Leadership and readiness
Data licensing and business differentiation: what a scope can reveal
By SourceX Editorial · Reviewed by Noah Loul ·
10 sections
Short answer
To assess whether a proposed data scope reveals how your business wins, first write down the three to five things that actually win you work, then ask what a capable analyst could infer about each from the whole package, not field by field. Score who could learn it, how quickly and how much it would matter, then keep, transform, exclude or defer each part, and record the evidence behind every call.
License terms
Data licensing business
Key takeaways
- Start from a named list of how the business wins; a field list alone cannot tell you what matters.
- The assembled package can reveal patterns no single field shows, such as which jobs you decline or which supplier you rely on.
- Have someone outside the deal team write down what they can infer from a prepared sample, before and after transformations.
- Score inferability, recency, uniqueness, harm and audience; the highest factor drives the treatment.
- Exclude what is current, unique and obvious now; leave trade secret status and contract limits to counsel with the documents in hand.
How should an owner assess whether a proposed data scope reveals competitive advantages?
Start from how the business wins, not from the field list. Write down the few advantages that actually win you customers, then ask, for each one, what a capable analyst holding the whole proposed package could work out about it. If the answer is close to your playbook, change the scope before anything else is discussed; if the answer is a general picture any competitor already has, the exposure is low.
Field-by-field reviews miss this. A price column, a customer name and a margin field are easy to spot and remove. What a field review does not catch is the pattern that only appears when thousands of records sit side by side: which jobs you decline, how fast you answer a certain kind of request, which suppliers you switch to when lead times slip.
The worksheet on this page turns that question into a repeatable review: name the advantages, map them to records, run an inference test, score the exposure and choose a treatment. It does not tell you whether to license data, and it does not replace counsel; it gives leadership a defensible basis for scoping a conversation.
Name how the business actually wins before reading a single field
An advantage you cannot name is one you cannot protect. Most owners describe their edge in general terms, such as service or quality. The review needs something sharper: the specific decision, method or relationship that a competitor would copy if they could see it.
Legal definitions help frame the list. U.S. federal law treats information as a trade secret when the owner takes reasonable measures to keep it secret and it derives independent economic value from not being generally known or readily ascertainable. The USPTO's trade secret toolkit lists customer lists, pricing schedules, manufacturing techniques and marketing strategies among the kinds of information that can qualify. Not every advantage is a trade secret, and the worksheet does not need to settle that question, but anything that might be one deserves the strictest treatment.
| Type of advantage | Where it lives in records | What reveals it in aggregate | Example systems |
|---|---|---|---|
| Pricing and bidding judgment | Quotes, bids, change orders, win and loss notes | Which jobs are bid high, low or declined, and why | Salesforce, HubSpot, estimating tools |
| Speed or service model | Ticket timestamps, dispatch logs, SLA fields | Staffing levels and response patterns by customer tier | Zendesk, ServiceTitan, Jira Service Management |
| Process know-how | Work orders, technician notes, checklists, QA records | Repeated fixes and settings that reveal a method | ERP work orders, field service apps |
| Supplier and partner relationships | Purchase orders, vendor records, expedite emails | Preferred suppliers, switching triggers, negotiated lead times | NetSuite, SAP Business One, shared inboxes |
| Customer concentration and mix | Invoices, accounts, contract records | Revenue dependence on a few accounts or niches | Accounting systems, CRM |
| Sales playbook | Call notes, email sequences, proposal templates | The order and wording of what closes deals | CRM activity logs, sales engagement tools |
Why the whole package can say more than any single field
Combined records reveal what separated records hide. The EU Trade Secrets Directive recognizes this in its definition: information can be secret as a body or in the precise configuration and assembly of its components even when the individual pieces are known. A set of ordinary records, assembled, can be the valuable thing.
Two research results show how little an outsider needs. Narayanan and Shmatikov showed that in sparse datasets of transactions and preferences, very little background knowledge is needed to pick out an individual record. The same logic applies to a niche business: if only a handful of customers buy an unusual part, the part number alone can name the customer. Statistical agencies have handled the business version of this for decades. The Census Bureau describes suppressing cells where one contributor could estimate another's value, and then suppressing additional cells because published totals would otherwise let a reader calculate the hidden one by subtraction.
Finally, data used to train a model does not stay neatly inside the model. Carlini and colleagues extracted hundreds of verbatim sequences from a language model's training data, including sequences that appeared in only one document. A licensing agreement can limit use, but the review should assume that distinctive text, such as a proposal template or a pricing rationale written in your words, could resurface.
The competitive exposure review worksheet
Run the worksheet once per proposed scope, with one row per advantage from the earlier table. The output is a short record that leadership signs and that the licensing review can rely on. Each step names the evidence to keep, so the decision can be checked later rather than remembered.
The inference test is the step that makes the worksheet more than a checklist. Give the reviewer the prepared sample and one instruction: describe how this company wins work, who its important customers and suppliers are, and how it decides what to charge or prioritize. Collect the answer in writing before the reviewer sees the advantage list, then compare. Anything the reviewer described correctly is exposed; anything they missed or got wrong is, for now, protected by the scope as prepared.
- Use someone outside the deal team for the inference test; people who know the business read their own advantages into the data and overestimate exposure, or know too much to notice it.
- Test free text separately from structured fields; notes and emails carry reasoning that columns do not.
- Keep the sample small enough to read but drawn from the full date range, so seasonal or recent patterns are not missed.
| Step | Input | Question to answer | Evidence to record | Owner |
|---|---|---|---|---|
| 1. Advantage list | Leadership interview | What three to five things win us work? | Named advantages in one sentence each | CEO or owner |
| 2. Scope map | Proposed systems, tables, date range, free text | Which records touch each advantage? | Mapping of fields and document types to advantages | Data owner |
| 3. Inference test | A sample of the prepared package | What could an analyst conclude about each advantage from the sample alone? | Written conclusions from a person outside the deal team | Independent reviewer |
| 4. Exposure score | Inference test results | Who could learn it, how fast and how much harm? | Scores with a one-line reason each | Owner with finance |
| 5. Treatment | Scores | Keep, transform, exclude or defer? | Chosen treatment and the field-level change | Data owner with counsel |
| 6. Re-test | Transformed sample | Does the inference still work? | Second set of written conclusions | Independent reviewer |
| 7. Sign-off | Steps 1 to 6 | Is the remaining exposure acceptable? | Signed decision and named open questions | CEO or owner |
Scoring exposure: who could learn it, how fast, and what it would cost you
Score each advantage on three factors and let the highest factor drive the treatment. An advantage that is hard to infer but devastating if inferred still needs strong treatment; an advantage that is easy to infer but already common knowledge in your market needs little.
Recency matters more than most owners expect. Records from several years ago often describe a business that no longer operates that way, while last quarter's quotes describe the one competing today. Many scopes become acceptable simply by ending the date range earlier or excluding the most recent period.
| Factor | Low | Medium | High |
|---|---|---|---|
| Inferability | Needs outside data and expert effort | A skilled analyst could infer it from the full package | Visible in a small sample or a single document |
| Recency | Describes practices since abandoned | Partly current | Describes how the business wins today |
| Uniqueness | Common practice in the industry | Known to a few close competitors | Specific to this business |
| Harm if copied | Minor inconvenience | Loss of some deals or margin | Loss of a core customer group or method |
| Who could see it | Narrow, contractually restricted use | Model developers under confidentiality terms | Outputs or data reachable by competitors |
Keep, transform, exclude or defer: choosing the treatment
Pick the lightest treatment that defeats the inference you observed, then prove it with the re-test. Heavier treatments are not automatically safer for the business: they cost usefulness, and a buyer who cannot use the data will ask for a wider scope later, which reopens the review.
Each treatment protects against a different kind of inference. Coding identifiers stops an outsider from naming accounts or suppliers but leaves behavior visible. Coarsening a field, such as turning five priority codes into two, hides the tiering behind a service model. Shifting the date range hides how the business works now while keeping how it used to work. Excluding a record type is the only treatment that reliably protects a method written out in free text, because reasoning cannot be partially removed from a note without rewriting it.
Deferral is a legitimate answer. If an advantage depends on a product launch, a pending contract or a pricing change that will be public within a year, the records describing it may be low risk later and high risk now. Write the date on the worksheet and revisit it rather than excluding the record type for good.
| Treatment | What it changes | What it protects | What it costs |
|---|---|---|---|
| Keep as is | Nothing | Nothing beyond contract terms | No usefulness lost |
| Code identifiers consistently | Names of accounts, suppliers and staff become stable codes | Who you serve and buy from | Little; patterns stay analyzable |
| Coarsen or band fields | Detailed codes, amounts or timings become ranges or fewer levels | Tiering, pricing bands, staffing patterns | Some precision for the buyer |
| Shift or shorten the date range | Drops the most recent period | How the business competes today | Less recent behavior in the data |
| Exclude a record type | Removes notes, templates or files entirely | Methods written in free text | Often the most useful content |
| Defer | Holds a record type for a later scope | Plans that become public soon | Delay, plus a second review |
Illustrative: a commercial HVAC contractor reviews its dispatch history
This scenario is illustrative; the counts are invented to show the method, not measured. A commercial HVAC contractor is asked to license several years of service tickets and technician notes for training a field service assistant. Leadership names three advantages: a same-day response promise for its largest commercial accounts, a diagnostic checklist its senior technicians developed, and favorable lead times from one parts distributor.
The inference test reviewer reads a sample of about 400 tickets. From priority codes and timestamps alone, the reviewer correctly identifies the top accounts and the staffing pattern behind the same-day promise. From technician notes, the reviewer reconstructs most of the diagnostic checklist. From parts fields, the reviewer names the distributor in roughly 70 percent of sampled jobs.
The treatments follow from the scores. Account identifiers are replaced with consistent codes and priority codes are coarsened to two levels, which breaks the account inference. The checklist is the asset the buyer most wants, so leadership decides to keep the notes but end the date range two years before today, when the checklist was less mature. Supplier names are generalized to a category. The re-test reviewer can no longer name the top accounts or the distributor and recovers only an older, partial version of the checklist, which leadership accepts and signs.
What leadership can decide now and what needs more evidence
Leadership can decide immediately which advantages are off the table. If an advantage is current, unique and visible in a small sample, and no transformation preserves the data's usefulness, exclude that record type and move on. That decision needs no outside input and prevents wasted preparation work.
At a glance
What leadership can decide now and what needs more evidence
Other questions need evidence the business does not yet have. Whether a transformation still leaves the data useful depends on what the buyer is building, which comes from the proposed scope. Whether a record type is a trade secret, or whether licensing it is consistent with customer contracts and confidentiality obligations, is a question for counsel with the specific contracts in hand. Mark these as open on the worksheet rather than guessing.
- Decide now: the advantage list, record types that are excluded outright, and the latest acceptable date.
- Decide after the inference test: transformations, sample re-test and the residual exposure.
- Decide with counsel: trade secret status, customer and supplier contract limits, and the restrictions to require in the license.
- Decide with the buyer's scope: whether transformed data still serves the intended use.
Which assumption flips the decision, and how to show the uncertainty
Usually it is the assumption about who will see the data or what a model trained on it will reveal. If the data is used only to train a model under strict confidentiality terms, exposure is mostly through model outputs; if the license allows wider distribution of the dataset, every record is potentially readable by a competitor. Write the assumption on the worksheet and score both cases.
Show uncertainty as ranges and named gaps, not as a single verdict. A line such as 'the reviewer inferred the top accounts in the original sample but not after coding; confidence moderate, sample of about 400 tickets' tells the next reader exactly how much weight the conclusion can bear. Where the inference test was not run for an advantage, say so; an empty row is evidence of a gap, not of safety.
Taking the worksheet into a licensing conversation
A completed worksheet makes the first conversation with a data partner faster, because it answers the questions that usually stall scoping: what is out, what is transformed and why. It also separates the competitive question from the others a license raises. Systems access, privacy permissions, ownership and contractual licensing rights are separate checks, and de-identifying people does nothing to protect business secrets.
A business weighing a SourceX deal begins with a short company assessment that involves no data upload, so the worksheet can stay internal at that stage. Delivery happens only after the owner has approved the scope and terms and signed, and from then on SourceX may use, sublicense and distribute the de-identified dataset only within what that signed agreement permits. The worksheet is the owner's record of why the approved scope looks the way it does.
Frequently asked questions
Can licensing operational data reveal a competitive advantage?
Yes, when the package as a whole shows a pattern a competitor could copy, such as which jobs you decline, how you price unusual work or which supplier you rely on. Single fields are rarely the problem; combinations and free text usually are. A written inference test on a sample is the most direct way to find out.
Is removing customer names enough to protect the business?
No. Removing names protects identities, not methods. Pricing logic, process know-how and supplier relationships can survive name removal intact, so the competitive review has to be run separately from the privacy review.
Who should run the inference test?
Someone outside the deal team who understands the industry, such as a trusted adviser or a manager from a different function. They should read only the prepared sample and write down what they conclude about each named advantage before seeing the advantage list's details.
Does a confidentiality clause make the review unnecessary?
No. Contract terms limit what the recipient may do, but they cannot undo what has been seen, and research has shown that models can reproduce distinctive training text. Use the review to decide scope and the contract to back it up.
How recent can licensed records be without exposing current strategy?
There is no fixed cutoff. Score recency for each advantage: practices the business has moved past carry little risk, while the most recent quotes or notes usually describe how it competes today. Ending the date range earlier is often the simplest fix.
Sources
- 18 U.S.C. 1839(3) defines a trade secret as information whose owner takes reasonable measures to keep it secret and which derives independent economic value from not being generally known or readily ascertainable. Source
- Directive (EU) 2016/943 Article 2 covers information that is secret as a body or in the precise configuration and assembly of its components, has commercial value because it is secret, and is subject to reasonable steps to keep it secret. Source
- The USPTO trade secret toolkit lists customer lists, pricing schedules, manufacturing techniques and marketing strategies among information that can be a trade secret. Source
- Census Bureau research report on cell suppression for economic data: the p% sensitivity rule and secondary suppressions needed because published marginal totals would let users calculate a primary value. Source
- Narayanan and Shmatikov show that for sparse datasets of transactions and preferences very little background knowledge is needed to identify a record. Source
- Carlini et al. extracted hundreds of verbatim text sequences from a language model's training data, including sequences that appeared in a single document. Source
See if your company qualifies
A short company assessment. No data uploads are needed.