Skip to content
Outrich Data

How to brief your board on selling company data

Updated 2026-11-16 · 7 min read · By the Outrich Data team

A board memo on licensing company data should answer five things: what data is in scope, whether the company has the right to license it, what the risks are and how each is reduced, what the company receives, and what approval is being requested. Keep it short and attach the evidence.

Why this decision goes to the board

Licensing company data to AI labs is a new kind of transaction for most companies. It touches customer relationships, employee relations, privacy law and competitive position at once. It is also partly irreversible. Files can be deleted when a license ends, but a model that has learned from the data cannot unlearn it.

For those reasons the decision usually belongs above the level of the team that proposes it. Depending on the company's size and governance, that means the general counsel, the executive team or the board. Whether formal board approval is required depends on the company's own governing documents and on the size of the deal, and counsel should confirm it.

A good memo makes the decision easy to take. It is short, factual and candid about what is not known. It separates the question of principle, whether the company is willing to license data at all, from the question of the specific deal. This guide gives a framework, a risk register and an outline. It is general information and not legal advice.

A decision framework

Boards decide well when a proposal is broken into gates. Each gate is a question that must be answered yes before moving to the next. If any answer is no, the proposal is narrowed until it passes or it stops.

The order matters. Rights come before value, because data the company cannot license has no value in this context however much a buyer would pay. Reputation comes after law, because a deal can be lawful and still be one the company would not want to explain to customers.

  • Scope: can the dataset be described precisely, by system, record type and date range
  • Rights: does the company own it, and do contracts and privacy notices permit this use
  • Privacy: can personal and client-identifying details be removed to a documented standard
  • Strategy: would licensing it weaken the company's competitive position or its own AI plans
  • Reputation: could the company explain the deal comfortably to customers, employees and the press
  • Value: is the return worth the management time and the residual risk
  • Terms: are scope, term, exclusivity, payment security and liability acceptable
  • Control: does the company approve the package and the buyer categories in writing before delivery

The risk register

A risk register lists each material risk with what drives it and what reduces it. It is the part of the memo directors read most closely. Be specific to the dataset in question. The table gives the risks that appear in most data licensing proposals.

The register should not assign false precision. Describing what makes a risk more or less likely is more useful to a board than a score. Where a mitigation depends on a counterparty, such as escrow or a title search, say who provides it.

RiskLikelihood driverMitigation
Breach of a customer contractConfidentiality or use restrictions in customer terms and DPAsContract review; remove restricted accounts; seek consent where needed
Privacy violationPersonal data left in the dataset; notices that did not cover the useDocumented de-identification; notice review; counsel sign-off
Employee objectionMessages or recordings in scope; weak monitoring noticeLimit to work channels; exclude direct messages; notify staff
Third-party IP claimLicensed, open-source or NDA material mixed into the dataTitle search; remove third-party material; tailored warranties
Loss of competitive advantageData reveals methods, pricing or product logicExclude sensitive systems; field-of-use limits; non-exclusive scope
Reputational harmDeal becomes public without contextPrepared customer and employee communications; supplier not named
Non-paymentDelivery before funds are securedFunds in escrow before delivery; release on acceptance
IrreversibilityTrained models cannot unlearn the dataStart with a narrow scope; decide with that limit in view
Open-ended liabilityUncapped indemnities; long survival periodsNegotiate caps and survival; disclose exceptions
Security incident in transferCredentials in files; unsecured transferSecrets scan; encrypted transfer; checksum manifest

Questions the board will ask

Directors tend to ask the same questions, and a memo that answers them in advance saves a meeting. Prepare a direct answer to each of the following.

On rights: whether the company is certain it owns the data, which contracts were reviewed and by whom, and what was excluded. On customers: whether any customer will be surprised, and whether major accounts should be told first. On employees: whether staff communications are included and what staff have been told.

On the counterparty: who the buyer is, whether it is a direct buyer or an intermediary, which AI labs may receive the data and whether the company approves them. On money: what the company receives, when, and how payment is secured. On precedent: whether this sets a policy for future deals. On exit: what happens at the end of the term and what cannot be undone.

Where the honest answer is that something is unknown, say so and say how it will be resolved before signing. A board is more comfortable approving a proposal with stated conditions than one that claims certainty it does not have.

What the board needs to see about the money

Present the financial case plainly. State the offer, the payment schedule and the security for payment. State the internal cost: legal review, engineering time for exports and management attention. State what the company gives up, such as the ability to license the same data elsewhere if exclusivity is granted.

Avoid presenting market figures as a forecast. Publicly reported AI data deals mostly involve large publishers and platforms, and they do not predict what an operating company's dataset will fetch. The relevant number is the written offer in front of the board.

If the proposal is to explore the market and no offer exists yet, say that. Ask for approval in stages: first to run a rights review and seek offers, then to return with a specific deal. When Outrich Data is the buyer, its offer is in writing with a stated validity period, which lets a board consider it at a scheduled meeting.

A memo outline

The outline below fits on two to three pages with attachments. Write the summary and the request first. Directors should be able to stop after the first section and know what they are being asked to decide.

  • Request: the approval sought, in one sentence, with any conditions
  • Summary: the dataset, the counterparty, the consideration and the main risk
  • Background: why AI labs license company data and how the proposal arose
  • Dataset in scope: systems, record types, date range, volume and what is excluded
  • Rights position: ownership, contract review, privacy notices and earlier licenses
  • Privacy and de-identification: method, standard and who performs it
  • Proposed terms: scope of use, term, exclusivity, payment, escrow, warranties and liability
  • Risk register: the table above, adapted to this deal
  • Stakeholders: plan for customers, employees and any works council or union
  • Alternatives considered: doing nothing, a narrower scope, other routes
  • Recommendation and next steps: who signs, and what returns to the board
  • Attachments: data card, title report or rights memo, draft license, counsel's advice

Briefing the general counsel first

The general counsel should see the proposal before the board does. Counsel will want the contracts that touch the data, each version of the privacy notice, the employee policies, and any earlier data-sharing agreements. Collecting these in an indexed data room before the conversation shortens it.

Counsel will also decide how advice is recorded. Legal analysis of the company's exposure is sensitive. How it is shared with the board, and whether it stays privileged, is a matter for counsel to manage. The memo itself can summarize conclusions and leave the detailed advice in a separate privileged document.

A data title search supports this work. It gives counsel a structured review of ownership, contracts, notices and earlier licenses. It does not replace counsel's judgment.

After approval

Board approval should come with clear limits. Record the dataset approved, the scope of use, the counterparty or buyer categories, and any conditions such as completion of the rights review. Name the officer authorized to sign. State what would need to come back to the board, such as a wider scope, exclusivity or a new counterparty.

Then keep the record. File the final license, the data card, the title report, the delivery manifest and the closing statement together. Add the license to a register of data grants, so that the next proposal starts from an accurate picture of what has already been licensed.

Common questions

Does licensing company data need board approval?
It depends on the company's governing documents, its delegation of authority and the size of the deal. Many companies treat a first data license as a board matter because of its novelty and reputational weight. Counsel should confirm what is required.
Who should write the memo?
The executive sponsoring the proposal, with input from legal, security and the team that owns the data. The general counsel should review it before it is circulated and decide how legal advice is presented.
How long should the memo be?
Two to three pages plus attachments is enough. Put the request and summary first. Detail on rights, de-identification and terms belongs in the attachments, where directors who want it can find it.
What is the most common reason a board says no?
Uncertainty about rights and about customer reaction. Both are reduced by doing the contract and notice review before the meeting and by showing exactly which records were excluded as a result.

Find out if your data qualifies.

Eleven questions, about five minutes, no call required. You get an answer on screen and a written follow-up.

esc