Data privacy in synthetic market research: what to check before a study

A research team wants to explore a new offer using past interviews and customer information. The material is useful because it contains context: what people bought, the problems they described and the circumstances around their choices. Some of that context may also identify them.

Using a simulated audience changes how a study gathers reactions. It does not, by itself, decide whether the information supplied to the study is appropriate to share or how it should be handled. A practical review starts with the material the project needs, where it will go and who will use the result.

This guide explains the questions to resolve and what our public policy says about Sapien. The answers for a particular project should be reflected in the arrangements agreed for that work.

Start with the information the question needs

Imagine Vale, a fictional homewares retailer comparing two delivery offers. Its team has purchase summaries, support transcripts and a spreadsheet linking orders to customer names and addresses. It wants to understand whether a more predictable delivery window would make an offer more useful to different types of buyer.

The study may need context about delivery constraints, order frequency and the alternatives buyers use. It may have no need for names, phone numbers, full addresses or individual order identifiers. Starting with the decision makes it easier to distinguish useful context from information included simply because it was in the original export.

Prepare a version of the material suited to the work. An aggregate purchase summary and reviewed excerpts might answer the question more directly than a full customer database. Read free text as well as column headings: a transcript can contain a name, an unusual incident or a sensitive personal detail even after the obvious contact fields have been removed.

Check that your organization is permitted to use and share the source for the proposed purpose. An agency receiving a client file should establish that permission with the client. Access to a document does not settle the terms on which another service may process it.

Separate the website relationship from study data

Our privacy policy, effective October 13, 2026, identifies Sapien as a trade name of Outrove Inc. It covers account information, information submitted through forms and communications, technical and usage information, and business contact information supplied by our visitor-identification provider.

For that policy data, Outrove is the controller. The policy separately states that when customers upload information about their own respondents or contacts, Sapien processes it on their behalf as a processor, while the customer is the controller.

That distinction matters when preparing a project. A person requesting a demo has a different relationship to the service from someone whose information appears in a customer's research file. For Vale, the review should cover the actual customer material proposed for the study, the purpose of processing it and the people authorized to provide it.

Agree the project arrangements before transferring that material. Where a data-processing agreement or a client approval is required, a general website notice does not replace it. The relevant documents should describe the work being done and the responsibilities attached to it.

“Synthetic” does not mean every input is anonymous

A simulated response is different from a response collected from a recruited person. The information used to ground or contextualize that simulation can still concern real people. Removing a name does not necessarily remove the possibility of identification through other details.

The UK Information Commissioner's Office explains that effective anonymisation depends on people no longer being identifiable, including through information combined with other sources. Its guidance distinguishes this from pseudonymisation, where personal data remains personal data. These are context-dependent judgments; the ICO also flags its current anonymisation guidance as under review following legislative changes.

For Vale, replacing names with customer numbers would not resolve an excerpt that describes a recognizable person and delivery incident. Review the whole material and the way it will be used. If the question can be answered with less identifying detail, prepare that smaller source set rather than treating the word “synthetic” as an exemption from the review.

Ask about the actual processing path

The useful security questions concern the proposed project: who can access its inputs and outputs, which services process them, where that processing occurs, and what happens when the work ends. Connect each answer to the data and workflow your team intends to use.

Our policy identifies service providers used to operate Sapien, including Google, Vercel, PostHog, RB2B, scheduling and email providers. It describes PostHog site analytics and privacy-protected session recordings, with form and input values masked, and RB2B business-visitor identification. Those descriptions should not be read as a complete technical map of how a particular research file is processed.

If your procurement decision depends on a particular control, request evidence for that control and its scope. Examples include the handling of uploaded research by model providers, access restrictions, security assurance reports and any required regional processing arrangement. A claim about one provider or part of a service should not be extended to the entire workflow.

Our public policy states that we use reasonable safeguards. It does not establish a security certification or specify an encryption architecture. Its statement that we do not sell personal information also answers a different question from whether a particular input may be retained or used by a model provider. Resolve the question your project actually depends on in the applicable terms and supporting documentation.

Make retention and location concrete

The policy says we retain information for as long as needed to provide the service, maintain legitimate business records and comply with law. That is a retention principle, rather than a fixed deletion date for every research artifact.

For a study, agree which inputs, working copies and outputs need to remain available and for how long. If the project requires deletion by a particular date, establish the applicable arrangement, including the treatment of backups and legally required records. The same review should cover copies retained by your own team and anyone receiving the results.

Sapien is based in the United States, and the policy says service providers may process data in the US and other countries. It describes safeguards for relevant international transfers, including Standard Contractual Clauses and the UK Addendum in providers' processing terms, or another lawful transfer mechanism. A team with a regional hosting requirement should resolve that requirement specifically before sharing data.

Carry the review through to the report

A report can reintroduce information the preparation stage tried to remove. A memorable quote or a narrowly defined group may make an individual recognizable even when the main results are aggregated. Decide who needs the detailed material and what belongs in a wider presentation.

Keep modeled reactions identifiable as modeled reactions. A synthetic interview excerpt should not appear as a verbatim statement from a real customer. Equally, a real customer quote included as background should retain its source and permitted use. That separation helps readers understand both the evidence and the information they are handling.

Vale's useful outcome is a study with enough context to compare the delivery offers, supported by an agreed treatment of its source material. The review should make the research usable, with clear responsibilities for the information it actually needs.

For questions about Sapien's privacy policy or a privacy request, contact saif@asksapien.ai. Bring the proposed data types, intended use and any procurement requirements so the discussion can address the actual project.