Data licensing board scenario model: conservative, base and upside cases

How do you model conservative, base and upside cases for a data licensing deal?

Model the three cases from one shared input dictionary, change only the inputs that are genuinely uncertain, and use the result to decide whether to fund a scoped investigation, not to forecast income. Each case states the record families in scope, how much clears rights triage, the staff time and calendar time the work takes, the open risks, and the point at which the answer would change.

The model leaves the deal amount blank until a real offer exists. Leadership teams that plug in figures from press coverage of other companies' AI data deals produce a precise-looking number with no evidence behind it, and directors then argue about that number instead of about the facts the company actually controls. What the board can evaluate today is cost, effort, exposure and fit, all of which come from inside the company.

This article gives the input dictionary, the rules that separate conservative, base and upside assumptions, an illustrative worked scenario in counts, percentages and durations, a sensitivity method, and a set of decision rules. All example figures are fictional.

What is the board actually deciding at this stage?

At the first board discussion the decision is almost always whether to spend a bounded amount of leadership and staff time finding out if a licensing opportunity is real, not whether to sign a license. Framing it that way keeps the model small and keeps the downside visible.

Under Delaware's corporate statute, for example, the business and affairs of a corporation are managed by or under the direction of its board, so directors set the frame even when management runs the work. In practice the board approves a scope, a time box, a named decision owner and the conditions that bring the question back to the board, such as a signed term sheet or a rights issue counsel cannot resolve. Whether a specific license later needs board or shareholder approval depends on the company's documents and the deal itself, which counsel confirms.

A useful test for whether the model is aimed at the right decision: every output should map to a yes or no on the next step. If an output only satisfies curiosity, such as an estimate of what the whole archive might be worth, drop it from the board version.

Input dictionary: what every case must state

Every case uses the same inputs, each with a definition, an evidence source and the rule for setting its conservative, base and upside values. Keeping one dictionary means the three cases differ only where the company is truly unsure.

Leave any input blank rather than guessing when no evidence exists yet, and mark it as an open question with an owner. A blank that is labeled honestly is more useful to directors than a filled cell nobody can defend.

What separates a conservative assumption from an upside one?

A conservative assumption uses only evidence already in hand, a base assumption adds what the responsible owner expects with a stated reason, and an upside assumption takes the favorable end of what the evidence still supports. None of the three cases is allowed to contain a value nobody can source.

Three rules keep the cases honest. First, the upside case is not a wish; if the best supporting evidence for a value is a hope that customers will not object, it does not belong in any case. Second, the conservative case carries every known cost and every open risk, including staff time already committed elsewhere. Third, an input moves between cases only if a named owner can explain why; inputs nobody is unsure about stay identical across all three.

Write the reason for each non-identical value in one sentence beside it. When directors ask why the base case assumes more records clear rights triage than the conservative case, the answer should be a sentence like counsel reviewed a sample of the largest customer contracts and found no AI or data reuse restriction, not a general sense that it will probably be fine.

Illustrative worked scenario: one company, three cases

The scenario below is illustrative. It describes a fictional 300-person software company that runs field service scheduling and holds support tickets and work orders. Every figure is a count, a percentage or a duration made up for the example; none is drawn from a real company or deal, and no deal amount appears because none would exist at this stage.

Reading across the rows shows what the model is for. In all three cases the archive is large enough to discuss, so volume is not the question. The cases diverge on how much clears rights triage and how much staff time the review consumes, and the conservative case still contains two open risks. That points to one next step: fund a limited contract review and a metadata-only external assessment, then rerun the model.

Which assumption flips the decision, and how should the uncertainty be shown?

The assumption that flips the decision is the one whose plausible range crosses your decision threshold; find it by moving one input at a time and noting where the recommendation changes. Show the uncertainty as ranges and switch points, not as a single blended number.

This is standard practice in public cost and benefit analysis. The U.S. Office of Management and Budget's Circular A-4 asks analysts to test whether results are sensitive to plausible changes in the main assumptions and to identify the critical values at which the conclusion switches. The GAO Cost Estimating and Assessment Guide treats sensitivity analysis and risk and uncertainty analysis as separate steps, the first changing one assumption at a time and the second producing a range. A board scenario model can borrow both habits without borrowing the government's discounting or dollar conventions.

For a licensing investigation, express each switch point in the input's own units. In the worked scenario, for example, the board might agree in advance that if counsel's sample shows less than a fifth of records clear rights triage, the work stops after the contract review, and if staff hours to a recommendation would exceed what the company can spare in one quarter, the scope narrows to a single record family. Listing two or three such thresholds tells directors exactly which facts to watch.

Why rights triage usually dominates the model

Rights triage usually dominates because it is binary at the record level: a record the company cannot lawfully license contributes nothing, however clean or large the archive. Effort and timing estimates move the model by degrees; a contract or privacy restriction can remove an entire record family.

Three distinctions keep the triage input honest. Having access to records is not the same as owning them; owning them is not the same as having privacy permission to reuse personal data in them; and none of those is the same as having contractual freedom to license them, because customer agreements can restrict reuse. De-identification reduces privacy exposure but does not by itself create a right to license.

Regulators have also signaled that quietly expanding data use is risky. In a February 2024 post, the FTC's Office of Technology warned that adopting more permissive data practices, such as using data for AI training, and disclosing that only through a retroactive change to terms or a privacy policy could be unfair or deceptive. A conservative case should therefore assume that records collected under older notices stay out of scope until counsel reviews those notices.

What can leadership decide now, and what waits for more evidence?

Leadership can decide now whether to fund a bounded investigation, who owns it, and which thresholds bring it back. It cannot decide yet whether to license, at what price or under what terms, because those depend on rights review results and an actual offer.

The decision rules below are a starting set; adjust the thresholds to the company's own capacity and risk tolerance, and record the version the board approved.

Putting the model in front of directors

Give directors one page with the three cases side by side, the switch points, the open risks and the recommended next step, and keep the full input dictionary in an appendix. The page should answer what the company is being asked to approve, what it costs in people and time, what would stop the work, and when the board will hear back.

Context belongs in a short paragraph, not a sales case. The U.S. Copyright Office's report on generative AI training, released in pre-publication form in May 2025, includes a section examining the practicality of licensing options for training material, and the Data Provenance Initiative's Consent in Crisis study documents how quickly websites added restrictions on crawling for AI. Both explain why licensed business records are being discussed at all; neither says anything about what a particular company's records are worth, and the board paper should not imply otherwise.

Record the decision, the approved thresholds and the model version in the minutes or the board pack. When the investigation reports back, rerun the same dictionary with the new evidence so directors can see which inputs moved and why.

• 1Decision requested, scope and time box in two sentences.
• 2Three-case table with only the inputs that differ highlighted.
• 3Two or three switch points in the inputs' own units.
• 4Open risks, each with an owner and a date.
• 5Staff time and elapsed weeks, with named projects that would slip.
• 6What comes back to the board, and when.

Where an outside assessment fits in the model

The most expensive inputs to estimate internally are buyer fit and rights scope, and an outside assessment can supply part of that evidence without moving any data. The opening step with SourceX, a short company assessment for which nothing is uploaded, sits naturally at the investigation gate in the decision rules above.

For the history and volume rows of the dictionary, SourceX offers a general guide rather than a threshold: buyers usually want two or more years of consistent records and tens of thousands of them in one system, rare work can qualify on smaller sets, and every deal is assessed on its own facts. If an opportunity proceeds, the company approves the scope and terms and signs the agreement before any data is delivered, so the board's later approval point stays in place. Nothing here replaces counsel's review or the company's own decision owner.