Hey Juno

What is PII in customer research, and how do you protect it?

By Hey Juno · Published · Updated

Personally identifiable information (PII) in customer research is anything in an interview that points to a real person, such as a name, an email address, a phone number or an account number. You protect it by collecting less, limiting who sees it, redacting it where you can, and deleting it when the study is finished. Redaction helps, but it does not make a transcript anonymous. Juno, the AI researcher from Hey Juno, does not redact transcripts today, so plan on the basis that yours contain personal information.

Qualitative interviews are more exposed than surveys, because free talk is full of identifying detail. The risk is not only what you ask for. It is what participants volunteer.

What counts as PII in qualitative research

Regulators define personal data broadly. The EU's General Data Protection Regulation (GDPR) covers any information relating to an identified or identifiable person, and says a person can be identifiable directly or through other details combined (GDPR, Article 4). The UK Information Commissioner's Office (ICO) and the Office of the Australian Information Commissioner (OAIC) take a similar view. Their own guidance is the place to check how your law defines it: the ICO's what is personal data and the OAIC's what is personal information.

In practice it helps to split PII into two groups.

  • Hard identifiers point at one person on their own: full name, home address, email, phone number, card number, bank account number, national ID or passport number.

  • Quasi-identifiers do not identify anyone alone but can when combined: employer, job title, suburb, school, a hospital, a rare condition, a date.

A transcript that says "I'm the only vet in a town of 900" has no name in it. It still identifies someone.

Where PII hides in an interview transcript

  • Names. The participant's own, and everyone they mention: a manager, a partner, a child, a named colleague.

  • Contact details. People offer an email or phone number to be "followed up with".

  • Account and card numbers. Often read aloud when someone is explaining a billing problem.

  • Addresses. Given when describing a delivery, a move or a local service.

  • Things said about other people. A participant describing a difficult relative or a former boss has put a third party in your data who never agreed to be there.

  • Sensitive detail. Health, finances, family circumstances and similar topics. Many privacy laws treat some of these as special categories that need more care.

Voice interviews add another layer. The recording itself is personal data, because a voice can identify someone.

How to protect it

1. Collect less

Data minimisation is a principle in the GDPR: collect what is adequate, relevant and limited to what the purpose needs (Article 5(1)(c)). Before you launch, ask whether the study needs the participant's full name, email or company at all. If you recruit through a panel, you may not need any contact details from the participant. Also check that your questions are not inviting disclosures you have no use for.

2. Decide who can see what

Not everyone on a project needs raw transcripts. Give raw access to the few people doing the analysis and the rest the findings and quotes. Use named accounts, not shared logins, and remove people when they leave.

3. De-identify where it helps

De-identification means removing or masking identifiers so the data is harder to link to a person. Terms get mixed up here.

  • Pseudonymisation replaces identifiers with labels, such as "Person 1", while a key or the original data still exists somewhere. The GDPR treats pseudonymised data as still personal data (Article 4(5)).

  • Anonymisation means nobody can reasonably re-identify the person, including you. Data that is truly anonymous falls outside the GDPR (Recital 26). This is a high bar. The ICO's anonymisation guidance explains how it judges it.

Most research transcripts are pseudonymised at best.

4. Set retention and deletion

Keep data only as long as the research needs it. Decide the period before the study starts, write it in your consent wording, and delete on schedule.

Tell participants plainly what is collected and how long it is kept. The ICC/ESOMAR International Code and the Market Research Society Code of Conduct set professional expectations for research. Check the current versions.

What redaction can and cannot do

Redaction finds identifying details in text and masks them. It is useful, and it is easy to oversell.

It can:

  • Remove the strongest re-identification keys, such as a full name, an email address or a card number, before a transcript is shared widely.

  • Keep the shape of the answer. A tag such as [EMAIL_1] tells you an email was mentioned without showing it.

It cannot:

  • Make a transcript anonymous. Employer, city and dates can still combine to point at someone.

  • Catch everything. Detection is imperfect, so spoken numbers, unusual names and misheard words can slip through.

  • Touch the original recording. Text redaction does not change audio.

  • Replace access control and retention. It is one layer.

Masking too much also strips out detail the analysis needs.

How Hey Juno handles it

Juno, the AI researcher from Hey Juno, interviews participants and returns findings traced back to quotes and transcripts. Here is what Hey Juno does about PII.

  • Redaction is not switched on today. Juno has built text redaction of hard identifiers: full names, addresses, identity numbers, financial numbers, email addresses and phone numbers. It is a single platform-wide setting, not something a customer or team can turn on or off, and it is not on in production. Your transcripts, quotes and exports are not redacted. Switching it on is planned, and so is redacting audio. Neither has a date.

  • Access is limited to your team. Transcripts, quotes, recordings and exports can be seen only by people in your organisation who can see the study.

  • Participants agree to recording. For voice interviews, participants agree to recording before the interview starts. Your team hears the original recording.

  • Other controls. Juno captures what the study needs and no more. Answers about health, finances and family get tighter handling. You set how long answers are kept, and can have them deleted. When a participant mentions someone else, Juno treats that person as context and does not profile them. For security status, see trust.heyjuno.co.

Until redaction is on, treat transcripts and exports as containing personal information. Collect less, and share findings rather than raw transcripts.

For participant wellbeing and what people are told, see participant wellbeing in AI research and Hey Juno's safety standards.

A practical checklist

  • Decide whether you need names or contact details at all.

  • Treat transcripts and exports as containing personal information, and share findings rather than raw transcripts beyond the core team.

  • Restrict raw transcript and audio access to named people.

  • Set a retention period before fieldwork starts, and diarise the deletion.

  • Never call transcripts anonymous, redacted or not.

  • Ask your own legal or privacy team which laws apply to your participants.

Where this doesn't apply

This guide is general information and is not legal advice. Laws differ by country, by participant location and by the kind of data. Research on children, health or other special categories may carry stricter duties. Check with your own legal or privacy team. Redaction, where it is used, does not cover audio or data you collect outside Juno.

Frequently asked questions

Is a redacted transcript anonymous?

No. Redaction masks hard identifiers, but employer, location and dates can remain, so a determined reader may still work out who someone is. Treat redacted transcripts as personal data.

What is the difference between pseudonymisation and anonymisation?

Pseudonymisation swaps identifiers for labels and keeps the means to link them back, so it is still personal data under the GDPR. Anonymisation means nobody can reasonably re-identify the person, which is much harder to achieve.

Does Juno redact transcripts or audio?

Not today. Juno has built text redaction of hard identifiers, but it is not switched on in production, so transcripts, quotes and exports are not redacted. Switching it on is planned, and so is redacting audio. Neither has a date. Until then, treat transcripts and exports as containing personal information.

Can my team turn PII redaction on?

No. Redaction is a single platform-wide setting, and a customer or team cannot switch it on or off. Access to transcripts, quotes, recordings and exports is limited to people in your organisation who can see the study.

Does GDPR apply to customer interviews?

It can, if participants are in the EU or UK or you operate there, but it depends on your situation. Read the regulation and your regulator's guidance, and check with your own legal or privacy team.