AI Agents for Law Firms: Intake Without the Missed Call

An AI agent for law firm intake answers calls and messages around the clock, qualifies potential clients by practice area, and routes conflict-check flags to the attorney before the first consultation is ever scheduled. The Legal Services Corporation found that 22% of people with civil legal problems don't know where to find help. A firm whose intake answers in seconds removes that friction the moment someone reaches out, whether the call comes in at 2 PM on Tuesday or 10 PM on Sunday. This post gives you the framework for deciding which intake step to automate first and exactly how to set the agent's qualification criteria before you go live.

Why does law firm intake keep losing potential clients?

The missed call is the oldest problem in legal services, and it has not gotten easier. A person decides they need a lawyer when their situation demands it: often in the evening, often on a weekend, often when they are already stressed and comparing options quickly. They call the first three firms they find. The first two don't answer. The third firm answers and books the consultation. Monday morning your staff returns the call and hears: "I already found someone else."

The window that closes before Monday morning

Law firm inquiry volume is not evenly distributed across a 9-to-5 window. Personal injury and family law inquiries cluster around evenings and weekends, the moments when an accident, a confrontation, or a difficult conversation prompts someone to search for representation. A firm without after-hours coverage hands those callers to the competitors who do answer.

The scale of unmet need makes this more significant than a single firm's callback rate. The LSC Justice Gap Report found that 71% of low-income households experienced at least one civil legal problem in the past year, and only 20% of those problems initially prompted anyone to seek professional legal assistance. One reason is simple: 22% of people with legal problems don't know where to find help. The caller who doesn't reach someone on the first try often stops calling altogether.

What a missed evening call costs a contingency firm: an illustrative example

Illustrative example, not a client result: a personal injury firm handles 30 inbound calls per week, with 30% arriving after hours. If nine of those calls go to voicemail and the Monday callback converts at 15% versus 40% for calls answered live, the firm passes on roughly two retained clients per week. At an average personal injury attorney fee of $8,000, that gap is $16,000 per month in potential revenue. Your own missed-call log tells the real figure for your practice.

What does an AI intake agent actually do for a law firm?

An AI intake agent for a law firm is not a phone tree with a transcription layer. A phone tree routes by key press. An AI intake agent reads what the caller says, classifies the matter type, collects the information that practice area requires, screens for conflict signals, and either schedules a consultation or routes the flagged call to the attorney with a full summary. The agent acts on each message; a phone tree waits for a button press.

The five tasks the agent handles without staff

After-hours call answering. The agent picks up every call the office does not answer, regardless of time. It identifies itself as an automated intake assistant and asks the caller to describe their situation in plain terms.

Practice area classification. Based on what the caller describes, the agent routes the conversation into the correct qualification flow: personal injury, family law, estate planning, business disputes, criminal defense, or other. Misclassified calls are rare when the opening question is open-ended.

Initial qualification. The agent collects the specific information each practice area requires before the first consultation. This is not a generic contact form. The question set is written for the matter type.

Conflict flag collection. The agent asks the caller to provide the names of all adverse parties. Those names are logged for the attorney to run against the existing client list before the consultation is confirmed.

Consultation booking. For non-flagged matters, the agent offers available times from the attorney's calendar and confirms. The appointment is in the system before the caller hangs up.

Where the handoff line sits

The agent stops when the caller describes an immediate safety concern, uses emergency language, provides a name that requires human conflict verification, or explicitly asks a legal question about the merits of their case. Every conversation that reaches a handoff point sends the attorney a transcript and a classification within seconds. The GPT-4 Technical Report documents that large language models can pass a simulated bar exam at the top 10% of test takers, but that capability is not why you deploy one for intake. You deploy it to route correctly, collect accurately, and hand off cleanly.

How do you build the qualification layer for a law firm agent?

The qualification layer is the set of questions the agent asks to determine whether a caller is a fit and what information the attorney needs before the consultation. Getting this right is the highest-leverage part of the build. Too few questions and the attorney walks into consultations missing critical details. Too many and callers drop off before the booking step.

The four questions every practice area shares

Across practice areas, four questions appear in nearly every intake workflow:

1. What is the nature of the matter? (Caller's own words, then agent classification.)
2. When did the relevant event or situation occur, or when is the filing deadline?
3. What jurisdiction is involved?
4. Have you spoken to or retained any other attorney regarding this matter?

Practice areas then add their own layer. Estate planning asks about asset type and approximate estate size. Family law asks whether children are involved and the current filing status. Business disputes ask for both parties' business names. Personal injury asks about insurance coverage and whether emergency services responded.

LegalBench (Guha et al., Stanford, 2023) evaluated 20 large language models across 162 legal reasoning tasks covering six types of legal reasoning. The benchmark confirms that current LLMs can reliably identify matter type, extract procedural deadlines, and classify jurisdiction from free-form descriptions. These are exactly the tasks the qualification layer asks the agent to perform.

Example: qualifying a personal injury caller step by step

Step 1. Agent confirms caller name and preferred contact number.

Step 2. "Can you describe what happened and when it occurred?" Agent classifies: auto accident, slip and fall, medical negligence, or other.

Step 3. "Did this happen in [state]?" Agent flags out-of-state matters for attorney review.

Step 4. "Have you spoken with another attorney about this incident?" Agent logs the response.

Step 5. "Could you tell me the names of any other parties involved?" Agent logs names for conflict review.

Step 6. Agent offers available consultation times and confirms.

A five-step qualification flow typically takes three to five minutes. Flows longer than seven questions see completion rates drop. See how lead qualification logic works across industries for the underlying framework that applies here.

What happens when a conflict check or sensitive disclosure comes up?

Conflict checks and sensitive disclosures are the two places where an AI intake agent must stop and escalate immediately. These are not edge cases. They happen on a meaningful share of inbound calls, and handling them incorrectly creates ethical exposure for the firm. The agent's job is to recognize these signals and hand off, not to resolve them.

The two scenarios that always go to a human

Conflict check scenarios. The agent collects adverse-party names and logs them. It does not run the check itself. The check requires access to the firm's matter management system and, in most states, a licensed attorney's review. What the agent does is ensure the attorney has the names before the consultation, not after. A conflict check the agent cannot complete is a conflict check queued for the attorney the next morning, not a failed intake.

Sensitive disclosures. A caller who mentions immediate danger, describes an ongoing safety situation, or discloses something that could trigger mandatory reporting gets routed to a human immediately. The agent identifies these signals and routes within two turns of conversation. No intake script overrides a person in distress.

Two additional scenarios trigger escalation regardless of practice area. First: callers who state that a prior attorney reviewed and declined their matter. The prior declination often signals complexity the intake agent cannot assess. Second: calls that open with a specific attorney's name. Those callers are existing relationships or referrals that need direct handling, not a new-matter intake flow.

Which practice areas see the fastest return from AI intake?

Not all practice areas benefit equally from an AI intake agent. The fastest payback comes from practices where inbound inquiry volume is high, the initial qualification is straightforward enough to script, and a missed call has a calculable cost to the firm.

The three strongest first builds

Personal injury. Contingency-fee practices have a direct link between intake conversion and firm revenue. Every retained client has an expected fee. Missed calls are lost revenue, not just missed opportunities. Accidents happen evenings and weekends, which is exactly when offices are closed. An agent that captures those calls and qualifies them before Monday morning converts a voicemail stack into a sorted case list with names, matter types, and conflict flags already logged.

Estate planning. Estate planning inquiries are high-intent but unhurried. A person who has decided to update beneficiaries or write a will is not comparing three firms in 10 minutes. They are, however, likely to call after hours when they have time to think. An intake agent that answers, qualifies by estate complexity, and books a consultation captures a significant share of after-hours inquiries that currently go unanswered.

Family law. Divorce and custody inquiries are sensitive and sometimes urgent. A firm that answers immediately and books within the same call often outperforms a firm that takes 24 to 48 hours to return a call. Speed matters more in family law than in almost any other practice area, because the decision to retain is often made while the caller's situation feels most acute.

How do you measure whether the intake agent is working?

Three numbers tell you whether the agent is performing: contact rate, qualification rate, and booking rate. Contact rate is the percentage of inbound contacts the agent picks up. Qualification rate is the percentage of those contacts that complete the intake questionnaire. Booking rate is the percentage of qualified callers who accept a consultation time. Everything else is secondary until those three are clean.

What to check in week one

In the first week, focus on contact rate and qualification rate only. Booking rate depends on attorney availability, which takes a week to calibrate against real inquiry volume.

If contact rate is high and qualification rate is below 60%, the qualification flow is too long or the agent's opening prompt is confusing. Pull the transcripts at the dropout point and shorten the question set. If both rates are strong, let the booking rate data accumulate for seven days before drawing conclusions about scheduling gaps.

The Wharton 2025 AI Adoption Report found that 72% of enterprise leaders formally measure Gen AI ROI. Law firm intake is one of the more measurable AI deployments available: every contact is a logged event, every qualification is a data point, and every booking ties directly to a consultation that either converts to retained or does not. The measurement framework is not complicated; it requires consistent logging from day one.

Track these three from the same inbound follow-up framework to measure how consultation requests convert to retained clients after the intake step. The intake agent's output is the follow-up agent's input.

What should stay with a human in law firm intake?

The question is not whether to automate intake but which parts to automate. Some elements carry ethical and relationship stakes that an agent cannot substitute.

Initial legal advice. An AI intake agent collects facts. It does not analyze them. A caller who asks "do I have a case?" gets a consistent response: "The attorney will review the details you have shared and answer that question in your consultation." The agent does not offer opinions on the merits of any matter, ever.

Conflict determination. The agent collects the names needed for a conflict check. The determination itself is the attorney's responsibility. Do not attempt to automate the conflict screening step with a database comparison unless a licensed attorney has reviewed and approved the technical process under the firm's applicable bar rules.

Fee discussion. Contingency percentages, retainer amounts, and billing structures are not topics for the intake agent. Any caller who raises fees gets a redirect: "The attorney will go through the fee structure with you during your consultation." The agent books the meeting; it does not negotiate the terms.

Any caller in distress. A caller in immediate danger, describing an ongoing safety situation, or clearly breaking down during the call gets routed to a human within two turns. No intake flow overrides a person in crisis. This escalation rule is configured before the agent goes live, not added later.

If your firm is ready to see which intake step applies first, get your free AI System Plan and walk through the qualification logic with your practice area in focus.

Methodology

This post draws on four primary sources. Access-to-justice statistics are from the Legal Services Corporation Justice Gap Report, a federal government survey on unmet civil legal needs. AI capability benchmarks reference the GPT-4 Technical Report (OpenAI, arXiv:2303.08774, 2023) and the LegalBench benchmark (Guha et al., Stanford, arXiv:2308.11462, 2023), which tested 20 large language models across 162 legal reasoning tasks. Adoption data is from the Wharton 2025 AI Adoption Report. The illustrative cost example uses hypothetical round numbers explicitly labeled as such and is not a client result. The target keyword "ai agent for law firm intake" appears in the title, lead paragraph, one H2, and this section.