Key takeaways
- A DPIA is required before processing likely to result in a high risk, particularly with new technologies, and always for large-scale special category or criminal offence data[1].
- It must describe the processing, assess necessity and proportionality, assess the risks and set out the measures to address them[1].
- Public concern is real evidence of risk: 56% of legal service users named AI acting without their knowledge or consent as their most urgent AI concern[2].
- Human review is the norm, not the exception: 84% of businesses using AI apply some checking to its outputs[3].
- A good DPIA for a small firm can fit on four pages; the discipline matters more than the length.
When you need one
Article 35 requires a data protection impact assessment before any processing that is likely to result in a high risk to people's rights and freedoms, in particular where new technologies are involved[1]. It is always required for three kinds of processing: systematic and extensive evaluation of people based on automated processing, including profiling, that produces significant effects; large-scale processing of special category or criminal offence data; and systematic monitoring of a publicly accessible area on a large scale[1].
Law firms meet these triggers more often than they think, because so much of their data is sensitive. Projects that usually warrant a DPIA include:
- Migrating to a new case or document management system that will hold health, family or criminal files.
- Introducing call recording, transcription or AI call handling on client lines.
- Using AI to review documents, summarise files or draft advice from client material.
- Adopting identity verification that uses facial matching, which is biometric data.
- Monitoring staff activity on devices, email or the network.
- Installing CCTV covering reception or public areas outside the office.
- Any new use of automated decisions about enquiries, such as automatic rejection or prioritisation.
If you are not sure, do a two-page screening assessment and record the answer. Deciding that a DPIA was not needed is itself a decision you should be able to show.
Screening questions before you start
Most firms do not need a full DPIA for every software change, but they do need a consistent way of deciding. A one-page screening form, completed by the person proposing the change and checked by the data protection lead, catches the projects that matter. If the answer to any of these questions is yes, carry out a full assessment:
- Will the tool process health, criminal, family, immigration or other sensitive information about clients or third parties?
- Will it make or shape decisions about people, such as whether to take on an enquiry, without a person reviewing each outcome?
- Does it use new technology the firm has not used before, such as AI models, voice analysis or biometrics?
- Will it record, monitor or track people, including staff, callers or website visitors?
- Will data be combined from different sources, for example linking call records with case files and marketing data?
- Will data be stored or accessed outside the UK?
- Could vulnerable people, such as children, victims of abuse or people in police custody, be affected?
- Would a client be surprised to learn their information was being used in this way?
File the screening form even when the answer is no. It shows the regulator, a client or an insurer that the firm considered the question at the right time.
Where DPIAs usually fall short
- They describe the supplier's product rather than the firm's actual use of it, so configuration choices that create or reduce risk are never assessed.
- Risks are listed but never scored, which makes it impossible to see whether the measures changed anything.
- Measures depend on staff behaviour, such as always checking AI summaries, without any way of confirming it happens.
- Nobody signs off the residual risk, so accountability is unclear when something goes wrong.
- The document is never reviewed after launch, even when the supplier changes the product significantly.
Fix these and even a short DPIA becomes a working management tool rather than a compliance artefact. For the security measures that sit underneath most assessments, see cyber security basics for small law firms.
What the law says it must contain
Article 35(7) sets the minimum content: a systematic description of the processing and its purposes; an assessment of necessity and proportionality in relation to those purposes; an assessment of the risks to individuals; and the measures to address those risks, including safeguards and security measures[1]. If you have a DPO, you must seek their advice[1]. If the assessment shows a high risk you cannot reduce, you must consult the regulator before starting.
A short template
1. The project in plain words
What is the tool, who supplies it, which teams will use it, and what problem does it solve? Attach the supplier's product description.
2. The data flow
What personal data goes in, whose it is, where it is stored, who can access it, which other suppliers touch it, whether it leaves the UK and how long it is kept.
3. Lawful basis and conditions
The Article 6 basis, any Article 9 or Schedule 1 condition, and any PECR requirement, such as cookies or marketing.
4. Necessity and proportionality
Why this tool and this data are needed for the purpose, what less intrusive options you considered, and what you will not collect.
5. Risks to people
For each risk, likelihood and severity: confidentiality loss, inaccurate outputs, unfair decisions, loss of access, unexpected reuse, and distress for vulnerable people.
6. Measures
Contract terms, configuration, access controls, human review points, staff guidance, testing and how people can raise concerns or challenge outcomes.
7. Residual risk and sign-off
The risk remaining after measures, the DPO's or data protection lead's advice, the partner who accepts the risk and the review date.
Using evidence on public attitudes
A DPIA asks you to assess risk from the point of view of the people affected. For AI tools, there is now good evidence about how clients see it. In the Legal Services Board's public panel research, AI acting without the user's knowledge or consent was the single most urgent AI concern for legal service users, at 56%[2], and 81% agreed that AI legal tools should never act automatically for a vulnerable person in high-stakes situations[4]. Those findings point directly to the measures a DPIA should test: transparency at the point of contact and a human in the loop for consequential steps.
Businesses are already moving that way. Among businesses using AI, 84% apply at least some human input or checking to outputs and 67% apply significant checking[3]. Under the Data (Use and Access) Act 2025, significant solely automated decisions require safeguards including the ability to challenge decisions and obtain human intervention[5]. Building those safeguards in at the design stage is far cheaper than retrofitting them.
Adoption barriers tell a similar story. Ethical concerns, high costs and unclear regulation were rated the most significant barriers to AI adoption by businesses[6]. A proportionate DPIA addresses the first and third directly, and often shows that the second is lower than feared once scope is controlled.
A worked example: AI call answering
Example: a fictional six-partner firm wants out-of-hours calls answered by an AI Employee such as Ava, which follows the firm's approved intake, routes urgent matters to the duty solicitor and books initial consultations. The firm's data protection lead runs the template.
| Risk | Likelihood | Severity | Measures |
|---|---|---|---|
| Caller does not realise they are not speaking to a person | Medium | Medium | Clear disclosure at the start of every call; option to leave a message for a person |
| Sensitive details recorded beyond what is needed | High | High | Intake script collects outline only; retention limit on recordings and notes |
| Urgent matter, such as a police station call, not escalated | Low | High | Urgency rules tested monthly; escalation to a named duty solicitor; daily review of notes |
| Inaccurate summary leads to wrong conflict decision | Medium | Medium | Fee earner reviews every note before acting; conflicts checked by staff |
| Call data reused by supplier | Low | High | Contract prohibits secondary use; confirmed in writing |
| Data stored or accessed abroad without safeguard | Low | Medium | Supplier locations recorded; transfer safeguard checked |
The residual risk is assessed as low to medium, the partners sign off with a three-month review, and the privacy notice and call greeting are updated before launch.
Keeping it useful after launch
- Review the DPIA at the date you set, and whenever the supplier changes the product, the data, the sub-processors or the location.
- Feed incidents and complaints back into the risk assessment.
- Keep each version, so you can show what you knew and decided at the time.
- Share a short summary with the teams using the tool, so the measures are actually followed.
- Store DPIAs with the supplier record and the data processing agreement.
A DPIA is not a document you write to justify a decision already made. If the assessment shows the risk is too high, the right outcome is to change the design, choose a different tool or not proceed.
For tools that send data abroad, read our guide to international transfers. For the confidentiality questions AI tools raise, see confidentiality versus legal professional privilege.
Frequently asked questions
Do small firms really need DPIAs?
The duty depends on the risk of the processing, not the size of the firm. A sole practitioner adopting AI to process crime or family files should still assess the risk, though the DPIA can be short.
Can we rely on the supplier's DPIA?
You can use the supplier's information, but the DPIA is yours as controller. It must reflect your data, your clients, your configuration and your measures.
When must we consult the regulator?
When the DPIA shows a high risk that you cannot reduce to an acceptable level. Most firm projects can be redesigned to avoid this.
Is a DPIA needed for a new case management system?
Usually, if it will hold special category or criminal offence data at scale, which most law firm systems do. It is also the best time to fix access controls and retention.
References
legislation.gov.uk (The National Archives) (2016). UK GDPR Article 35: Data protection impact assessment (opens in a new tab) ↑
Legal Services Board (2026). AI in Legal Services: Research Report (LSB Public Panel) (opens in a new tab) ↑
Department for Science, Innovation and Technology (GOV.UK) (2026). Artificial Intelligence Adoption Research (human oversight) (opens in a new tab) ↑
Legal Services Board (2026). AI in Legal Services: Research Report (LSB Public Panel), vulnerable users (opens in a new tab) ↑
GOV.UK (DSIT) (2025). Data (Use and Access) Act 2025: data protection and privacy changes (opens in a new tab) ↑
Department for Science, Innovation and Technology (GOV.UK) (2026). Artificial Intelligence Adoption Research (barriers) (opens in a new tab) ↑