BetterUp's privacy-notice archive needs a cohort-to-version applicability map

Treat the archive as version evidence, not applicability evidence

The direct answer is to use the archive to establish which public notice versions exist, while requiring separate buyer evidence for which version governed a specific coaching population at a specific time. A current notice may describe today's provider position without resolving an earlier pilot, a continuing cohort, a regional deployment, or a feature introduced under different terms. Conversely, an older notice in the archive does not prove that every participant enrolled during that period received it or that its provisions controlled over the contract, data-processing agreement, product terms, or a later communication. Version history is necessary evidence, but applicability is a relationship among documents, dates, configuration, and people.

The accountable team should translate this point into a named workflow, affected population, source data, human owner, approval right, exception path, retained evidence, and review date. That translation is what separates an interesting AI development from a decision that can be governed and evaluated.

Connect the participant cohort to the governing record

For each materially distinct cohort, the buyer should be able to identify the employer or sponsor, participant population, region, enrollment and access dates, product and coaching mode, enabled AI features, source of company or participant data, contract and data-processing agreement, notice version delivered, and any consent or acknowledgment the operating model relies on. This map should also show whether human coaches, AI coaching, simulations, organizational analytics, integrations, or administrator views were in scope. A platform-wide label such as AI coaching cannot establish that all populations encountered the same data path or notice. Where the governing record is uncertain, the buyer should preserve the uncertainty rather than assign the latest page retroactively.

The accountable team should translate this point into a named workflow, affected population, source data, human owner, approval right, exception path, retained evidence, and review date. That translation is what separates an interesting AI development from a decision that can be governed and evaluated.

Separate notice wording from configured practice

Even when the applicable notice is known, it does not independently establish what the buyer enabled, what information entered the system, which roles could see it, how a model or human coach used it, what moved through integrations, or how retention and deletion operated. Those questions require current contract, configuration, access, data-flow, and service evidence. The same distinction applies to participant rights: a published description of access, correction, choice, or deletion does not show how a request is made, which records are covered, what the employer controls, what must be retained, or whether the process worked for a representative participant. The notice sets an evidence boundary; it is not an implementation certificate.

The accountable team should translate this point into a named workflow, affected population, source data, human owner, approval right, exception path, retained evidence, and review date. That translation is what separates an interesting AI development from a decision that can be governed and evaluated.

Make notice changes visible to people and governance owners

A material notice change should be reviewed against existing cohorts, contractual commitments, enabled features, employment-use boundaries, works-council or other consultation obligations, participant expectations, and the buyer's own communications. Governance evidence should distinguish a newly posted version from a delivered and applicable change, identify who assessed it, and show whether the buyer narrowed, paused, re-notified, or continued the affected service. The archive can support that review by preserving prior public versions, but it cannot show the buyer's decisions or participant experience. If the organization cannot connect a person and use to a governing version and configured practice, it should not claim that privacy treatment is settled.

The accountable team should translate this point into a named workflow, affected population, source data, human owner, approval right, exception path, retained evidence, and review date. That translation is what separates an interesting AI development from a decision that can be governed and evaluated.

Limitations and unknowns

BetterUp is the provider source. Its current AI privacy index identifies a May 2025 current notice, gives a May 9, 2025 update date, and links previous versions from September 2023, December 2023, and June 2024. The index does not independently establish which notice applied to a buyer, participant, cohort, region, date, product, or feature configuration, or establish contract precedence, delivery and acknowledgment, configured data flows, role visibility, retention, deletion, subprocessors, participant-rights operation, legal applicability, coaching quality, behavior change, or workforce outcome. Current contracts, data-processing agreements, cohort and enrollment records, delivered notices, configuration and data maps, access and request evidence, participant review, and qualified coaching, workforce, privacy, security, procurement, and legal review control.