By Oliver · AI Architect, BuildAClaw · Sep 9, 2026 · 10 min read

How to Build an AI Agent That Screens New Client Intakes for a Law Firm

The average small law firm loses 38% of inbound leads before an attorney ever hears about them. Here's how to build a local AI agent that screens, qualifies, and routes every intake call in under 90 seconds — without sending a single case detail to the cloud.

Most solo and small law firms run intake the same way they did in 2015: a paralegal or receptionist answers the phone, scribbles notes on a pad or types them into a CRM field, and hopes the right facts make it to the right attorney before the caller hangs up and calls the next firm on Google. It doesn't work. Attorneys I've talked to estimate they lose a third or more of qualified leads simply because intake was slow, inconsistent, or handled by whoever happened to pick up the phone that day.

An AI agent fixes the structural problem, not just the speed problem. It asks the same qualifying questions every time, checks for conflicts before anyone gets excited about a new matter, and hands the attorney a clean summary instead of a voicemail. I've built this exact workflow for personal injury, family law, and immigration practices running on a Mac Mini M4 with OpenClaw as the orchestration layer — no case data ever leaves the building.

Why Cloud-Based Intake Tools Fail Law Firms Specifically

Generic AI receptionist tools exist — Podium, Ruby, various chatbot vendors — but almost none of them are built for the constraints law firms actually operate under. Attorney-client privilege attaches the moment a prospective client shares confidential facts, even before engagement. Routing that conversation through a third-party cloud API creates a discoverability and confidentiality exposure that most malpractice insurers flag immediately.

There's also the conflicts problem. A generic chatbot has no way to check an incoming caller's name, the opposing party's name, or related business entities against your firm's existing matter database before continuing the conversation. That check has to happen locally, against your live client records, in real time — not after the fact when someone's already spent 20 minutes on a call they should never have taken.

Step 1 — Define the Qualifying Criteria Before You Touch a Prompt

Before building anything, sit down with the managing partner and write out the actual intake rubric your firm uses today, even if it's informal. For most practices this comes down to five checks:

• Practice area match — does the matter fall inside what the firm actually handles?
• Jurisdiction — is the incident or matter in a state/county the firm is licensed and willing to work in?
• Timeline — is the caller inside or outside the relevant statute of limitations?
• Conflict check — does any named party match an existing client, opposing party, or related entity on file?
• Case value threshold — does the estimated value or fee structure clear the firm's minimum?

Every one of these needs to become an explicit rule in the agent's system prompt, not a vague instruction like "screen for good cases." Vague instructions produce inconsistent screening, which defeats the entire purpose.

Step 2 — Wire Up the Intake Channel

Most firms want this running on the phone line, the website contact form, or both. OpenClaw handles this through two connectors: a voice pipeline (speech-to-text into the agent, text-to-speech back out) for calls, and a webhook listener for form submissions. The agent doesn't need to sound like a human — callers respond well to a clear disclosure upfront ("I'm an automated intake assistant, an attorney will follow up") because it sets expectations and actually increases completion rates versus a caller who suspects they're being deceived.

Step 3 — Build the Conflict Check Against Your Live Database

This is the step most DIY chatbot builds skip, and it's the one that actually matters for a law firm. The agent needs read access to your case management system (Clio, MyCase, PracticePanther all expose an API) so it can run every named party mentioned in the intake — caller, opposing party, related businesses, co-defendants — against existing matters before the call ends.

Configure this as a hard gate: if the agent finds a potential match, it does not continue gathering case facts. It thanks the caller, tells them an attorney will call back to confirm details, and immediately flags the matter for manual review. This protects the firm from continuing a conversation that later has to be unwound for a conflict.

Step 4 — Route Qualified Leads With Full Context, Not Just a Notification

A ping that says "new lead, call John" is barely better than a voicemail. The agent should generate a structured summary — practice area, jurisdiction, timeline flags, estimated value, conflict check result, and a clean transcript — and drop it directly into the matching attorney's queue in your case management system. Some firms also want a Slack or SMS alert for anything flagged urgent, which the agent can trigger the same way an agency intake-to-delivery pipeline routes new client work to the right team member.

For firms already using an AI agent for contract review and document workflows, the intake agent can hand off directly into that pipeline once a matter is confirmed — the engagement letter and initial document requests go out automatically the same day, instead of sitting in a queue behind whatever the paralegal gets to next.

Step 5 — Build the Emergency Escalation Path

Some intakes can't wait for the normal review queue. A caller describing a recent arrest, an eviction notice, or a filing deadline inside 30 days needs to bypass everything and reach a live person. Hard-code trigger phrases and category flags that force immediate escalation — a text to the on-call attorney, a transfer to a live answering service, whatever your firm's emergency protocol already is. The agent's job here is narrow: recognize urgency and escalate, never assess or advise.

What This Actually Costs to Run

On a Mac Mini M4 running the full OpenClaw stack locally — reasoning, transcription, and the conflict-check logic — most small firms land between $60 and $140 a month in token spend, depending on call volume. That's running against the case management API directly, with no case data ever transiting a third-party cloud service. Compare that to a part-time intake coordinator at $2,000–$3,500/month, or a cloud-based legal chatbot subscription that still requires a human to re-key everything into the case management system afterward.

Frequently Asked Questions

Can an AI agent legally screen potential clients for conflicts of interest?

Yes, as a screening tool that flags potential conflicts for attorney review — it does not make the final conflicts determination. The agent cross-references intake names, opposing parties, and related entities against your existing client and matter database, then surfaces matches for a human attorney to confirm before engagement.

Does an intake screening agent count as giving legal advice?

No, if it's scoped correctly. The agent should only gather facts, categorize the matter type, and check basic qualifying criteria. It should never answer "do I have a case" — that line has to be hard-coded into the prompt and confirmed in testing before deployment.

How much does it cost to run an AI intake agent for a small law firm?

Most solo and small firms spend $60-$140/month in API tokens on a Mac Mini M4, plus the one-time hardware cost. That typically replaces a part-time intake coordinator role running $2,000-$3,500/month.

What happens if the AI agent misses a real emergency?

You build a hard-stop rule: any matter mentioning injury, arrest, eviction, or a filing deadline within 30 days gets flagged urgent and routed to a live attorney immediately, bypassing the normal queue. The agent should always err toward escalation.