An AI consultant for a law firm works out where fee-earner time goes, picks the few tasks worth handing to AI, sets confidentiality and supervision rules, configures the tools (often ones the firm already licenses) and hands over checking routines staff can run. Cost is consultant days times a day rate, so ask for fixed fees per phase.
The figure most firms miss is their own time. A consultant can't find the right tasks without interviewing the people doing them, and partner hours are the most expensive hours in the building. And some small firms don't need a consultant at all: if you're under about five people and your practice-management software already includes AI features, you may get most of the value by switching them on carefully yourself. Consultant or DIY for a small professional firm sets out that choice.
The work, phase by phase
A sound engagement for a firm of five to fifteen people usually runs through five phases. The day counts below are a rough guide for comparing proposals, not a price list; a firm with one practice area and tidy systems sits at the low end.
1. Discovery: where the hours go (roughly 2–4 consultant days)
Interviews with partners, fee earners and support staff; a look at a handful of matters from opening to closing; time-recording data if you have it. The output is a ranked list of tasks, each with hours per month, risk level and whether AI can do it safely.
Take a family law practice as an illustration: interviews show paralegals spend around 40 hours a month building chronologies from client emails and disclosure, and partners spend 12 hours a month rewriting routine update letters. Both rank high. Drafting financial remedy arguments ranks low: high risk, low volume, and exactly where judgement lives. The top of the ranked list the consultant hands over might read:
| Task | Hours a month | Risk | Recommendation |
|---|---|---|---|
| Chronologies from emails and disclosure | 40 | Medium: dates and sources must be checked | AI first draft, paralegal checks every entry against its source |
| Routine client update letters | 12 | Low to medium: tone and hearing dates | AI draft from matter notes, partner approves |
| First-call enquiry notes | 9 | Low: internal only | Transcribe and summarise with consent |
| Financial remedy arguments | 6 | High | Leave with lawyers |
A list like this is the discovery phase's whole value. If the partners look at it and say "that's not where our time goes", the interviews missed something, and it's cheaper to find out now than after configuration.
2. Risk and rules (roughly 1–2 days)
Approved tools and plans, what client material may go into them, how AI use is supervised and recorded, and the wording for engagement letters. The consultant should work from your regulator's or bar association's guidance on AI, not a generic template. A filled-in clause from a policy of this kind might read:
Client documents may be processed only in the firm's approved AI tools (currently Claude Team and the AI features of our practice-management system), signed in with a firm account. Any AI-assisted research must have every authority opened and read by a fee earner before it is relied on. AI drafts sent to clients or third parties must be reviewed and approved by the supervising lawyer, who remains responsible for them.
3. Tool choice (roughly 1–3 days)
For each ranked task: can a tool the firm already has do it, does it need a general AI plan, or does it need a legal-specific tool with a research database behind it? A good consultant tests two or three options on your own anonymised material and shows you the results side by side, rather than recommending from a slide.
A results sheet from that kind of test might look like this (illustrative, five anonymised client update letters per tool, scored by the supervising partner):
| Option | Usable with light edits | Needed rewriting | Factual errors introduced | Cost |
|---|---|---|---|---|
| AI features in the practice-management system | 3 of 5 | 2 | 0 | Add-on quote |
| Claude Team with a precedents project | 4 of 5 | 1 | 1 (a wrong hearing date copied from an old letter) | $20 per seat a month, annual |
| Legal drafting tool on trial | 4 of 5 | 1 | 0 | Quote, annual seats |
The interesting line is the error. The hearing date came from an old precedent letter left in the project, not from the model's imagination, which points to a fix (clean the precedents) rather than a reason to reject the tool. A consultant who reports errors like this, with causes, is doing the job properly.
4. Configuration (roughly 3–10 days)
The practical work: setting up shared projects with firm precedents, connecting the AI features to your document management or practice-management system, writing prompt templates, building any automations such as intake triage. Here's a before-and-after for an out-of-hours enquiry process (illustrative):
- Before: web-form enquiries land in a shared inbox; the receptionist reads them at 9am, forwards them to whoever seems relevant, and the fee earner phones back by early afternoon, sometimes after the prospect has instructed another firm.
- After: each enquiry is classified by practice area and urgency within minutes, a conflict-check request is created in the practice-management system, the prospect gets an acknowledgement saying when they'll hear from a named lawyer, and urgent matters (an injunction, a hearing this week) go to the duty partner's phone. Nothing that could be advice is sent automatically.
The design choices in that workflow are covered in AI client intake for law firms.
5. Handover (roughly 1–2 days)
Written instructions for each workflow, a named owner in the firm, a checking routine, an error log, and a review date. If the firm can't run everything without calling the consultant, the handover isn't finished. One runbook entry, filled in for the update-letter workflow (illustrative):
- Workflow: routine client update letters, family matters.
- Owner: the senior associate; cover when she's away: the practice manager.
- Steps: open the matter's latest attendance note, run the "update letter" prompt in the firm's Claude Team project, check every date and figure against the file, send for partner approval.
- Checks: hearing dates against the court diary; no statement of the client's prospects unless the partner wrote it.
- Error log: any wrong date, name or fact, with the cause, in the shared log. Three entries with the same cause in a month means the prompt or the precedents need fixing.
- Accounts: the project, prompts and automations sit in firm-owned accounts; the office manager holds the admin log-ins.
- Review: first Monday of each quarter, 30 minutes.
The accounts line exists because of a realistic failure. A consultant builds the intake automation in their own automation-platform account to save time, intending to transfer it later. The engagement ends, the transfer never happens, and months afterwards the consultant closes the account. Enquiry acknowledgements stop without an error message, and the firm notices only when a prospect complains about hearing nothing for two days. Everything you pay for should be built in accounts the firm owns from day one, with the consultant added as a user and removed at handover.
What to get in writing from each phase
| Phase | Deliverable | How you'll know it's done |
|---|---|---|
| Discovery | Ranked task list with hours, risk and recommendation | Partners recognise the numbers and agree the top three |
| Risk and rules | AI policy, approved-tool list, engagement letter wording | Signed off by the compliance lead and circulated |
| Tool choice | Test results on your own material, with a recommendation and costs | You could explain the choice to a client who asked |
| Configuration | Working set-up, prompt templates, automations | A fee earner completes a real task with it unaided |
| Handover | Runbook, checking routine, owner, review date | Two weeks pass with no calls to the consultant |
How the fee is built
Consultants price in four main ways: a day rate, a fixed fee per phase, a monthly retainer after handover, or occasionally a share of savings. Day rates vary so widely by experience and market that any single figure would mislead; compare proposals by the number of days and what each day delivers. For the general picture, see what an AI consultant costs a small business.
Here's an illustrative proposal for an eight-person firm, written in days so you can apply any rate:
| Phase | Consultant days | Firm time needed |
|---|---|---|
| Discovery | 3 | 2 partners × 3 hours; 4 fee earners × 2 hours; 2 support staff × 2 hours |
| Risk and rules | 1.5 | Compliance lead, 4 hours |
| Tool choice | 2 | 2 fee earners × 2 hours testing |
| Configuration | 6 | Office manager, 6 hours |
| Handover | 1.5 | All staff, 1 hour each |
| Total | 14 days × your quoted rate | about 40 hours |
Put a value on the firm's column. If those 40 hours were billable at an average of $200, that's $8,000 of fee-earning time, which may be close to the consultant's own fee. This is the strongest argument for a tight discovery phase: a consultant who needs 12 hours of each partner's time is expensive however low their rate. Software costs sit on top and are separate; a general business AI plan such as Claude Team or ChatGPT Business is $20 per seat a month billed annually, and legal-specific tools are mostly quoted.
Two structures deserve caution. Open-ended hourly billing on configuration invites scope creep. And a share-of-savings fee sounds aligned but depends on measuring savings you'll argue about later; if you agree one, define the baseline in writing first.
The sum shows why. Suppose the consultant takes 20% of the value of fee-earner hours saved, valued at $200 an hour. Their measurement, timing drafts alone, finds 40 hours a month saved: $8,000, so a $1,600 monthly fee. The firm's measurement counts the time spent checking the drafts and finds 25 hours: $5,000, so $1,000. That $600 a month gap is $7,200 over a year, and neither side is wrong about its own method. Agreeing in advance that review time is counted, and who does the measuring, is what stops the dispute.
Retainers after handover need the same scrutiny. A retainer of one consultant day a month is 12 days a year. If the firm's realistic need is a half-day check-in each quarter plus one or two days when a vendor changes something, that's four or five days, and paying for them as needed costs well under half the retainer. A retainer earns its place when there's a steady stream of new workflows to build, not as insurance.
What pushes the price up or down
- Number of practice areas. Each one adds interviews and its own risk profile. Conveyancing and litigation share very little.
- Where documents live. Matters organised in a document management system make configuration faster. Matters scattered across email and personal drives add days.
- Integration. Advice on tools is cheap; connecting AI to your practice-management system, intake form and email is where days accumulate.
- Build versus advise. A consultant who builds automations charges for build and testing time. One who advises and leaves your IT provider to build charges less but hands you a coordination job.
- How much you've already done. A firm with an AI policy and a trial already under way can skip most of phase 2.
When a small firm can skip the consultant
You probably don't need outside help if all of these are true: one partner has the time and interest to own AI for a quarter; your practice-management system already includes AI features; the first tasks are drafting and summarising rather than integrations; and you're content to start with a written policy and one general AI plan. The 90-day plan for a small law firm is built for firms taking that route.
You probably do need help if you want automations that touch client communications, if your documents are spread across several systems, or if two partners disagree about whether AI should be used at all and need a neutral view grounded in your own numbers.
There's a middle route for firms that sit between the two. A three-partner practice with one interested partner might buy only discovery and the risk-and-rules phase, roughly three to six consultant days, and do the configuration itself from the ranked list and the policy. That keeps the part where outside eyes help most (spotting where the hours really go, and writing rules that match regulatory guidance) and drops the part a capable partner can do from vendor documentation. Ask for a separate fixed fee for each phase so this split is possible; a proposal that only prices the full engagement is telling you something.
Red flags in a legal AI consultant's proposal
The wording of a proposal tells you a lot. Compare these two descriptions of the same deliverable (both illustrative):
- Weak: "We will deploy an AI legal assistant trained on your precedents to answer client questions automatically and cut research time by 70%."
- Strong: "We will set up a shared project holding your ten most-used precedents, write prompt templates for first drafts of three letter types, and agree a review step with the supervising partner. We'll measure drafting time on 20 letters before and after."
The first promises an outcome nobody can guarantee, puts AI in front of clients without supervision, and gives you nothing to test. The second names what will exist and how you'll check it, and it gives you a test to hold the consultant to. Ninety days after handover, the firm times 20 more letters the same way. An illustrative result: 35 minutes a letter before, 21 minutes after including the partner's review, and three error-log entries in the quarter, all caught before sending and two of them traced to one outdated precedent. That's a result the partners can see and a fix they can make, which is what a well-written deliverable should produce. Other warning signs:
- No mention of confidentiality, privilege or your regulator's guidance.
- Offering to process client files in the consultant's own accounts or tools.
- Suggesting AI for tasks that should stay with lawyers; the legal tasks a small firm should never hand to AI is a useful cross-check.
- A single recommended product regardless of what discovery finds.
- No handover phase, or a handover that is really a retainer.
Before signing, run the contract past the clauses in AI consulting contracts: clauses to check, particularly ownership of the prompts and configurations you're paying for.
Questions firms ask before hiring
Should an AI consultant see our client files?
Usually they don't need to. Most of the work can be done from process interviews, anonymised samples and test matters. Where they must see live material, treat them like any other supplier with access: a confidentiality agreement, a data-processing agreement if they handle personal data, access limited to named matters, and a record of what they saw. Check your regulator's rules on outsourcing and supervision.
Who is responsible if AI output is wrong after the consultant leaves?
The firm. Professional responsibility for advice and documents stays with the lawyers, whatever tool produced the draft. That's why the handover should include written checking routines, named owners for each workflow and an error log. A consultant's contract may limit their liability for configuration mistakes, but it won't move your duty to clients.
Is it a problem if the consultant also resells software?
Not necessarily, but you need to know. Ask in writing whether they receive commission, referral fees or reseller margin from any vendor they might recommend, and ask them to include at least one option they don't earn from. A recommendation that always lands on the same product, whatever your practice areas, is a sign the advice isn't independent.
Further reads
- What Does an AI Implementation Consultant Actually Do? — The general version of this role, outside legal work.
- How to Choose an AI Consultant: 20 Questions to Ask First — Twenty questions to shortlist consultants before any quote.
- AI Consultant Red Flags: 12 Warning Signs to Walk Away From — Twelve general warning signs, beyond the legal-specific ones.
- How to Vet an AI Consultant's Case Studies and References — How to check the examples consultants show you.
- Per-Seat or Pay-As-You-Go? Legal AI Pricing for Small Firms — The software costs that sit alongside the consulting fee.
- What an AI Consultant Can't Do for You, and What You Must Own — The decisions that stay with the partners, whoever you hire.
- Pros and Cons of Using AI in Your Business, With Real Costs — Every pro and con of business AI with a price attached, a first-year ledger for a small law firm, and the costs that never reach an invoice.
- AI for Small Law Firms: What to Automate First — A scoring method and a firm-tested order for what a small law firm automates first, with the legal work that should wait.
- AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.
Sources: no vendor pricing is quoted for consulting, because consultants' rates are not published consistently; software prices referenced are from OpenAI and Anthropic pricing pages.