Begin by listing records your business can describe and has permission to offer, not by exporting them. If you're asking how to sell AI data, a buyer first needs to understand what the records cover, how they were made and whether their data quality suits a particular task. ADP (Accord Data Partners) can facilitate an introduction, but it doesn't receive company datasets or decide eligibility.
Your first output is a record inventory, not a folder of files. List the record types, date ranges, systems and people responsible for them. If you're helping someone else's business, your introducer role is to establish interest and permission for a conversation, not to inspect or promise access to its records. The business makes the disclosure and licensing decisions.
Start with the records you hold
Choose one record type you can explain instead of offering ‘everything we have’. Maintenance logs, for example, are easier to describe than an entire customer system. The sell your company data page covers the owner route for anyone considering how to sell AI data; you don't need to send files to explore whether a discussion makes sense.
Calling yourself the data owner doesn't mean you can license every field in a database. Your company may have created operational entries while customers supplied documents or another firm supplied photographs. Note who contributed each part and which team knows the history. That gives you a sensible starting point for checking rights before anyone requests a sample.
Record data provenance in plain terms: which system produced an entry, whether staff corrected it and what happened when the system changed. Keep gaps visible. If older records use different categories, say so rather than merging them silently. A buyer can then judge whether the history fits its purpose, while your team knows what would take work to prepare.
Decide who can authorise a discussion
Confirm who can speak for the company before a buyer conversation becomes a data conversation. Before you decide how to sell AI data, identify who controls access to the records and who must approve the proposed use. An employee's ability to export a file isn't authority to disclose or license it.
If you introduce someone else’s business, get its written consent before naming it to a buyer. Record that written authority and what you may describe. If you run the business, confirm the appropriate internal approval before discussing disclosure. Business consent to meet a buyer is separate from the lawful basis for processing personal data. A privacy assessment may be needed before disclosure, especially where people remain identifiable. The ICO's sharing agreement guidance explains why the parties should document purposes and responsibilities.
Read customer contracts before assuming records are yours to offer. Check restrictions on reuse, onward disclosure and material supplied by others. For a business considering how to sell AI data, copyright permission is another question: GOV.UK's copyright guidance explains that using protected material generally requires permission or an applicable exception. Resolve uncertain rights before promising a buyer access.
Prepare a description before a sample
Hypothetical worked example. A repair firm has dated equipment service logs from several branches. Its initial description could state the years covered, approximate number of entries, typical fields and known changes in recording practice. It could propose a sample scope limited to equipment type and fault categories, without sending customer names, free text or actual entries at this stage.
Make an exclusion list for material you won't discuss without further review. Free text may contain names, photographs may show people and attachments may belong to customers. Separating these from routine operational fields helps you describe a possible dataset honestly. It also stops a general request for ‘a few examples’ from turning into an uncontrolled release.
Describe fields, coverage, missing periods and known limitations before considering a sample. These details help a buyer assess whether further discussion is useful when you're deciding how to sell AI data without sending files prematurely. If a sample becomes necessary, agree its purpose and contents separately, then check permissions and a suitable delivery route. A sample is still a disclosure, not a harmless preview.
Check what the buyer actually needs
Confirm buyer identity before discussing a specific disclosure: which legal organisation would receive the records, who would work with them and whether another party would receive them later. Ask for the intended project and the fields it needs. A broad claim that records are useful for AI isn't enough to decide which subset, if any, your business should consider.
Ask whether the project needs historical patterns, accurate labels, particular dates or ongoing updates. If you're asking how to sell AI data for that project, check whether your inventory actually contains what the buyer needs. If the fields don't exist or require extensive manual interpretation, say so early rather than preparing an unsuitable export.
Assess fitness against the proposed task, not how tidy a spreadsheet looks. Check how consistently staff used categories, whether dates mean the same thing across systems and how often key fields are blank. Distinguish a gap you can explain from one you would have to investigate. That gives both sides a basis for deciding whether further work is worthwhile.
Review the proposed use and rights
Put the proposed licence terms in writing before treating a price discussion as an agreement. Ask whether the buyer wants to train a model, test one or use records for another defined purpose. Cover duration, who may receive the records, whether onward licensing is proposed and what happens when the arrangement ends. Your business and the buyer negotiate their own licence.
Define the permitted use narrowly enough to distinguish a particular training project from unrelated projects or unrestricted sharing. If you're considering how to sell AI data for more than one purpose, remember that a licence granted for one project doesn't automatically cover another. Signing a licence also doesn't settle privacy obligations; where records contain personal data, the parties must address their respective duties before disclosure or processing.
Keep confidentiality separate from ownership and privacy. An internal document might be commercially sensitive even if it contains no personal data, while a photograph may involve someone else's copyright. Ask what the buyer may retain, who may inspect the material and how it will handle information outside the agreed purpose. Don't rely on a single broad permission clause to answer every question.
Agree how records will be transferred
Agree what may be sent and on what basis before preparing an export. If you're exploring how to sell AI data, keep the transfer separate from the initial introduction. Name the approved dataset version, responsible contacts and fields to remove; decide who checks the final export against that description. Release is the business's decision, not an invitation to upload a dataset to ADP.
Choose a transfer method with the buyer that suits the material and the agreement. Specify where the files will be placed, who can retrieve them and how receipt will be confirmed. Never give a buyer or introducer your system credentials. If the parties agree to transfer an encrypted package, send its one-time decryption key through a separately agreed secure channel. Avoid an email attachment simply because it's the quickest route for a colleague.
Set access controls around the agreed transfer, including who can prepare, approve and receive the export. Keep a record of the version sent and the approval behind it, so later queries refer to the same material. Agree how mistaken transfers will be reported and what happens to superseded copies. Those details matter more than the name of the file-sharing tool.
Track acceptance and payment terms
Define acceptance criteria using things both parties can check: the agreed fields, coverage, format and stated limitations. A licence agreement sets rights and duties; dataset acceptance confirms whether the delivery meets the agreed checks. If you're deciding how to sell AI data, don't tie acceptance to an undefined promise that a model will perform well. Write down how the buyer raises a mismatch, what the business can correct or explain, and how an acceptance decision affects payment under the agreed milestones.
Match payment milestones to the actual work and decisions in the agreement. For example, distinguish an agreed preparation task from delivery and from the buyer's review of that delivery. State what triggers an invoice, when a dispute must be raised and what happens if a proposed update is declined. Don't assume a quoted figure is earned merely because talks began.
Plan for a rejected delivery before either party has to respond to one. Decide whether further releases pause while a mismatch is reviewed, what evidence each side will use and whether copies must be returned or deleted if the arrangement ends. Keep the commercial terms consistent with the agreed scope rather than adding open-ended repair work.
Choose the right ADP route
Keep the owner and adviser routes separate. If you're asking how to sell AI data as an owner, decide whether an introduction is appropriate and what you're authorised to describe. An adviser helping a client should establish the client's interest and permission to be introduced, then leave record access and licensing decisions with that business. An initial expression of interest isn't permission to share customer information.
ADP is independent of buyers. It provides a portal and a paid introduction course, but the course explains the introduction model, not how to engineer a model or upload training files. Taking it cannot secure buyer acceptance or earnings. A business and a buyer must decide between themselves whether to proceed and what their licence will say.
Once you have an inventory and confirmed authority, read ADP's information for businesses to decide whether you want to explore an introduction, using your record description and unresolved questions rather than sending files to ADP.
Frequently asked questions
Can pseudonymised records be shared without a privacy check?
Not necessarily. Pseudonymised records can still be personal data. Assess the proposed disclosure and each party's obligations rather than assuming that removing direct identifiers settles the question.
Does ADP approve my records or issue the licence?
No. ADP can facilitate an introduction, but it cannot approve eligibility or offer a licence. Your business and the buyer decide whether to proceed and agree their own terms.
Must I send a sample before speaking to a buyer?
No. Start with a description of the records. If a sample becomes necessary, agree its purpose and contents and check authority, permissions and the delivery route before sharing it.
