Automate the repeat admin around matters before the legal work itself: enquiry intake and triage first, then client status updates, then first drafts of routine letters, then time capture. Choose between them by scoring each task on weekly hours and risk. Leave legal research, contract review and anything filed with a court until the firm has a proven checking routine.
Most small firms get the order backwards. The first thing a partner tries is legal research, because it's the most impressive demonstration and the closest to their own expertise. It's also where general AI tools invent authorities and misstate rules, so the first experience is an embarrassing error and the project loses its sponsor. Admin comes first for a better reason than caution: it's where the hours are, where clients feel the difference, and where the firm learns the checking habit it will need later for legal work.
Stage 1: Log a week of repeat tasks (about 90 minutes in total)
For one normal week, everyone jots down each task they did more than twice, with a rough count and minutes. Fee earners and support staff both, because the admin is often split between them. Don't aim for precision; a tally on a sticky note is enough. Here is an excerpt from a support assistant's log (illustrative):
| Task | Times this week | Minutes each | Notes |
|---|---|---|---|
| New enquiry from website form: read, sort, reply, collect conflict details | 23 | 12 | Nine arrived after 6pm |
| "Any update on my purchase?" calls and emails | 31 | 6 | Mostly conveyancing clients |
| Client-care letters from template | 9 | 20 | Retyping matter details each time |
| Chasing clients for ID and forms | 14 | 5 | Some clients chased three times |
Add the fee earners' logs and you have a list of perhaps fifteen to twenty repeat tasks. That list is worth more than any vendor demo, because it describes your firm rather than an average one.
Expect the first logs to miss the smallest jobs. A disputes solicitor will often log "drafting witness statement, 6 hours" and little else, while the tiny repeats go unrecorded: filing each email to the right matter, forwarding documents to counsel, updating the diary. Ask everyone to add a line for anything under five minutes that happened more than ten times. In the illustrative firm below, that one request surfaced about 40 two-minute filing jobs a week for a single person, an hour and twenty minutes nobody had thought to count.
Stage 2: Score each task
For each task, multiply weekly hours by a risk weight. The risk weight reflects what happens if the automated version gets something wrong and nobody notices.
- Low risk (weight 1.0): errors are visible and cheap, such as a misfiled email or a clumsy acknowledgement.
- Medium risk (weight 0.6): errors reach clients but are correctable, such as a wrong date in a routine letter.
- High risk (weight 0.2): errors could amount to wrong advice, a missed limitation date or a misleading court document.
The weights are a judgement call, not a formula from anywhere official. Adjust them, but keep the high-risk weight low enough that legal work can't win on hours alone. Here's how it came out for a four-lawyer firm with two property solicitors, one private-client solicitor, one disputes solicitor and two support staff (illustrative):
| Task | Hours a week | Risk | Score | Rank |
|---|---|---|---|---|
| Enquiry intake and triage | 8 | Low to medium (0.8) | 6.4 | 1 |
| Client status updates | 6 | Low (1.0) | 6.0 | 2 |
| Routine letters from templates | 7.5 | Medium (0.6) | 4.5 | 3 |
| Reconstructing time records | 3 | Low (1.0) | 3.0 | 4 |
| Email sorting and summaries | 3 | Low (1.0) | 3.0 | 5 |
| Lease and contract review | 6 | High (0.2) | 1.2 | 6 |
| Legal research | 4 | High (0.2) | 0.8 | 7 |
Time capture ranks fourth on hours, but it has a twist: automating it doesn't only save time, it recovers billable time that was never recorded. If fee earners typically lose some of each day's short emails and calls, that makes it worth moving up. Many firms promote it to second or third for that reason.
The order also shifts with the practice. Run the same score for an illustrative two-solicitor family law practice. Enquiries are fewer and more delicate, since a clumsy automated acknowledgement to someone describing abuse at home does real harm, so intake takes a medium weight: 5 hours × 0.6 = 3.0. Status updates score 3 hours × 1.0 = 3.0. Time capture scores 4 hours × 1.0 = 4.0, and because so much family work is short calls and emails that never reach the time record, it goes first. Same method, different answer, which is why scoring your own list beats copying another firm's.
Stage 3: Check what your existing systems already do
Before buying anything, find out what the firm's current software offers. Case management systems increasingly include an AI assistant. Clio, for example, calls its built-in assistant Manage AI (it grew out of Clio Duo). It summarises matters from the data already stored there, and when a task is completed or a matter changes stage it can suggest a client update, drafted for a person to review and send. In April 2026 Clio also made its separate legal AI workspace, Clio Work, available to solo and small firms without a Clio Manage subscription. Microsoft 365 business plans include Copilot Chat at no extra cost, and the paid Copilot Business add-on (about $21 a user a month on annual billing) works across Outlook, Word and the firm's files.
Three questions for your case management supplier, by email, before you plan anything else:
- Which AI features are included in our current plan, and which cost extra?
- Does the AI use our matter data to train anything, and where is that written in our contract?
- Can the system send automated client updates when a matter reaches a stage, and can we edit the wording?
The answers often remove the need for a separate tool for tasks two and three on the list.
Vague replies are common, so read each answer against its question. An illustrative response to question 2: "Security is our top priority and our AI features meet industry standards." That says nothing about training. The follow-up that gets a straight answer is narrower: "Is our matter data used to train or improve any AI model, yours or a third party's? Please answer yes or no and point to the clause in our agreement or your data processing terms." A supplier that can't answer that in writing is telling you something useful.
Stage 4: Set up the first automation, in the usual order
1. Enquiry intake and triage
Enquiries arrive through the website form, email and phone messages, often out of hours. An AI step reads each one, identifies the likely matter type, pulls out the details needed for a conflict check, and drafts an acknowledgement. A person decides whether to take the matter. Tools range from the case management system's own intake features to an automation platform such as Zapier (Professional from $19.99 a month on annual billing for 750 tasks) passing form entries to an AI step. The full set-up, including out-of-hours qualification, is in AI client intake for law firms.
What the triage output looks like for one enquiry, and what the reviewer fixed (illustrative):
Enquiry summary
Matter type: Residential conveyancing (purchase)
Parties for conflict check: enquirer; neighbour at no. 14
Urgency: Medium
Key facts: Enquirer says neighbour has moved the fence 1m onto
their garden; wants advice before selling next year.
Suggested reply: standard conveyancing acknowledgementIt's a boundary dispute, not a purchase, and the sale is a year away, so it belongs with the disputes solicitor. The model latched onto the word "selling". The conflict-check details were right, which is the part that saves the most time. After two weeks of corrections like this, the prompt gained one line: "If the enquiry mentions a neighbour, boundary, fence or access, classify as property dispute."
2. Client status updates
Conveyancing and probate clients ask "where are we?" constantly, and each answer means opening the file. An AI step can draft an update from the matter's latest stage and file notes, for a person to send. Before and after, for a purchase:
Before (typed from memory between calls):
"Hi, still waiting on searches, should be soon. Will let you know."
After (drafted from the file, checked, sent):
"Your property searches were ordered on 2 September. Two of the
three are back and show nothing unusual. The last one usually takes
three to five weeks, so we expect it by early October.
Once it arrives we'll review it and send you a summary within two
working days. Nothing is needed from you at the moment."The second takes the same time to send and prevents the next call. The method is in how small law firms use AI to answer client status questions.
3. Routine letters and documents
Client-care letters, completion letters, standard requests to the other side: first drafts from templates plus matter details, reviewed by the fee earner. Keep the firm's approved template as the base and let AI fill and adapt, rather than asking it to write letters from nothing. See how small law firms use AI to draft letters and routine documents.
What goes wrong is usually the wording your regulator or insurer expects to stay fixed. A prompt that protects it:
Fill this client-care letter template using the matter details below.
- Paragraphs marked [LOCKED] must be copied exactly, word for word.
- Fill every [field] from the matter details. If a detail is missing,
leave the field in square brackets and list it at the end.
- You may adapt the "Scope of work" paragraph to this matter, but do
not add any service the matter details don't mention.
Template: [paste]
Matter details: [paste]In an illustrative first run for a commercial lease renewal, the draft filled the client names, property and fixed fee correctly, then "tidied" the locked complaints paragraph into two friendlier sentences and added "advice on the tax position" to the scope, which the matter didn't include. The fee earner caught both by running Word's Compare feature against the template rather than reading the letter through, which takes seconds and shows every changed word in the locked sections.
4. Time capture
AI tools can suggest time entries from sent emails, calendar events and documents edited, for the fee earner to approve before logging off. A realistic first-week result: the tool proposes a 0.3-hour entry for "reviewing lease" on a matter where the solicitor only opened the document to find a phone number. The fee earner deletes it. Approving suggestions, rather than letting them post automatically, is what keeps the time record honest. AI time capture for solicitors covers the tools and settings.
5. Email sorting and summaries
Summaries of long threads and a suggested next action for each email. It's low risk and saves time for everyone, which is why it often spreads on its own once the first automation is running.
One rule matters more than the others here: summaries should quote deadlines, never calculate them. In an illustrative thread, the other side's letter required a response "within 14 days of the date of this letter", and the summary helpfully turned that into a calendar date, counting from the day the email arrived rather than from the letter's date three days earlier. Tell the model to copy deadline wording exactly under a heading "Deadline to diarise", and leave the date to the person who owns the diary.
Stage 5: Run it in shadow mode for two weeks
Whichever task comes first, run the automated version alongside the old one before clients see it. For intake, the AI triages every enquiry but the support assistant still does her own triage, and they compare at the end of each day. Track three things: how often the AI's classification was wrong, how much time the draft saved, and anything it did that a client would have disliked. The general method is in how to pilot AI in shadow mode before customers see it.
A realistic mistake from this stage, and a good reason for it: in one test, the drafted out-of-hours acknowledgement promised that "a solicitor will call you back within two hours". The enquiry came in at 11pm. Nobody had told the model the firm's hours, so it offered a service the firm doesn't provide. The fix was two lines in the instructions: office hours, and "never promise a call-back time". Shadow mode caught it before any enquirer did.
Stage 6: Decide, then pick the next one
At the end of two weeks, go live if three conditions hold: errors are rare and caught by the review step, the time saving shows up in the log, and the person doing the task prefers the new way. If one fails, fix the prompt or the settings and run another week. If two fail, drop it and move to the next task on the list. Only start the second automation once the first has run without surprises for a month.
For the intake pilot in the four-lawyer firm, the two-week tally (illustrative) read: 41 enquiries; 5 misclassified (three boundary disputes filed as conveyancing before the prompt fix, one enquiry about managing a parent's affairs filed as a will, one sales email treated as an enquiry); all 5 caught at review; average handling down from 12 minutes to about 5; and one acknowledgement the assistant rewrote because it sounded cold to a bereaved enquirer. Errors rare and caught, saving visible, the assistant in favour: all three conditions met. It went live with a line on tone added to the instructions.
In the four-lawyer firm above, the likely first quarter looks like this: intake in month one, status updates in month two (using the case management system's own feature), and routine letters in month three. Time capture follows in month four.
What waits until later, and why
- Legal research. Useful with a legal research tool that cites real sources, and even then every authority is checked at source. Not a first project.
- Contract and lease review. Good at spotting missing or unusual clauses, poor at judging what matters for this client's deal. Worth trying once the firm has a review habit, with the tool treated as a first pass.
- Anything filed with a court. The consequences of an invented citation or a wrong fact are too serious for early experiments.
- Advice to clients. AI can help a solicitor write advice more clearly. It shouldn't decide what the advice is.
The full list, with reasons, is in which legal tasks a small firm should never hand to AI. Most firms that start with admin find they reach the legal work within six months, with a checking routine already in place and a team that trusts the tools because it has seen them corrected.
Further reads
- AI Implementation Plan for a Small Law Firm: The First 90 Days — Turn the first automation into a 90-day plan for the firm.
- AI Email Triage for Professional Firms: Sort, Summarise, Draft — The fifth job on the list, set up properly.
- Per-Seat or Pay-As-You-Go? Legal AI Pricing for Small Firms — How legal AI tools charge before you commit to one.
- AI Contract Review for Small Firms: What It Catches and Misses — What contract review can do once the admin is running.
- Can Solicitors Use ChatGPT Without Breaching Confidentiality? — Settle the confidentiality rules before any matter data goes in.
- What an AI Consultant Does for a Law Firm, and What It Costs — What outside help with this costs, if you want it.
- ChatGPT or a Legal AI Tool: Which Should a Small Firm Use? — Which legal tasks a business ChatGPT plan handles safely, which need a legal research tool, and how a three-lawyer firm split the budget.
- AI Readiness Checklist for Accountants, Solicitors, Consultants — Twenty checks, grouped and scored, that tell an accountancy, law or consulting firm whether it is ready to pilot AI or has gaps to fix first.
- How Conveyancers Use AI to Cut Admin on Each Transaction — Stage-by-stage admin savings for a conveyancing file, a title-summary prompt with sample output, chaser wording and the tasks AI must never touch.
- AI Marketing for Small Law Firms: Content, Reviews and the Rules — How a small law firm can use AI for guides, posts and review replies while staying inside platform, consumer-law and professional conduct rules.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: Zapier pricing page; Microsoft 365 Copilot Business pricing; Clio help pages on Manage AI (matter summaries, drafted client communications) and Clio's April 2026 announcement on Clio Work availability. Firm examples and scores are illustrative.