Key takeaways
- Admin is the lowest-risk place to start with AI: errors are easier to catch and rarely reach a client as advice.
- Most UK businesses using AI to improve operations report no change in headcount; the gain is time, not redundancy.
- Good first uses are transcription, internal meeting notes, email sorting, file opening data and time capture.
- Keep people in charge of client money, complaints, decisions about clients and decisions about staff.
Start where the risk is lowest
Much of the debate about AI in law firms focuses on legal research and drafting advice, where errors are serious and checking is demanding. Practice management and administration are a better place to start. The tasks are repetitive, the outputs are easy to check, and a mistake in a diary entry or a file-opening form is far less harmful than a mistake in advice. They also take up a large share of support staff time, which is where many small firms feel most stretched.
Starting with admin has another advantage: it builds the habits the firm will need for higher-risk uses later. Staff learn to use approved tools, to check outputs and to report problems, and the firm learns how its suppliers behave, before anything important depends on them.
What happens to jobs
The fear that AI means immediate job cuts is not borne out by the data so far. Of UK businesses using AI to improve business operations, 63% reported no change in headcount, 6% a decrease and 1% an increase[1]. Only 10% of businesses using AI say they use it extensively[1].
Source: Office for National Statistics, 2026[1]
Show the data
| Item | Value |
|---|---|
| No change | 63% |
| Decrease | 6% |
| Increase | 1% |
For a law firm, the practical aim is to give support staff time back for work that needs a person: looking after clients on the phone, chasing third parties, preparing bundles and keeping files in order. Say this clearly to the team at the start. People who believe a tool is there to replace them will not help you make it work.
Where businesses are aiming AI
Across the economy, customer service is the most common target. Among UK businesses using or planning to use AI, 46% use or plan it for customer service, against 16% for legal functions[2]. In a law firm, customer service means enquiries, appointment booking, progress updates and the many small questions clients ask between meetings. Those are also tasks where getting a person on the line quickly matters, so any automation should make that easier, not harder.
An admin task inventory
| Task | Usually done by | AI fit | Human check |
|---|---|---|---|
| Transcribing attendance notes and dictation | Secretaries | Good | Fee earner reads and corrects before filing |
| Summarising internal meetings | Whoever chairs | Good | Chair confirms actions before circulating |
| Sorting and routing incoming email | Reception, secretaries | Good | Spot-check misrouted messages daily |
| Extracting details for file opening | Secretaries, accounts | Good | Check names, addresses and dates against source |
| Suggesting time entries from activity | Fee earners | Moderate | Fee earner approves every entry |
| Drafting routine admin letters and reminders | Secretaries | Good | Sender reads before sending |
| Answering staff questions on internal policies | Practice manager | Moderate | Answers drawn only from current policies |
| Taking enquiry details out of hours | Nobody, or voicemail | Good with clear limits | Team reviews every note next morning |
| Handling client money or bank detail changes | Accounts | Not suitable | Always a person, with call-back checks |
Valuing support time properly
Admin time is cheaper per hour than fee-earner time, but it is not cheap in total and it is hard to recruit. Median gross hourly pay in 2026 was £13.51 for receptionists and £14.95 for legal secretaries[3]. The bigger value is often indirect: when secretaries are not typing up dictation, they can chase the search that is holding up an exchange or call a client who has gone quiet.
Example: a fictional six-fee-earner firm estimates its two secretaries spend about eight hours a week between them typing attendance notes. A transcription tool cuts that to about three hours of checking and correcting, freeing around five hours a week for client contact and file management.
Five low-risk first uses
Transcription. Speech-to-text tools have become accurate enough for routine attendance notes, especially when the fee earner dictates clearly. The fee earner must read the result before it goes on the file, because a misheard figure or name can matter. Decide how long audio is kept, and delete it once the note is checked if you do not need it.
Internal meeting notes. Summaries of partners' meetings, team huddles and training sessions are a safe place to learn what these tools do well and badly. Keep client matters out of recorded meetings unless the tool is approved for client data.
Email sorting. Rules and AI classification can route messages to the right team, flag likely completions or court deadlines and pull out attachments. Check misrouted messages daily for the first month, then weekly.
File opening. Extracting names, addresses and key dates from an instruction email or form into the file-opening screen saves retyping. The person opening the file still checks each field, especially spelling of names used in conflict searches.
Time capture. Tools that suggest time entries from emails, calls and documents help fee earners who forget to record short tasks. Every suggestion must be approved by the fee earner; nothing should post to a bill automatically.
Check what you already pay for
Before buying anything new, look at the tools you already license. Many case management, dictation and office platforms now include transcription, summarising or classification features, sometimes switched off by default. Using them can be simpler than adding a new supplier, because the data already sits under an existing contract. It still needs a check: confirm whether the feature sends data to another provider, whether inputs are used to improve models, and whether an administrator can control who uses it.
Bringing the team with you
Admin tools succeed or fail on the goodwill of the people who use them. Involve support staff in choosing the first tasks, because they know which jobs are tedious and which only look tedious from the outside. Agree in advance what the freed time will be used for, and say so openly.
- Ask support staff to list the three tasks they would most like help with.
- Let one or two volunteers try the tool first and report back to the team.
- Make it safe to say a tool is not working, and drop it if it is not.
- Recognise the new work people take on with the time saved.
Keep these with people
- Anything involving client money, payments or changes to bank details.
- Responses to complaints, which need judgement, empathy and accountability.
- Decisions about whether to take on a client or matter.
- Decisions about staff, such as recruitment, performance and absence.
- Conversations with vulnerable or distressed callers, beyond taking basic details and passing them on urgently.
Confidentiality applies to admin too. The Law Society's generative AI primer flags the risk of misuse or disclosure of confidential information[4]. A dictated attendance note is as confidential as the advice letter that follows it, so admin tools need the same contract and settings checks as any other.
A worked workflow: file opening
Receive the instruction
The instruction arrives by email or web form with the client's details and a brief description.
Extract
An approved tool pulls names, addresses, contact details, other parties and key dates into a draft file-opening form.
Check
The secretary compares every field with the source and corrects any errors, paying particular attention to names and spellings.
Run conflict and ID steps
The usual conflict search and identity checks run on the corrected details, exactly as before.
Open and record
The matter is opened, and the form records that AI extraction was used and who checked it.
Example: a small team's first month
Example (fictional): a firm with four fee earners and two support staff introduces transcription and email sorting in the same month. Transcription works well for three fee earners but poorly for one who dictates in a noisy car; he switches to dictating at his desk. Email sorting misroutes a handful of messages in the first week, mostly between two teams with similar names, and the rules are adjusted. After a month, the support team estimates it has gained most of a day a week, and uses it to start a proactive weekly update call for conveyancing clients.
Measuring results
- Hours per week spent on the task before and after, estimated by the people doing it.
- Errors caught at the checking stage, and any that got through.
- What the freed time was used for.
- Staff views after one month and three months.
Once admin uses are working, look at enquiry handling. Our guides to AI receptionists for law firms and AI receptionist vs human receptionist cover the options. For the wider tool landscape, see AI tools for legal research and drafting compared, and for what to tell clients, telling clients when you use AI.
Frequently asked questions
What admin tasks can AI do in a law firm?
Good starting points are transcription of attendance notes, internal meeting summaries, email sorting, extracting file-opening details and suggesting time entries, each with a person checking the output.
Will AI replace legal secretaries and receptionists?
The evidence so far suggests not, at least in the short term. Most UK businesses using AI to improve operations report no change in headcount. The usual gain is time freed for client contact and file work.
Is AI transcription safe for client attendance notes?
It can be, using an approved tool under a firm contract, with the fee earner checking every note and a clear rule on how long audio is kept.
Which admin tasks should stay with people?
Anything involving client money or bank details, complaints responses, decisions about taking on clients, staff decisions, and substantive conversations with vulnerable or distressed callers.
References
Office for National Statistics (2026). Artificial intelligence in UK businesses: 2023 to 2026 (opens in a new tab) ↑
Department for Science, Innovation and Technology (GOV.UK) (2026). Artificial Intelligence Adoption Research (opens in a new tab) ↑
Office for National Statistics (2026). Earnings and hours worked, occupation by four-digit SOC: ASHE Table 14 (2026 provisional) (opens in a new tab) ↑
The Law Society (2026). Generative AI: the essentials (opens in a new tab) ↑