AI Setup for a Two-Lawyer Estate Planning Firm: The First Month

Set up three things first: a written AI-use policy covering what client information may enter a tool, a business-plan assistant account that does not train on your content by default, and one low-risk workflow with a named reviewer. For two lawyers that takes about 25 to 30 hours across the month and costs roughly $40 to $50 a month.

The order matters more than the tool. Firms that buy software first usually discover in week three that nobody has decided what may be typed into it. Estate planning adds its own traps: clients disclose family conflict, health and money, a draft may later be read by a court, and thresholds and signing formalities vary and change. So put rules ahead of features and start where a mistake is cheap to catch.

Follow me on Instagram@sagnikteaches

The month at a glance, with hours and costs

Take an illustrative practice: two lawyers, one paralegal, one administrator, about 15 new estate planning matters a month and around 60 live files. Here is the plan in one table, followed by each week in detail.

Connect on LinkedInSagnik Bhattacharya

Costs. ChatGPT Business and Claude Team both need at least two seats. On the Claude pricing page, a Team standard seat is $20 a month billed annually or $25 monthly, so two seats cost $40 to $50. ChatGPT Business is priced the same way, $20 per seat annually or $25 monthly. Both vendors say business plans do not train on your content by default; confirm that on the current pages before client material goes in. If the firm already runs Clio Manage, check your plan before buying a second tool: Clio's pricing page, as of October 2026, includes AI on the Core plan and above rather than selling it as an add-on. Template automation for wills and trusts is a later decision with its own costs, so it stays out of this month's budget. The larger cost is time: 25 to 30 hours across the team, about 60 per cent of it lawyers' time, which at any realistic billing rate is the real investment.

Subscribe on YouTube@sagnikteaches

Week 1: write the policy before opening an account

The policy fits on one page. Its job is to tell staff, in plain words, what may go into which tool. Draft it in a single sitting, then have both lawyers sign it. Here is a filled-in version to adapt, illustrative:

AI USE POLICY - [firm name] - version 1, [date]
1. Approved tools: [assistant name] firm account only. No personal or free accounts for client work.
2. Allowed in the tool: client names and matter facts needed for the task, with account numbers, tax identifiers, dates of birth and addresses replaced by placeholders such as [ACCOUNT], [ID], [DOB].
3. Never in the tool: signed documents, identity documents, medical records, anything marked privileged for another matter.
4. Every output is a draft. A named lawyer reviews anything that goes to a client or into a legal document. The reviewer's initials go on the file note.
5. AI does not: assess capacity, advise on tax thresholds, state signing formalities, cite law, or draft dispositive terms without a lawyer rewriting them.
6. Record keeping: save the final reviewed text on the file; keep prompts only if they contain no client detail.
7. Review: this policy is reread at the end of each month for the first three months.

Point 2 is the one that matters most. Names and the facts of the task have to reach the tool for it to be useful, but identifiers do not. The placeholder habit takes seconds and removes the worst consequence of a mistake. Point 5 lists what AI does not do. Estate planning rests on points that change with place and year: thresholds, execution requirements, treatment of certain assets. A tool trained on mixed sources will produce confident text that may be wrong for your client, so the policy takes those topics out of the assistant's hands.

Then set up accounts. Create the business plan with the firm's email domain, make one lawyer the admin, switch off any optional features you have not reviewed (connected apps, memory, browsing) until you have read what each does, and turn on multi-factor sign-in. If the firm runs Clio Manage, its AI features may already be live: Clio's help pages describe a firm-level AI access setting under Settings, Security & Compliance, AI Features, but say it stays on where your licence includes AI, so ask Clio's support team if you want it off until the policy is signed. Read the vendor's privacy terms for retention and training; the privacy terms checklist lists the clauses worth reading. Note that ChatGPT workspaces on the Business plan have no self-service data export, which matters if you ever want to move your saved projects out.

Choosing the one workflow

Pick by three tests: the output is read by a lawyer before anyone else sees it, a mistake is obvious on a second read, and the task recurs at least ten times a month. Here are candidates for a two-lawyer practice, ranked by suitability for the first month.

1. Post-meeting summary letter. The lawyer dictates or types notes from the planning meeting; the assistant turns them into a plain-English letter confirming what was discussed and what happens next. Recurs on every matter; easy to check against the notes.
2. Document checklist. From the matter type, produce a list of what the client must bring or complete. Low risk, saves paralegal time.
3. Client status update. A short message saying where the matter stands, built from a file note. The status update tutorial covers this.
4. Intake questionnaire to draft instructions. Riskier because it feeds the draft; the intake questionnaire tutorial covers it and is a good second workflow.
5. Drafting a will or trust. Not a first-month task. It needs the comparison in the template automation against generative drafting tutorial first.

For the rest of this plan I will use the post-meeting summary letter as the example workflow, because it meets all three tests.

Week 2: build and test on dummy files

Build the workflow in a shared project inside the assistant, so both lawyers use the same instructions. ChatGPT Projects and Claude Projects both allow shared instructions and files on business plans; ChatGPT's custom GPTs are being retired, so do not build on those. Write the instructions once, then test them on invented matters before any client's facts go near them.

Here is a project instruction block, illustrative:

You help a two-lawyer estate planning firm draft plain-English client letters from the lawyer's meeting notes.
Rules:
- Use only facts in the notes. If something is missing, write [CHECK: what is missing]. Never fill gaps.
- Do not state any tax threshold, deadline, signing requirement or legal rule. Write [LAWYER TO CONFIRM] where one would normally go.
- Keep to 350 words, short paragraphs, no legal Latin, no jargon without a one-line explanation.
- Structure: what we discussed, what you decided, what happens next, what we need from you, who to call.
- Tone: warm, direct, no promises about timing unless the notes give a date.

Now a test input and a realistic output. The notes below are invented.

NOTES: Meeting with [client], 45 mins. Widowed. Wants a new will to replace the one made after her husband died. Two adult children, [child 1] and [child 2]. Residue equally to both; [child 2]'s share held in trust rather than paid outright - LAWYER TO ADVISE on structure. Executors: [child 1] and [friend]. Car to [grandchild]. Incapacity documents, property and health; decision-makers [child 1], then [friend]. She will bring the old will, the house deeds and a list of accounts. Next: draft 3 weeks after documents arrive.

A good output reads like a clear letter: it confirms a new will replacing the old one, the rest of the estate split equally between the two children with [child 2]'s share held in trust [LAWYER TO CONFIRM: structure], [child 1] and [friend] as executors, the car going to [grandchild], documents appointing [child 1], and then [friend], to act on property and health decisions if she loses capacity, the three things she will bring, and a draft about three weeks after they arrive. What you would fix in a first attempt, illustrative: the assistant wrote "a trust will protect [child 2]'s inheritance and any benefits they receive", a legal claim that is not in the notes and may be wrong for this client. The fix is a line in the instructions, "never describe what a document does or achieves beyond the notes", then retest.

Run five dummy matters with different features: a gift to a charity, a minor child who needs a guardian, a client who owns a small business, a client who wants to leave someone out, and a client with a pet to provide for (use generic descriptions only). Record each time the assistant adds a fact, drops one or states a rule. Expect three or four rounds of adjustment. Then write the review sheet:

REVIEW SHEET - summary letter
[ ] Every name, relationship and gift matches the meeting notes
[ ] No legal rule, threshold or deadline stated unless I have added it
[ ] All [CHECK] and [LAWYER TO CONFIRM] flags resolved
[ ] Executors and guardians listed in the order discussed
[ ] Nothing promised about timing or cost beyond the notes
[ ] Each next step names the right owner: the client or the firm
[ ] Tone suits this client
Reviewed by: ____   Date: ____

Week 3: pilot on live matters and tell clients

Use the workflow on ten real matters. The paralegal drafts with the assistant, using placeholders for identifiers, and the lawyer who held the meeting reviews every letter against their own notes. Time both steps. An illustrative result: writing a summary letter by hand takes 30 minutes; with the assistant, drafting takes 8 minutes and review takes 7, so about 15 minutes saved per matter. At 15 new matters a month that is around four hours saved, plus the benefit of a consistent structure. Do not expect to save more in month one, because review is deliberately slow at first.

Also in week 3, tell clients. Add a short statement to the engagement letter and the website's privacy page. Wording to adapt, illustrative:

We use secure, business-grade software, including AI tools, to help prepare routine letters and checklists. Your information is entered only into tools the firm has approved, identifiers are removed where possible, and every document is reviewed by your lawyer before it reaches you. AI does not make legal decisions or draft your will or trust. If you would prefer that AI is not used on your matter, tell us and we will work without it.

The last sentence is optional but useful. A client who objects costs you nothing if the workflow is easy to skip, and the offer builds trust. Check with your bar's ethics guidance whether your situation requires a specific disclosure or consent; I cannot tell you what applies to your practice.

A mistake the pilot should catch

An illustrative failure from a dry run: the notes said "she will bring the old will, the deeds and a list of accounts". The assistant's letter said "we will obtain your old will, the deeds to your house and details of your accounts". A reviewer skimming for tone would miss it; a reviewer working through the sheet's "right owner" line would not. The error stalls the matter quietly: the client waits for the firm, the firm waits for the client, and three weeks later nobody has the deeds. The lesson for the review sheet is to make the check specific, because "looks fine" is how this kind of slip goes through. If the letter is wrong in the first ten matters more than twice, stop and fix the instructions before widening use.

Week 4: review, measure and choose the second workflow

Hold a 90-minute review with both lawyers. Bring the counts.

• How many letters were drafted, and how many needed more than light edits?
• How many [CHECK] flags were raised, and what did they reveal about missing meeting notes?
• Minutes per letter before and after.
• Any client questions or objections about AI.
• Any policy breach: an identifier pasted in, a personal account used.

Decide one of three things: keep the workflow as is, change it, or stop it. If it saved time and the error rate is low, pick the second workflow. For an estate planning practice the usual next step is intake: a questionnaire that gathers family and asset details in advance, which can take much of the fact-gathering out of the first meeting. Run it through the same dummy-file routine as this month's letter before any live client sees it.

Update the policy at the same meeting. Add what you learnt, such as a new placeholder rule or a banned phrase, and increment the version number. Then add the pilot to a simple risk register so the risks you accept are written down; the risk register template is a good start.

What to leave alone in the first month

• Anything that decides or states law. Tax thresholds, execution formalities and treatment of particular assets vary and change. The assistant should flag, not answer.
• Capacity and undue influence. These are judgements made by the lawyer in the room. Do not feed an assistant notes asking whether a client "seemed to understand".
• Dispositive drafting. Clauses that distribute property need either a rules-based template that a lawyer built, or a lawyer's rewrite. See the comparison tutorial before deciding.
• Research citations. General assistants invent references; the checking routine for invented case law is the minimum if you ever use one for research.
• Anything involving a client who has asked not to use AI.

Where the 25 to 30 hours actually go

Partners rarely believe the estimate until they see it itemised, so here is the same month broken down by task, counting each person's hours. Drafting the policy takes about three hours, including a second read by the other lawyer. Choosing and configuring the accounts takes two hours, most of it reading terms and switching off features you do not need. The week 1 staff session is one hour for three people, so three hours. Writing the project instructions and the review sheet takes three hours, and dummy-file testing another five to six, because each of the five test matters needs a run, a read-through and a note of what went wrong, over three or four rounds. In week 3 the ten live pilot letters take about three hours of drafting, review and logging, the client notice (engagement letter edit and website line) two hours, and a 30-minute check-in with the same three people another hour and a half. The end-of-month review is ninety minutes for both lawyers, so three hours, and the policy update half an hour. The total lands at roughly 26 to 27 hours, split about 60 per cent lawyers and 40 per cent paralegal and administrator.

That split matters for cost. Lawyer hours are the dear ones, so push anything that does not need a lawyer to the paralegal: building the checklist, running the dummy tests, timing the steps. The lawyers' real jobs are the policy, the review sheet and the review meeting.

A one-hour session for the paralegal and administrator

A policy that only the partners have read protects nobody. Run a one-hour session in week 1 and keep it practical. Spend ten minutes on why: client information is confidential, and a free chatbot is a place where it can leak or be stored in ways the firm does not control. Spend twenty minutes walking through the policy line by line with real examples of what goes in and what does not. Spend twenty minutes having each person use the approved tool on a dummy matter, replacing identifiers with placeholders, so the habit is physical and not theoretical. Spend ten minutes agreeing how to report a slip. The rule should be that nobody is blamed for telling a partner they pasted something they should not have, because the alternative is a firm that never hears about it.

Before and after: the same paragraph, twice

The point of the summary letter is plain English. Here is a typical paragraph from a hand-drafted letter, illustrative:

Pursuant to our meeting, we confirm your instructions that your residuary estate be divided equally between your children, the share of [child 2] to be retained upon trust on terms to be settled. Upon receipt of the documentation enumerated below, we shall prepare a draft will for your consideration and approval.

And the version the assistant produced from the lawyer's notes, before the lawyer's edit:

You'd like everything left after your gifts to be split equally between your two children, with [child 2]'s half kept in a trust. We'll send you your draft will in three weeks.

The second is shorter and clearer, but it has changed something. The draft depended on the documents arriving first, and the plain version turned that condition into a promise with a date. If the deeds turn up late, the firm has broken a commitment it never meant to make. The lawyer's edit restored it: "Once we have the documents listed below, we'll send your draft will, usually within three weeks." That is the pattern to expect: the assistant improves readability and quietly drops conditions, which is why the review sheet checks every promise about timing against the notes.

Which estate planning tasks are safe to hand over, and which are not

How to tell the pilot worked, and when to stop it

Set the thresholds before the pilot, not after. These are illustrative; adjust them with your partners.

• Continue if at least eight of ten letters needed only light edits, no letter contained an uncorrected fact error at review, and the average time per letter fell by at least five minutes.
• Change if three or more letters needed heavy edits. Look at the pattern: the same missing instruction usually causes it. Fix the project instructions and run five more.
• Stop if an error reached a client, a policy breach happened twice, or reviewing took longer than writing by hand. Stopping is a valid result; it costs you a month of subscriptions, not a client.

Keep the evidence simple: a spreadsheet with one row per matter, the minutes spent drafting and reviewing, the number of flags raised, and a column for what you changed. A half-page of notes from the review meeting is enough for the file, and it becomes the start of the risk register.

If you only have ten hours

A leaner version still works: write the one-page policy (2 hours), set up the business accounts (1 hour), build the summary letter workflow with three dummy tests (3 hours), pilot on five live matters with full review (3 hours) and hold a 30-minute review. You will save less time, but you will have the safeguards. A first month that ends with a signed policy, a client notice and one workflow you trust is a better result than one that ends with five tools and no rules. If you would rather map your own matters with someone, that is what my AI implementation consultation covers, and the contact page is the way to reach me.

Questions two-lawyer firms ask before starting

Which AI account should a two-lawyer firm start with?

A business plan, not personal accounts. ChatGPT Business and Claude Team both need at least two seats, which fits two lawyers, and both say they do not train on business content by default. Cost is about $40 to $50 a month for two seats. Read the current terms yourself before client material goes in.

Do clients need to be told the firm uses AI?

Many firms say yes, and your bar's ethics guidance may require it in some situations, such as when client information goes into a tool or when it affects fees. A short line in the engagement letter is cheap and builds trust. Ask your regulator or ethics adviser what applies to you.

Should the first workflow touch wills or trusts?

No. Start with something where an error is caught easily and cannot become a legal instrument: a post-meeting summary letter, a document checklist or a status update. Drafting dispositive documents comes later, and needs a rules-based or tightly reviewed process.

How do we stop staff putting everything into a chatbot?

Give them an approved tool that is easier than the alternative, write one page saying what may and may not go in, and review a sample of use each month. People paste sensitive material into free tools when the approved route is slower or unclear.

Further reads

• ChatGPT Business for Law Firms: Seats, Privacy Settings and Cost — Seats, privacy settings and cost of the business plan used here.
• Clio Manage AI Pricing: What It Costs and How to Switch It On — What Clio's built-in AI costs if the firm already uses Clio.
• How Small Law Firms Use AI to Draft Letters and Routine Documents — More letter types once the summary letter works.
• How to Set Up Human Review for AI Work Without Slowing Down — Review steps that do not slow the firm down.
• AI Implementation Plan for a Small Law Firm: The First 90 Days — The 90-day version of this month's plan.
• What an AI Consultant Does for a Law Firm, and What It Costs — What outside help costs, if you want it.
• Gavel Pricing: Which Workflows Plan a Small Firm Needs — Gavel Workflows from $83 to Scale at $417 a month, matched to a small firm's seats, templates and integrations, with break-even sums and a 7-day trial plan.
• AI for Bankruptcy Lawyers: Intake, Schedules and Document Chasing — A document-to-draft route for consumer bankruptcy matters, with missing-evidence checks before any client signs.
• How to Use AI in a Law Practice: A Four-Week Starter Plan — A week-by-week rollout for a small law practice: write the policy, pilot one low-risk task, verify every citation, then settle client disclosure and review.
• Using AI in Law: Which Tasks a Small Practice Should Start With — A risk-ranked list of what a small law practice can hand to AI first, with supervision and confidentiality rules for each rung.
• Which AI Is Best for Lawyers in Australia? Choose by Task — A decision rule for choosing AI by task for an Australian law practice, comparing general assistants with the AI built into practice software.
• AI Tools and AI Development: The Complete 2026 Guide — the AI hub, including every tutorial in the AI-for-business series.

Sources: Claude pricing page (Team), OpenAI ChatGPT Business pricing and help pages, Clio pricing page and AI features help page (checked October to November 2026).