Key takeaways
- Your agency stays the controller for client data put into an AI tool, so the supplier's terms, storage and retention are your responsibility to check.
- UK GDPR requires an impact assessment before processing that uses new technologies and is likely to result in a high risk to people[1].
- Significant decisions made solely by automated processing need safeguards, including the right to human intervention and to contest the decision[2].
- Security is part of the same check: 43% of UK businesses reported a breach or attack in the latest year, most often phishing[3].
Why an AI tool is a data protection decision
Almost every AI tool an agency might use touches personal data. Call answering and transcription capture voices, names, phone numbers and sometimes details about health or money. Listing tools may see a seller's reasons for moving. Lettings tools can handle identity documents, bank statements and references. As soon as that information goes into a tool, the agency is processing it, and the usual UK GDPR principles apply: a lawful basis, transparency, data minimisation, accuracy, limited retention and security.
In most arrangements the agency is the controller and the supplier is a processor acting on its instructions. That means the agency decides whether the processing is justified, must have a written contract with the supplier, and answers to clients and the ICO if something goes wrong. Free consumer tools rarely offer processor terms at all, which is why staff pasting client details into them is one of the most common gaps in agencies today.
The check, step by step
Map the data
List what personal data the tool will receive, from whom, where it is stored and who can see it. Include recordings, transcripts, summaries and any data used to improve the tool.
Choose the lawful basis
For most agency uses this will be contract or legitimate interests. Write down why, and do a short balancing test where you rely on legitimate interests.
Screen for a DPIA
Decide whether the processing is likely to result in a high risk to people. If it is, complete an impact assessment before going live.
Check the supplier
Get processor terms, the list of sub-processors, where data is held, how long it is kept and whether your data trains the supplier's models.
Set retention
Agree how long recordings, transcripts and uploaded documents are kept and make sure deletion actually happens in the tool and in any exports.
Update transparency
Change your privacy notice, call greeting and website wording so people know the tool is used and how to reach a person.
Handle automated decisions
If the tool makes or heavily shapes significant decisions about people, put human review and a challenge route in place.
Train staff and review
Tell staff what may be entered, who to ask and how to report a mistake. Review the assessment after three months of live use.
When an impact assessment is required
Article 35 of UK GDPR requires a data protection impact assessment where a type of processing, in particular using new technologies, is likely to result in a high risk to the rights and freedoms of individuals, and the assessment must be carried out before the processing starts[1]. AI tools are new technology almost by definition, so the real question is the level of risk.
- Likely high risk: tenant screening or scoring, tools that infer health, ethnicity or financial difficulty, recording and transcribing every call at scale, and monitoring staff performance automatically.
- Usually lower risk: drafting listing copy from a property fact sheet, summarising a negotiator's own notes, or routing calls without recording them.
- Always document the screening decision, even when the answer is that no full assessment is needed. That record is what the ICO will ask to see.
Questions for the supplier
| Question | A good answer | A red flag |
|---|---|---|
| Are you our processor, and will you sign processor terms? | Yes, with a standard data processing agreement | Terms of use only, no processing agreement |
| Where is our data stored and processed? | Named locations, with transfer safeguards for anything outside the UK | Not sure, or it depends |
| Is our data used to train your models? | No, or only with our opt-in | Yes by default, with no opt-out |
| How long do you keep recordings and transcripts? | A period we set, with deletion on request | Indefinitely |
| Who in your company can access our data? | Named roles, logged access | Any support engineer |
| How do you tell us about a breach? | Without undue delay, with a named contact | Not covered in the contract |
| Can we export and delete everything if we leave? | Yes, in a usable format | Export only on request, for a charge |
Automated decisions need extra safeguards
Where a significant decision about a person is based solely on automated processing, UK GDPR now requires safeguards. The controller must give the person information about the decision, let them make representations, let them obtain human intervention and let them contest the decision[2]. In agencies the obvious case is tenant selection: a system that rejects applicants automatically, or ranks them in a way staff simply follow, is close to that line.
The practical answer is to keep a person genuinely in the decision. That means staff who can see why a score was given, have the authority to overrule it and actually do so when the evidence supports it. A human who signs off every automated result without looking is not meaningful review. Our article on AI in tenant selection covers the discrimination risks in more depth.
Security sits inside the same check
A new tool is also a new place for data to leak. In the government's 2025/26 survey, 43% of UK businesses reported a cyber security breach or attack in the previous 12 months, and phishing was the most common, affecting 38%[3]. In the administration and real estate sector group, 83% of businesses said cyber security was a high priority[3].
- Turn on multi-factor authentication for every account that can see recordings, transcripts or documents.
- Give staff the least access they need, and remove leavers on their last day.
- Check how exports work: a transcript downloaded to a personal laptop is outside the supplier's controls.
- Ask how the supplier tests its own security and whether it holds a recognised certification.
Telling people about it
Transparency matters legally and commercially. In the government's tracker survey, 91% of UK adults said they were concerned about important decisions being made by computers without human input[4], yet 44% used an AI chatbot at least monthly[4]. People accept AI help when they understand it. Update your privacy notice to name the purpose, the type of supplier, the retention period and how to reach a person.
Sample privacy notice wording: 'We use automated tools to answer and record calls, create call summaries and log enquiries in our client system. Recordings and transcripts are kept for 12 months and then deleted unless they relate to an open complaint. A member of our team reviews summaries and makes all decisions about offers, tenancies and valuations. To speak to a person, ask at any point in the call.'
Marketing calls and retention traps
If a tool makes outbound calls, PECR applies. Under regulation 21, businesses must not make unsolicited direct marketing calls to numbers on the Telephone Preference Service register or to people who have told them not to call[5]. Canvassing past applicants about valuations is marketing, so screen the list first. Automated calls that play recorded marketing messages need the person's prior consent.
Retention needs care in the other direction too. Money laundering rules require customer due diligence and transaction records to be kept for five years from the end of the business relationship or the completion of an occasional transaction[6]. If an AI tool handles identity documents, make sure the records you must keep are exported to your own system before the tool's shorter deletion schedule removes them.
Worked example: a one-page assessment for call transcription
Example: a fictional three-branch agency plans to record and transcribe all inbound calls. Its screening note records that the processing is new technology, covers every caller and may capture sensitive details, so a full assessment is completed. The assessment lists the purpose (accurate enquiry records), lawful basis (legitimate interests, with a balancing test), data (voice, name, number, call content), supplier location (UK), retention (12 months) and access (branch managers and the operations lead).
Example, continued: the main risks identified are callers disclosing health or financial details and transcripts being exported to personal devices. The agency adds a greeting that tells callers about recording, disables transcript downloads for negotiators and sets a monthly check that deletion has run. The operations lead signs off and diarises a review after three months.
Frequently asked questions
Do estate agents need a DPIA for AI tools?
Only where the processing is likely to result in a high risk to people, but you should always record a screening decision. Tenant scoring, large-scale call recording and tools that infer sensitive details usually need a full assessment.
Can staff use free AI chat tools with client data?
Not safely. Free consumer tools rarely offer processor terms and may use what is entered to improve their models. Approve specific tools and tell staff what data may never be entered.
Do we need consent to record calls with AI?
Consent is not the only lawful basis; many agencies rely on legitimate interests. You must tell callers about the recording and its purpose, keep it only as long as needed and protect it properly.
What if an AI tool makes a decision about a tenant?
Significant decisions based solely on automated processing need safeguards, including information about the decision, the chance to make representations, human intervention and a way to contest it. Keeping a person genuinely in the decision is the simplest approach.
References
legislation.gov.uk (The National Archives) (2026). UK GDPR, Article 35: Data protection impact assessment (opens in a new tab) ↑
legislation.gov.uk (The National Archives) (2026). UK GDPR, Article 22C: Safeguards for automated decision-making (opens in a new tab) ↑
Department for Science, Innovation and Technology (GOV.UK) (2026). Cyber security breaches survey 2025/2026 (opens in a new tab) ↑
Department for Science, Innovation and Technology (GOV.UK) (2024). Public attitudes to data and AI: Tracker survey (Wave 4) report (opens in a new tab) ↑
legislation.gov.uk (The National Archives) (2019). The Privacy and Electronic Communications (EC Directive) Regulations 2003, regulation 21 (opens in a new tab) ↑
legislation.gov.uk (The National Archives) (2017). The Money Laundering Regulations 2017, regulation 40: Record-keeping (opens in a new tab) ↑