Law firm automation: start with the phone, not the documents

When a lawyer says “automation” they usually mean documents: the engagement letter that assembles itself, the discovery responses drafted from a template. Those are real, and they are the wrong place to start at a firm of one to five lawyers.

The reason is arithmetic. A drafting automation saves minutes on work you were going to do anyway. An intake automation recovers a matter you were going to lose without knowing it. One is an efficiency; the other is revenue. Start where the money leaks.

Six automations, in order of payback

The order below is a judgement from building these, not a measurement. Each entry says what it needs, where a person has to decide, and how it fails.

1. Missed-call text-back

A call goes unanswered and the caller gets a text within a minute: sorry we missed you, what is it about, here is a time.

• Needs: a business number that can send texts, a trigger on a missed call, and somewhere the replies land. An afternoon.
• The gate: none on the first text; it is a fixed message. A person reads the replies.
• How it fails: the number is not registered for business messaging and the texts are silently filtered. Test it as a stranger. The full setup, the message and the rules are in the text-back guide.

2. Intake and booking

After hours, or when everyone is with a client, a receptionist takes the caller’s details, asks the five questions you approved, and books a time from the real calendar.

• Needs: the questions written and approved per practice area, the calendar connected, a morning routine for reading the log.
• The gate: acceptance. The booking is provisional until a person reads the intake. The receptionist never accepts a matter, quotes a fee or gives advice. How the call goes, step by step.
• How it fails: it acts when it should pause. A published scheduling build computed a confidence score on every decision and read it nowhere.

3. Confirmations, reminders and no-show recovery

A confirmation when the consult is booked, a reminder the day before, and a message offering a new time when someone does not turn up.

• Needs: the booking in a calendar the system can read, and three messages written once.
• The gate: none. These are fixed messages about a meeting the person asked for.
• How it fails: it chases someone about a meeting they already moved. The system has to re-check the calendar before every message, not assume.

4. The review request

After a consult or a closed matter, one message with the review link. Anyone unhappy is routed to you first, not to the public page.

• Needs: a trigger on “matter closed”, one message, and the routing.
• The gate: the lawyer decides which matters get the request. Not every closed file should.
• How it fails: asking at the wrong moment. A review request to someone who just lost costs more than it earns.

5. Follow-up for leads that go quiet

A person inquired and then stopped answering. Two weeks of spaced follow-up, then a close.

• Needs: every inquiry in one place with a status, and the messages written to sound like the firm.
• The gate: a person can stop the sequence at any point, and the sequence stops itself on any reply.
• How it fails: a follow-up that reads as marketing. Keep it to a response about the thing they asked, and respect the solicitation rule.

6. Drafting and extraction

Now the documents. A first draft assembled from the firm’s own precedents, or the same fields pulled out of every document in a set.

• Needs: a precedent bank worth drawing on, and a lawyer who checks every citation and every number before relying on it. Two guides on this site cover drafting from your own precedents and checking citations.
• The gate: every draft is reviewed by the lawyer who signs it. The ABA’s Formal Opinion 512 applies the duties of competence and supervision to exactly this.
• How it fails: a fabricated citation reaches a filing. The check is not optional.

The rule for where the gate goes

Every item above follows one rule, and it was learned the hard way in the builds published here. The model does judgement. The system does everything the model should not be trusted with.

Reading a caller’s intent is judgement. Deciding to accept a matter is not something to hand over. Composing a message a client will read is not something to improvise; the words are fixed and approved, and the model chooses among them. Where a decision would embarrass the firm if it were wrong, a person makes it, and the system’s job is to make that decision take two minutes instead of twenty.

The failure that gives automation away

Clients do not mind that a text was automated. They mind being chased about a meeting they already booked, being texted by a “Sarah” who does not exist, or being asked for a review the week they lost. Every one of those is a system acting on stale information or on a decision it should have left to a person. Bar associations writing about these tools, including the North Carolina Bar, tend to reach the same place: the value is in routine, and the risk is in letting routine make decisions.

What is hype and what is not

Some of the trade press, such as Bloomberg Law’s review of what is working in 2026, notes that most lawyers are still not using the collaboration and workflow tools already available to them. That matches what we see: the gap at an independent firm is rarely the absence of clever software. It is the phone ringing out at 6pm and nobody knowing.

So: the text-back this week. The receptionist when the text-back is working. The reminders and the review request when there are consults to remind people about. The documents after that. Each one only when the one before it has earned it, which is also how the offer is built.