“For AI” is not a complete description of the use.
Ask the prospective buyer to describe the activity in writing. Training, evaluation and retrieval are different uses. An agreement can cover one, several, or broader activities; a short label on an application form does not settle the scope.
| Proposed use | What to clarify | Question to ask |
|---|---|---|
| Training or fine-tuning | Using records to develop or adapt a model. | Which models, recipients and future training uses are covered? |
| Evaluation | Using records to test a system’s performance. | Will the records or answers become a published benchmark? |
| Retrieval or a knowledge base | Making records available to a system when it answers a query. | Who can retrieve the material, and what may be shown in outputs? |
These are questions for defining the transaction, not conclusions about any particular buyer’s terms. Use the agreement checklist to record the answer and the clause that supports it.
Begin with complete units of work.
A record is easier to evaluate when its parts connect. A support history may link an issue to troubleshooting and resolution. A project file may link the brief to decisions and delivery. A procedure may connect the normal workflow to exceptions and checks.
An initial inventory should describe those relationships without exposing the underlying contents. Record the source system, date range, approximate volume, responsible owner, format and known restrictions. Include a way to distinguish a measured count from an estimate.
Large archives can contain repetitive exports, unfinished work, attachments from third parties and records that should stay outside a proposed license. Treat “available in our system” and “available to license” as separate columns. See the inventory method and worked example.
Prepare a scope before an export.
- Choose a responsible owner. Someone needs to coordinate operations, technology, contract and privacy questions. They do not need to answer every question themselves.
- Describe one candidate collection. “Resolved tickets from one service line over an agreed period” is more reviewable than “all of our data.”
- Record the restrictions. Include client commitments, staff information, third-party content, security secrets and any known hold or retention requirement.
- Get the proposed use. Ask who receives the material, what happens to it, how long it is used, and whether anyone can pass it on.
- Agree on an evaluation method. If a sample is appropriate, approve its scope, recipient and safeguards before assembling it. The first step can remain a description.
The FTC has warned AI companies that their use of data must remain consistent with their privacy and confidentiality commitments. That is a reason to examine existing promises before considering a new use. Read the FTC’s explanation.
A first scope an operator can actually review.
The outcome of that pass is a candidate description and an exclusion list. It is not a valuation, legal clearance, buyer approval or permission to export. If exclusions remove the useful context, record that finding instead of treating the original ticket count as the available dataset.
Our MSP guide works through that decision in detail. For recruiting operations, the staffing guide separates process knowledge from candidate records.
Compare what you keep, what you grant, and what you net.
A proposed payment is only one part of an offer. Put the licensed collection, use rights, exclusivity, intermediaries, preparation work, acceptance test and payment trigger side by side. An estimate before diligence is not the same as an unconditional amount due under a signed agreement.
Retaining ownership does not, on its own, tell you how broad the granted rights are. Termination, deletion of delivered files and treatment of models or other derivatives are separate questions to take into a contract review. Do not assume that one promise answers all three.
The offer evaluation guide includes a hypothetical net-proceeds comparison. The resource library provides a blank comparison sheet and a buyer questionnaire you can use without uploading files.
Leave the first conversation with fewer unknowns.
You do not need a perfect archive to start. You need a description, someone accountable for each open question, and an agreed next step. Start with the five-question readiness check, then use the inventory worksheet to make your answers specific.
Data Specialists provides these preparation tools and a way to request a conversation. A readiness result does not establish buyer demand or an offer. Any specific commercial role, fee arrangement and proposed buyer should be established before you proceed.