Leadership and readiness
Enterprise data licensing readiness survey: a protocol for measuring it
By SourceX Editorial · Updated
10 sections
Short answer
A study can measure businesses' readiness to license operational data by surveying a defined population of companies on six checkable conditions: a record inventory, a rights review, a privacy approach, an approval path, preparation capacity and stated willingness. This page publishes that protocol and a 24-item question set, with sampling, scoring, response-rate and uncertainty rules. It reports no results.
Checklist
Enterprise data licensing
Key takeaways
- This page is a survey protocol and question set; it reports no results and no estimates.
- Readiness is defined as six observable conditions scored separately, not as one blended index.
- The unit is the enterprise that would sign a license, drawn from a disclosed frame with fixed strata.
- Outcome rates follow AAPOR standard definitions, and a nonresponse bias analysis is triggered below 80 percent response.
- A validation subsample checks self-reported answers against documents, because self-report flatters.
- Meeting every condition says nothing about the value of the records or the legal right to license them.
How could a study measure businesses' readiness to license operational data?
A study can measure readiness by asking a defined population of companies whether six concrete conditions hold today: they know which records they have, they know whether they may license them, they have a privacy approach for the personal details inside them, they know who would approve a deal, they have people who could prepare the records, and they would consider doing it. Each condition is asked as a fact about the company, not as an opinion about AI.
This page is the protocol for that study and the question set it would use. It reports no results. Nothing here should be read as a finding about how many businesses are ready, because that number does not exist until the survey has been fielded, weighted and checked against the rules below.
Publishing the method first matters for a topic like this. Readiness surveys are easy to write in a way that flatters the sponsor: ask leaders whether their data is valuable and most will say yes. A preregistered design that defines readiness as observable conditions, discloses the sample and reports uncertainty is the only version a journalist, an investor or a company owner should cite.
Readiness as six observable conditions, not a feeling
The study treats readiness as a profile of six conditions, each scored from answers a respondent can check against a document or a system. A company is not scored as ready because its leaders are enthusiastic; it is scored on what it can show.
Keeping the six conditions separate is the main design choice. A company can have an excellent record inventory and no idea whether its customer contracts allow licensing, and a single blended score would hide that. Results are therefore reported condition by condition, and any combined index is secondary and fully decomposed.
| Condition | Definition used in the study | Observable indicator | What does not count |
|---|---|---|---|
| 1. Record inventory | The company can name its systems of record and the record types and years each holds | A written list of systems with owners and date ranges exists | General awareness that the company has a CRM |
| 2. Rights position | The company has reviewed whether contracts, terms and ownership allow licensing of a named record set | A review by counsel or a contracts lead has been completed for at least one record set | A belief that the company owns its data |
| 3. Privacy approach | The company knows what personal information the records contain and what was promised when it was collected | Personal data fields are classified and the governing notices are identified | Having a privacy policy on the website |
| 4. Approval path | The company knows who must approve a license and whether investors or lenders must consent | A named approver and a documented authority rule exist | The CEO would decide |
| 5. Preparation capacity | The company has people or vendors who could export, clean and document a record set | Named roles with available hours, or a vendor in place | An IT team exists |
| 6. Willingness | Leadership would consider licensing a defined record set under acceptable terms | A stated position, recorded with the conditions attached | Interest in AI in general |
Who is surveyed, and which companies fall outside the frame?
The unit of analysis is the enterprise, meaning the legal company that would sign a license, not a single site or branch. AAPOR's 2025 standard definitions for establishment surveys stress starting from a clearly defined sampling unit and define an establishment as one unit or location inside a business, which is exactly why this protocol fixes the enterprise as its unit and records the number of establishments as a variable instead.
At a glance
Who is surveyed, and which companies fall outside the frame
The target population is operating companies above a stated headcount floor in a stated set of countries and sectors. The frame, its supplier and its known coverage gaps are disclosed, following AAPOR's disclosure standard, which asks researchers to say whether a sample is probability-based and, where a list or panel is used, to name its supplier and describe the list.
Two groups are excluded on purpose and named in the report: companies whose main business is selling data, because their readiness is a different question, and companies in active insolvency, because authority over their records sits with a trustee or assignee. The respondent is the person best placed to answer about records and contracts, usually a COO, CFO, general counsel or head of IT, and the role is recorded for every completed case.
- Unit: the enterprise that would sign; establishments counted, not sampled.
- Frame: a named commercial or official business list, with supplier, date pulled and coverage notes.
- Strata: sector group, headcount band and country, fixed before fieldwork.
- Respondent: one key informant per enterprise, with a second informant invited for a validation subsample.
- Window: fieldwork dates fixed in advance and reported exactly.
The question set: 24 items across the six conditions
The instrument has a short screener, 24 core items mapped to the six conditions and a firmographic block. Items ask about facts the respondent can verify, offer a do-not-know option on every item and avoid any wording that implies a value, a price or a buyer. The wording below is the version to preregister; changes after the first pilot are logged.
- S1. Is this company an operating business that would sign contracts in its own name? (yes, no)
- S2. Which best describes your role? (owner or CEO, COO, CFO, general counsel, head of IT or data, other)
- I1. Does the company keep a written list of the software systems that hold its operating records? (yes and current within 12 months, yes but older, no, do not know)
- I2. For each of these record types, which system holds them: customer support, sales and CRM activity, field service jobs, projects, finance, engineering tickets? (system name or none)
- I3. For the record type with the longest history, how many years of continuous records does the company hold? (under 2, 2 to 5, 6 to 10, over 10, do not know)
- I4. Has anyone in the last two years exported a full history from that system for any purpose? (yes, no, do not know)
- R1. Has counsel or a contracts lead reviewed whether customer agreements restrict reuse of records the company holds? (yes for all, yes for some, no, do not know)
- R2. Do any software vendor terms you rely on limit how you may use exported data? (yes, no, not reviewed)
- R3. Has the company ever licensed, shared or sold records to an outside party other than a service provider? (yes, no, do not know)
- R4. Are any record sets subject to an exclusive arrangement with another party? (yes, no, do not know)
- P1. Have the fields containing personal information in your main record systems been identified? (yes, partly, no, do not know)
- P2. Can you locate the privacy notices or terms that applied when those records were collected? (yes, partly, no)
- P3. Has the company ever removed personal details from a record set for any purpose? (yes, no, do not know)
- P4. Does the company hold records from people in more than one legal jurisdiction? (yes, no, do not know)
- A1. Who would have final authority to approve licensing a set of company records? (board, owner, CEO, other officer, not decided)
- A2. Do investor, lender or shareholder agreements require consent for licensing company assets outside the ordinary course? (yes, no, not reviewed)
- A3. Is there a written signing-authority rule for contracts of this kind? (yes, no, do not know)
- C1. Who would export and clean records if the company decided to license them? (internal team named, vendor, no one identified)
- C2. Could those people give the work a few days per month for a quarter without delaying other priorities? (yes, with difficulty, no, do not know)
- C3. Has the company produced documentation of a dataset, such as a field dictionary, for any purpose? (yes, no, do not know)
- W1. Would leadership consider licensing a defined, de-identified set of operating records to an AI developer? (yes, only under conditions, no, not discussed)
- W2. Which conditions would matter most? (choose up to three: customer reaction, confidentiality, effort, approval, terms of use, other)
- W3. Has the company received an inbound request to license or share records for AI training? (yes, no, do not know)
- W4. Has the board or owner discussed licensing company records in the last 12 months? (yes, no, do not know)
- F1 to F4. Sector, headcount band, country of headquarters and number of locations, taken from the frame where possible and confirmed by the respondent.
How answers become readiness scores
Each condition is scored from its items with a fixed rule written before fieldwork, so analysts cannot choose a scoring rule after seeing the data. The rule is deliberately strict: a condition counts as met only when the company can point to evidence, and do-not-know answers count as not met while also being reported as their own category.
| Condition | Items used | Met when | Reported alongside |
|---|---|---|---|
| Record inventory | I1 to I4 | I1 is current and I3 is 2 years or more | Share answering do not know on I1 or I3 |
| Rights position | R1 to R4 | R1 is yes for all or some and R2 is not left unreviewed | Share with prior outside sharing (R3) |
| Privacy approach | P1 to P4 | P1 is yes or partly and P2 is yes or partly | Share with multi-jurisdiction records (P4) |
| Approval path | A1 to A3 | A1 names a decision maker and A2 has been reviewed | Share where A2 is yes |
| Preparation capacity | C1 to C3 | C1 names a team or vendor and C2 is yes or with difficulty | Share with prior dataset documentation (C3) |
| Willingness | W1 to W4 | W1 is yes or only under conditions | Distribution of conditions chosen in W2 |
Sampling, response rates and disclosure: what the report must state
The report states the frame, the sampling method, the fieldwork dates, the sample size in each stratum and the outcome rates, using AAPOR's standard definitions for final case dispositions so that response, cooperation, refusal and contact rates are comparable with other surveys. AAPOR itself cautions that a response rate alone does not tell you how much nonresponse error a survey has, so the rate is reported together with a nonresponse analysis rather than as a quality badge.
Business surveys often draw low response, and the protocol plans for it. For scale, the Census Bureau's Business Trends and Outlook Survey splits a sample of about 1.2 million businesses into six panels that each report every 12 weeks, and a Census working paper on its AI questions reported an average biweekly response rate of about 16 percent for that content. A private readiness survey should expect response well below the levels federal statistical standards aim for.
That is why the nonresponse analysis is mandatory. The federal Standards and Guidelines for Statistical Surveys call for a nonresponse bias analysis when the expected unit response rate is below 80 percent; this protocol adopts the same trigger. Frame variables such as sector, headcount and region are compared between responding and nonresponding companies, and weights are adjusted on those variables with the method named in advance.
Results are published only in aggregate. Before any table is released, cells describing fewer than a preset minimum number of companies are suppressed or combined. NIST SP 800-188 describes de-identification as a way to limit disclosure risk to individuals and establishments while still supporting statistical analysis, and suggests a disclosure review board; the study uses a small review group that signs off every published table.
What sample, baseline and uncertainty make a result defensible?
A defensible result is a share with a confidence interval for each condition, by stratum, from a probability sample with disclosed nonresponse handling. A convenience sample from a newsletter list can still be useful, but it is labeled as such and reported without population estimates or intervals that imply representativeness.
At a glance
What sample, baseline and uncertainty make a result defensible
The sample size is set from the precision the study needs in its smallest reported stratum, not from the total. The study team writes down, before fieldwork, the margin of error it will accept for each condition in each sector group, the design effect it assumes from weighting and stratification, and the minimum completed cases per published cell. If a stratum falls short, it is merged with a named neighbor or reported as insufficient.
The first wave is the baseline. Readiness is a moving target, so the protocol plans repeat waves with identical core items and a fixed reference period, and it reports change only where the interval for the difference excludes zero. New items can be added in later waves, but the 24 core items stay frozen so comparisons survive.
- Primary estimates: share of companies meeting each condition, weighted, with 95 percent intervals.
- Secondary estimates: number of conditions met, shown as a full distribution rather than a mean.
- Comparisons fixed in advance: by sector group, headcount band and whether an inbound request was received (W3).
- Validation subsample: a random set of respondents is asked to confirm three answers against a document; agreement rates are reported.
- Sensitivity checks: results rerun with do-not-know treated as met, and with unweighted data, both shown next to the main estimate.
Conclusions this design cannot support
The survey measures whether companies say, and in a subsample show, that conditions hold. It does not measure whether a given company's records are licensable, what they are worth, or whether anyone would want them. A company that meets all six conditions may still hold records that no AI developer needs, and a company that meets none may hold rare records worth the work of getting ready.
It also cannot separate access from permission. A respondent who can export every ticket from Zendesk has access; whether privacy law, the original notices, customer contracts and ownership allow licensing those tickets are four different questions, and de-identification alone does not create a right to license. The rights and privacy conditions capture whether a company has asked those questions, not the answers a court or regulator would give.
Finally, self-report can flatter. Respondents may overstate inventories and understate contract restrictions they have never read. The validation subsample measures the size of that gap, and if agreement is poor the report says so in its first paragraph, not in a footnote.
Pre-registration checklist to complete before the first invitation
Pre-registration means filing the plan in a time-stamped public registry before data collection. The Center for Open Science describes its purpose as separating hypothesis-generating from hypothesis-testing research, so readers can tell which findings were planned and which were found along the way. For this study the registration covers the following.
At a glance
Pre-registration checklist to complete before the first invitation
Checklist · 12 items
- Research question and the six condition definitions, word for word.
- Target population, frame supplier, date pulled and known coverage gaps.
- Strata, sample allocation and the minimum completed cases per published cell.
- Final instrument text, including the screener and every response option.
- Fieldwork mode, contact schedule, number of reminders and any incentive, with the rule that incentives never depend on answers.
- Case disposition codes and the outcome rates to be reported, following AAPOR definitions.
- Scoring rules for each condition and the treatment of do-not-know answers.
- Weighting variables, method and trimming rule.
- Nonresponse bias analysis plan and its trigger.
- Planned comparisons, interval method and the sensitivity checks listed above.
- Disclosure rules: cell suppression threshold and who signs off each table.
- Conflict-of-interest statement naming the sponsor and its commercial interest in the topic.
0 of 12 ticked. Tick each item as you go.
Using the instrument in your own company, and where SourceX comes in
A company does not need to wait for survey results to use the question set. Answering the 24 items internally, with the strict scoring rule, shows which of the six conditions are missing and which document would close each gap.
Licensing business data directly from companies is SourceX's work, from qualification and rights review through contracting, preparation, delivery and payment. Its first step is a short company assessment that needs no data upload, which fits a company that has just scored itself and wants an outside view of whether its records are worth the remaining work.
As a general guide, AI buyers look for consistent history, typically two years or more, and tens of thousands of records or more in one system, though rare work can qualify with less, and every case is assessed on its own facts.
Frequently asked questions
Has this readiness survey been run yet?
No. This page publishes the protocol and the question set only, and it reports no results. Any figure claiming to show how many businesses are ready to license data should come with its own disclosed sample, dates and method; if it does not, treat it as a claim rather than a finding.
Why not ask executives directly whether they are ready?
A direct question measures confidence, not readiness. Leaders may rate their own data highly, and a single yes or no hides which condition is missing. Asking about checkable facts, such as whether a written system list exists or whether customer contracts were reviewed, produces answers that can be validated and acted on.
What population does the study cover, and what conclusion would its design not support?
It covers operating companies above a stated headcount floor in named countries and sectors, with data sellers and insolvent companies excluded. It would not support conclusions about the value of any company's records, about whether specific records may legally be licensed, or about companies outside the frame.
Can a company reuse the question set internally?
Yes. The 24 core items and the scoring rules work as a self-assessment. Score strictly, count do-not-know as not met, and note the document that would change each answer; that list becomes the work plan for getting ready.
How large does the sample need to be?
That depends on the smallest group the report will describe. The protocol sets precision targets per sector and headcount band before fieldwork, inflates for the design effect of weighting, and merges or suppresses any cell that falls short rather than publishing an unstable estimate.
Sources
- The Center for Open Science says preregistration separates hypothesis-generating (exploratory) from hypothesis-testing (confirmatory) research. Source
- AAPOR's disclosure standard asks researchers to state whether a sample is probability-based, to name the supplier and nature of any list or panel, to give collection dates and to give sample sizes. Source
- AAPOR's Standard Definitions set final case disposition codes and outcome rates (response, cooperation, refusal, contact) for surveys, and AAPOR cautions that a response rate alone does not show how much nonresponse error exists. Source
- AAPOR's December 2025 Standard Definitions for Establishment Surveys define an establishment as a single unit or location within a business and stress starting from a clearly defined sampling unit. Source
- The OMB Standards and Guidelines for Statistical Surveys call for a nonresponse bias analysis when the expected or achieved unit response rate is below 80 percent. Source
- The Census Bureau's Business Trends and Outlook Survey sample of about 1.2 million businesses is split into six panels of about 200,000 that each report every 12 weeks. Source
- A Census working paper reports an average biweekly response rate of about 16 percent for the BTOS AI-related core and supplement content. Source
- NIST SP 800-188 describes de-identification as limiting disclosure risk to individuals and establishments while supporting statistical analysis, and suggests a Disclosure Review Board. Source
See if your company qualifies
A short company assessment. No data uploads are needed.