Most Law Firm Intake Calls Are Not Cases. That's the Problem.
David Selva · · 3 min read
AI Voice Agents, Law Firms, Lead Qualification
A paralegal spending her morning on calls the firm will never take is the most expensive filter a firm can own.
Forty inbound calls in a week. Six of them are cases the firm would actually sign.
The other thirty-four are wrong practice area, outside the statute, no injury, already represented, or somebody who wants free advice about a neighbor's fence. Somebody has to sit through all forty to find the six, and at most small firms that somebody bills at a rate making it a genuinely bad use of the hour.
Worse: the six real cases are on hold behind the thirty-four.
Qualification is a script, not judgment
Here's the part firms underrate. The first ninety seconds of an intake call isn't legal work. It's a fixed set of questions with a decision tree behind it.
When did it happen. Were you injured. Were you treated. Was a report filed. Are you represented. What state.
That's a script. Your intake person runs it from memory already, and runs it slightly differently at 4:30 on a Friday than at 9 on a Tuesday. An AI voice agent runs it identically at six on a Sunday morning, and never puts call two on hold to finish call one.
What happens in the thirty seconds after
The value isn't that a machine asked the questions. It's what the answers trigger:
- Qualified callers get booked into an attorney consult while still on the phone
- Clear non-cases get a courteous, accurate close and a referral note, without burning staff time
- Anything ambiguous routes to a person with the answers already captured, so nobody restarts from zero
- Every call lands in the system with its answers and a source attached, signed or not
That last one quietly matters. Most firms can't tell you which marketing spend produced the cases they signed, because the evidence lives in a paralegal's recollection of a phone call.
The line you don't cross
It doesn't give legal advice. It doesn't evaluate the merits of a claim. It doesn't tell anybody whether they have a case. It gathers facts and it schedules.
That isn't a limitation to engineer around — it's the entire design constraint. Any question starting with do you think I can sue gets a scheduled consult, not an answer. And read the transcripts weekly, because the calls it handled badly are the ones telling you where your script is wrong.
Where I'm genuinely unsure
I don't know where the line sits for a caller in real distress.
A person calling three days after a serious accident isn't in a state to be efficiently qualified, and I'm not convinced any script handles that well — ours included. You can route on tone and keywords, and we do, but tone detection is crude and the cost of getting it wrong is a person in a bad moment feeling processed by a machine.
My working rule is to over-transfer: when in doubt, a human. That costs the firm efficiency in exchange for not being the firm that did that to somebody. I think that's the right trade, but I'd hold it loosely, and I'd rather a firm decide it deliberately than inherit it from a default setting.
Before you automate any of this, pull last month's intake calls and mark each one signed, not signed, or never a fit. If more than two thirds are in that third bucket, your intake isn't a staffing problem. It's a filtering problem, and filters can be built.