A U.S. injury firm's intake now runs 24/7, in two languages.

We designed, built and operate a multi-agent AI platform that runs the entire lead-intake operation of a U.S. personal-injury law firm: website chat, qualification, e-signed retainer, CRM case, follow-up of leads the team couldn't reach, and AI-assisted case work. Live in production since 2025.

Get this for your firm Technical write-up

24/7Intake, in English and Spanish

3Cooperating AI workers

0-touchRetainer signing

100%Server-side ad events accepted in the latest audit

The problem

Leads were slipping out of a full ad budget.

Personal-injury firms live and die by how fast they respond to a new lead. Before this project, the firm was paying for ads on Google, YouTube and Meta, but:

Website visitors abandoned intake forms and nobody followed up.
The intake team closed leads as "no contact" when they couldn't reach them, even when the claim was still viable and inside the statute of limitations.
Ad spend couldn't be traced to signed cases, so nobody knew which campaigns worked.
Every follow-up call and letter was done by hand.

What we built

Three AI workers and one command center.

Built on the firm's own AWS account, so the firm owns the system, the data and the code.

AI WORKER 01

Intake chatbot

A bilingual (English/Spanish) intake conversation on the firm's landing pages that:

Captures calling consent word for word, with a timestamp
Flags red-flag cases from the caller's own description
Generates and e-signs the retainer, then opens the CRM case
Recognises returning leads instead of creating duplicates

AI WORKER 02

Lead-chase engine

Goes after the leads intake gave up on. An AI reads each lead's CRM status and case notes and decides: chase, stop, or send to a paralegal, with a one-line reason.

16-day sequence across email, SMS and voice
A person approves before calls go out
Any AI failure routes the lead to a human, never silently drops it

AI WORKER 03

AI paralegal

A 10-stage pipeline that takes signed cases to first notice of loss, including letters of representation generated from firm-approved templates, with rules that keep injury details and ID numbers out, and delivery by email or fax per insurer.

COMMAND CENTER

Operations dashboard

Live pipeline and conversion stats, "cold list" alerts, a full view of every lead (chat transcript, emails, call recordings and summaries, CRM link), and the ad that produced each lead.

Compliance, built in

If a lawyer asks "can you prove it?", the system can.

Consent is a hard gate. A lead without recorded consent cannot be called, whatever else happens upstream.
Calls are predictable by design. Follow-up calls use a keypad menu, not open-ended AI speech, so an opt-out is always an explicit, logged keypress and is honoured instantly.
Every call leaves a record. Recording, transcript and a time-stamped PDF summary are attached to the case file automatically.
Calls come from the firm's own numbers, so they're answered as the firm, not flagged as spam.
Everything ships switched off. Kill switches and approval gates meant each part went live only after the firm's counsel signed off.

Deeper detail: voice engine case study.

Attribution

Every signed case traced back to the ad.

Each qualified lead sends a server-side event to Meta's Conversions API, matched to Meta's specification and de-duplicated against the browser pixel. Every lead is stamped with its source (Google, YouTube, Meta, TikTok, lead vendor, CRM import or direct) and the ad name that produced it.

The latest pipeline audit confirmed 100% of server-side events accepted (42 of 42). When lead volume once dropped, the evidence trail showed within days that the cause was a campaign setting, not the intake system.

Deeper detail: ad attribution case study.

Results

What changed for the firm.

Intake runs 24/7 in two languages, with retainers signed without anyone touching them.
Leads the team had given up on are re-engaged systematically, within consent and statute-of-limitations rules.
Marketing spend is traceable end to end: ad → landing page → chat → signed case → CRM.
Run in production by a two-person team with a clean incident record.

Stack

Amazon Lex V2Amazon Bedrock (Claude)AWS LambdaDynamoDBAPI Gateway + CognitoEventBridgeSESAmazon TranscribeTwilio VoiceMeta Conversions APIReact dashboard

Client details are kept confidential. Built by Kamogelo Mahlasela and Masego Letsoko.

Keep reading

Fixed-price proposal in 48 hours

Want this running at your firm?

Tell Nectar how your intake works today. You get a fixed-price proposal within 48 hours, and the system you get is yours.

Chat with Nectar business@gv-va.com