Building a Law Practice Automation, Field by Field

There is already a post arguing why an automation on a legal account is mostly a refusal list rather than an answering machine — what an automation for lawyers has to refuse. It covers the disclosure problem, the conflict problem, and why a fast confident reply is the worst available response.

This post does not restate any of that. It assumes you agree and want to know what to actually type into the screen. One afternoon, three automations, in order.

Before you open the builder: the route has to exist

The automation is downstream of your intake. It cannot create one.

So the first thing is not a field, it is a decision: where does a stranger who writes to your Instagram account go next? A consultation booking page. A number the office answers. An intake form that somebody empties daily. Pick one and confirm it works today, not in principle.

If that route does not exist, stop here. An automation that routes to nowhere is worse than silence, because silence at least does not promise.

Step one: write the negative keyword list before the trigger

This is the reversal that makes a legal build different from a bakery's. On most accounts the trigger field decides everything. Here the field underneath it does.

Open your own inbox and read a hundred messages before writing a single word. You are looking for the vocabulary that means a human must read this, and it is specific to what you practise. A criminal practice ends up with FIR, summons, bail, custody, hearing, remand. A family practice with the words of an ongoing proceeding, and anything mentioning a child. An employment practice with the words people type when they are about to do something irreversible on Monday.

Write the list in a document first, get it signed off by whoever signs off compliance at your firm, then paste it into the builder. The document matters as much as the field, because in eight months somebody will want to know why a word is on the list.

Add the money words too — fees, charges, kitna lagega, free consultation — not because they are dangerous in themselves, but because how a practice may talk about fees in public is regulated in a lot of places and you do not want a template deciding it.

Step two: the consultation automation

The first automation goes on one post. Not the account.

Write the private reply as a template rather than leaving it to the model. A template is a sentence you read in full before anyone else did, which is the property you want on a legal account and the property a generated reply cannot have.

Step three: the resource automation

The second one is the easiest and the most useful. A checklist, a document list, an explainer PDF on your site — the things people ask for that carry no advice at all.

Same shape, different reply. Trigger CHECKLIST on the post that offers it. Private reply: the link, one line on what it is, and the same non-advice sentence. This one can run for months without maintenance, because a document list does not change the way a fee structure does.

It is also the automation that earns its keep. Templated replies are free and unlimited, and this is pure template.

Step four: the office-logistics automation

Where the office is, which days somebody is at the desk, which languages the practice works in, what to bring to a first meeting. Fixed facts, identical for everyone, entirely safe.

Keep it separate from the consultation automation rather than folding it in. When the office moves floors, you want to edit one template, not find the sentence inside three.

Step five: test on yourself, then read the refusals

Test on myself runs the whole pipeline — the match, the ten checks, the generation, the send — and delivers to your own account. Nothing is faked. Read the result on your phone, at night, the way the sender will.

Then go live on one post and open Activity the next morning, filtered to blocked.

A stack of blocked rows is the correct first week. POLICY.WINDOW_CLOSED on older comments is the platform, not a fault. Takeovers mean somebody in the office answered by hand first, which is the outcome the whole build is arranged to produce.

Two settings worth arguing about

Quiet hours do not protect anybody here. The four-paragraph message still arrives at 11:40pm; quiet hours only hold your reply until morning. That is fine, and it is worth knowing it is a courtesy rather than a control.

First-time senders only, under Advanced, is more useful than it looks. It stops an existing client — who is mid-matter and messaging about their own file — from receiving a generic intake template. That is the failure that costs the most and nobody anticipates it.

What it will not do

It cannot message anyone who has not messaged you first. It does not do WhatsApp. It has no CRM sync; captured leads leave as a CSV. It does not run sequences and it cannot follow up on a schedule.

Official Graph API only, so the clocks are the platform's: seven days to reply to a comment, twenty-four hours inside a DM thread, and only their next message restarts the second one. A Friday-night enquiry left for Monday is frequently past the window, which is an argument for a free template answering it at midnight rather than for nobody answering it at all.

Templated replies are free and unlimited; a credit is spent only when the AI writes a new reply. Free tier is 100 credits with no card, packs from ₹499. A practice built this way spends almost nothing, because generating novel text is the exact behaviour the build exists to avoid.

If you want the full list of what stops a send, the checks are documented here, and the paced setup guide is the right order for the first week.