status, fields and exclusions, because a category name such as support data or the 2022 to 2023 tickets can describe several different record sets. The definition has to be one that an administrator could rerun and get the same records.

Date fields are the most common trap. Zendesk's incremental ticket export returns tickets that changed since a start time, so a pull made that way can include tickets opened years earlier and touched recently, while a contract that says tickets from 2023 probably means tickets created in 2023. Status is the second trap: Zendesk generally archives tickets 120 days after they close, and archived tickets do not appear in views, so a slice pulled from a saved view can silently miss most of the closed history it was meant to contain.

The table lists the boundary lines worth writing into the scope, the wording that holds up and the failure each one prevents.

Draw the slice in system terms, not category names
Boundary lineWrite it asWhat goes wrong when it is vagueHow to check it
System and objectZendesk tickets in the main support brand, not support dataChat, email and a second help desk get pulled in or argued aboutList every system that holds the object and name the ones excluded
Date field and windowCreated between two named dates, in a named time zoneAn export by last change adds old tickets and shifts the countsCount the slice by created, solved and last-changed date and compare
Record statusClosed tickets, including archived onesA view-based pull misses archived ticketsReconcile a search count with the export count
FieldsNamed fields kept, named fields removedFree-text or custom fields carry details the buyer did not ask forAttach the field list to the scope
Linked contextWhich linked records travel as references, and which do notThe slice arrives with broken links and fails reviewSample 50 records and follow every link
ExclusionsCustomers, regions or departments left out, and whyRecords under restrictive contracts slip into the deliveryMatch excluded account IDs against the export
Derived copiesWhether summaries, labels or embeddings made from the slice are coveredDerived material outlives the license with no ruleName each derived output in the terms

Which records outside the slice does the slice lean on?

Usually more than either side expects, so test the slice for dependencies before agreeing to its edges. A ticket slice typically points to customer account records in a CRM, help-center articles agents linked, earlier tickets and engineering issues, and a record cut off from what it points to can fail the buyer's review even when every file opens.

Two common systems show the pattern. When a requester replies to a closed Zendesk ticket, the system opens a follow-up ticket that references the original and copies most of its data, but not its comments, so the conversation history sits in the original ticket, which may fall outside a date-bounded slice. Jira lets work items link across projects with types such as relates to or is caused by, and linking requires permission in each project involved, so a slice drawn around one project can carry links to work items it does not contain.

The buyer will also ask how the slice relates to the whole. The Datasheets for Datasets template asks whether a dataset holds every possible instance or is a sample of a larger set, and if it is a sample, whether it is representative and why not. Answer that question in writing before the buyer does, because the honest answer, selected by the buyer for one task, protects both parties later. For each dependency found, choose one of three responses:

  • Extend the boundary to include the linked records, and re-run the effort estimate for the larger scope.
  • Carry the links as reference fields only, such as the original ticket number and date, with the linked content left out and the gap stated in the scope.
  • Accept the loss, tell the buyer which links will break, and record that the buyer agreed before signing.
A grid of an archive by system and year with one highlighted slice of support tickets, and three links crossing its boundary to records outside it.
Illustrative: a date-bounded ticket slice and the outside records it still points to.

Effort worksheet: who inside the company does the work, and when

A slice can take nearly as many internal hours as the whole archive, because the hardest tasks attach to the boundary rather than to the record count. Estimate effort by owner, and mark which tasks can wait until terms are signed.

In the illustrative worksheet below, the slice holds about a quarter of the archive's records, yet its internal estimate comes to 41 hours against 38 for licensing everything. Filtering and proving an exclusion take an administrator longer than a full export, the link check exists only because of the cut, and someone has to keep the remainder intact and documented. The rows that move least are contract review and sign-off, which depend on how many customer agreements the slice touches, not how many tickets it holds.

Replace every number with your own estimates. De-identification is not on the list because in SourceX deals it is performed by SourceX directly or through its processing partner; the company's privacy reviewer still checks what the slice contains. The last column is the most useful one: anything marked after signing should not start on an indicative offer.

Effort worksheet: who inside the company does the work, and when
OwnerTask the slice createsIllustrative hours, sliceIllustrative hours, whole archiveStart before or after signing?
Systems administratorBuild the filter, prove exclusions, reconcile counts148Count test before; full extract after
Contracts or legal leadCheck customer agreements touching the slice1012Before: it decides whether to counter
Operations leadSample records and follow links across the boundary86Before, on a small sample
Privacy reviewerFlag sensitive fields inside the slice410Before, at field-list level
Executive sponsorApprove scope and terms, sign22At signing
Records ownerKeep the remainder intact and record what was licensed30After signing

What selling one slice does to the rest of the archive

It can keep the remainder fully available or quietly lock it up, and the difference is written in the exclusivity, disclosure and record-keeping terms. Settle all three before accepting the slice.

Exclusivity is the largest lever. In SourceX deals, the signed agreement for that deal is where exclusivity and term are settled, so the question is what that clause covers. An exclusive grant defined with the same boundary lines as the slice leaves the rest of the archive free; one defined by a category name, such as customer support records, can reach every ticket the company holds. Ask for the exclusivity clause to repeat the slice definition word for word.

Keep a machine-readable record of what was licensed. Licensing details tend to fall away as data moves: an MIT Media Lab audit covering more than 1,800 text datasets (Nature Machine Intelligence, 2024) reported high rates of missing or incorrect license information on popular hosting sites. A manifest that states the slice's rules travels better than a memory of a call. MLCommons' Croissant format, for instance, describes a set of files with include and exclude patterns, evaluating the includes first and then removing the excludes, and can carry usage conditions alongside them.

Then run a second-buyer test on the remainder. If another buyer asked next year for the records left out, could you license them without overlapping the first deal, breaking its links or revealing what the first buyer chose? If the answer is no, counter now on boundary, exclusivity or disclosure, while the terms are still open.

Reconcile the indicative offer against terms you would actually sign

Compare every item discussed with the draft agreement line by line, because an indicative figure describes a scope and a set of conditions, and either can change before signing. The draft you would sign is the only reference point; call notes and emails are not.

At a glance

Reconcile the indicative offer against terms you would actually sign

Work through the checks below with the slice definition and effort worksheet beside the draft. Where the draft is silent, treat the item as open and raise it in the counter. Timing commitments belong in the agreement too, but this memo takes no position on them; read what the draft says.

Six reconciliation checks against the draft terms:

  • Scope: the slice definition in the draft matches the boundary table line by line, including the date field, archived records and exclusions.
  • Conditions: the approvals, rights checks or sample reviews that must happen before the deal proceeds are named, with who performs each.
  • Payment trigger: the event that makes payment due is named in the agreement, and so is what happens if review narrows the slice further.
  • Exclusivity and term: what is exclusive, for how long, and whether the definition repeats the slice or reaches beyond it.
  • Changes: a redrawn slice is recorded as a signed amendment, not agreed by email.
  • End of the deal: what happens to delivered records and derived copies when the license ends.

Interest is not a deal: limits of this memo and how it was built

The strongest caution is that eligibility, buyer interest and an indicative offer do not ensure selection, closing or any particular payment. A buyer can narrow the slice after reviewing a sample, change its plans, or decide the rights work is too slow, just as the company can decline. Plan internal effort so the expensive tasks start only after signature, which is why the worksheet has its last column.

This memo also has blind spots. It cannot price the slice or the remainder, because most kinds of business records have no published market price. It cannot judge whether a buyer's selection rule is sound for the buyer's own model. And it assumes the slice can be separated at all; where nearly every record references every other, a partial deal may not be workable and the honest answer is a full scope or none.

Method: the boundary table, dependency test, effort worksheet and memo are editorial frameworks assembled for this article on 2026-10-26 from vendor documentation for Zendesk and Jira, the Datasheets for Datasets template, the MIT Media Lab licensing audit, the Croissant specification and OpenAI's published data criteria. The hours are illustrative, not measured, and no SourceX deal, offer or payment data was used; none has been published.

The supplier decision memo: accept, counter or decline

Write a one-page memo that ends in one of four decisions, so the company answers the buyer from evidence rather than from momentum. The memo is short on purpose: decision, slice definition, dependencies and how each is handled, effort by owner, effect on the remainder, the six reconciliation checks, open questions, and what would reopen the decision. When an item is missing, getting it is the next step: the buyer's selection rule in writing, a count of the slice by each date field, and the draft exclusivity clause.

Choose the decision that matches the evidence, using the rows below; the companion piece on declining an offer covers a full no in more depth.

The supplier decision memo: accept, counter or decline
DecisionWhen it fitsWhat to send the buyer
Accept as offeredThe slice is defined in system terms, dependencies are handled, exclusivity repeats the slice and the effort fitsWritten acceptance of the scope as drafted, with the field list attached
Counter with a redrawn sliceLinks break, the date field is wrong or a cleaner boundary sits nearbyA revised boundary table and the reason for each change
Counter on terms onlyThe slice is sound but exclusivity, conditions or the payment trigger reach too farMarked-up clauses that tie each term to the slice definition
Decline or deferThe slice cannot be separated, the effort outweighs the terms, or it would lock up the remainderA short no, a request to delete anything shared, and the change that would reopen talks

Taking a partial request through SourceX

SourceX handles partial requests on the same principle as this memo: delivery follows signature, and the company signs only once it has approved the scope and terms; after that, whatever SourceX does with the de-identified dataset, whether using, sublicensing or distributing it, stays within what the agreement allows. A redrawn slice is a change to that signed scope, not an informal adjustment.

SourceX starts with a short company assessment, no data upload required, and the fit check on the SourceX home page asks a few questions with no uploads and no commitment. Neither is an offer, and any figure an estimate shows is indicative. An assessment or an estimate does not mean a buyer will select the data, that a deal will close, or that any particular amount will be paid; those depend on the signed agreement. When a delivery is large, it stays in the company's own storage or goes out on encrypted drives, so a slice does not have to be copied into a new system to be licensed.

Frequently asked questions

Can a company refuse to license only part of its archive?

Yes. Nothing binds until the company approves the scope and terms and signs, so it can accept the slice, propose a different one, change the terms or decline. A counter is often the better first answer when the slice is sound but its boundary or exclusivity reaches too far.

Does licensing one slice stop a company from licensing the rest later?

Only if the signed terms say so. Exclusivity, disclosure limits and how the licensed records are recorded decide whether the remainder stays free. Ask for any exclusivity clause to repeat the slice definition exactly, and keep a manifest of what was licensed so a later deal does not overlap it.

Is carving out one slice less work than licensing the whole archive?

Not always. Filtering, proving exclusions, checking links that cross the boundary and keeping the remainder intact add work that a full scope does not need. Contract review depends on how many customer agreements the slice touches, so it shrinks less than the record count.

Which date should define a slice of support tickets?

Name the field explicitly, usually the date the ticket was created, plus a time zone. Exports that work by last change, such as Zendesk's incremental export, return tickets changed since a start time and can add older records. Count the slice by created, solved and last-changed date before agreeing the wording.

Does a SourceX assessment mean a buyer will take the data?

No. The assessment and the fit check on the home page give an early, indicative read and are not offers. Selection, closing and any payment depend on the agreement the company approves and signs.

Sources

  • OpenAI's Data Partnerships page asks for large-scale datasets not already easily accessible online and says it is not seeking sensitive, personal or third-party information. Source
  • Zendesk's incremental ticket export returns the tickets that changed since a given start time. Source
  • Zendesk generally archives tickets 120 days after they are closed; archived tickets do not appear in views but remain reachable through search and the API. Source
  • A reply to a closed Zendesk ticket creates a follow-up ticket that references the original and copies most of its data, but not its comments. Source
  • Jira work items can be linked with types such as relates to and is caused by, and linking across projects requires the link permission in each project. Source
  • The Datasheets for Datasets template asks whether a dataset contains all possible instances or is a sample of a larger set, and whether a sample is representative. Source
  • An MIT Media Lab audit of more than 1,800 text datasets (Nature Machine Intelligence, 2024) found high rates of missing and incorrect license information on popular hosting sites. Source
  • The Croissant specification describes file sets with include and exclude glob patterns, evaluating includes first and then removing excludes, and supports usage conditions. Source

See if your company qualifies

A short company assessment. No data uploads are needed.

See if your company qualifies