Reference workflow · Illustrative
AI Product Development | Legal Intake

AI Legal Intake Agent

Small firms lose hours gathering the same intake facts, checking parties, organizing documents, and chasing missing information. This reference workflow shows how an AI intake agent could prepare each inquiry — while routing conflicts, uncertain deadlines, and consequential decisions to an attorney. Not autonomous legal advice — a structured, auditable matter ready for review.

Khoda Consulting CA Employment Law Human-in-the-loop
Qualified intakescreened · reviewedBooked
Potential conflictmatch · restrictedStopped
Uncertain deadlineno guess · citedEscalated
Same agent · three inquiries · three endings

Where a small firm loses hours before a case even starts

Every inquiry runs through the same first questions — what happened, when, who’s involved, do we even take this — work that needs structure, not a partner’s time.
Conflicts get checked late, by hand. A prospect’s employer can be a current client under a different name — and finding that out after intake wastes everyone’s time and creates real exposure.
Real deadlines hide in vague facts. Statutes and filing windows turn on dates and jurisdiction a prospect rarely volunteers — and a missed one can have serious consequences for the prospect and the firm.
After-hours inquiries hit a voicemail. People in stressful situations reach no one, and many never call back.

Three inquiries. Three different endings.

The same agent, three prospects. One gets qualified, conflict-checked, and booked. One trips a conflict and stops cold — no assessment, no scheduling. One hides a deadline the agent refuses to guess, so it cites the source and escalates instead. The restraint is the part worth watching.


AI agent · employment intake walkthrough
Inquiry received
“I was let go a few weeks after I raised a harassment complaint about my manager. It feels like payback.”
web form · employer: Westside Logistics Inc.
Minimal capture
Captures just enough to screen — party and employer
prospect: Dana R.employer: Westside Logistics Inc.only what’s needed to run a conflict check
Conflict pre-screen
Resolves the employer and screens related entities first
searched: employer + parent / related entitiesno probable match — clear to proceed
Detailed intake
Now collects the situation — and flags what’s missing
situation: harassment complaint → terminationmissing: jurisdiction + incident date → 2 follow-ups
Documents → timeline
Builds a chronology from the uploads
termination_letter.pdf · hr_complaint.pdfFeb 12 complaint → Mar 1 terminationjurisdiction confirmed: California
Firm-fit → attorney review
Scores fit, then a lawyer approves
✓ CA jurisdiction✓ matter type accepteddisposition: attorney reviewmatter approved
Verified handoff
Books, files, and confirms — then verifies each action landed
consultation bookedCRM record + document folder createdconfirmation sent and verified
Booked and verified — appointment set, matter EMP-2026-0051 opened, CRM record and document folder created, confirmation sent and verified.
Illustrative workflow using fictional data — not a client deployment.

The restraint is the point

Conflicts get checked first. The agent screens the employer — including parent and related entities — before collecting sensitive detail, and stops the moment it finds a possible match. Late conflict discovery wastes intake work and creates confidentiality and representation risk.
The AI never sets a deadline. Extraction is the model’s job; the filing window belongs to deterministic rules and a reviewing attorney. On incomplete facts the agent escalates instead of asserting — because a wrong deadline becomes the client’s problem, not a demo bug.
A human gate before anything commits. No consultation is booked and no matter opened until an attorney approves. Human review here is a maturity signal, not a fallback.
Evidence becomes a timeline. Uploaded documents are read into a chronology with discrepancies and missing items flagged — so the attorney’s first thirty minutes start ahead.
Every decision is auditable. Each step, source, and attorney action is logged, so the firm can see exactly why the agent did what it did.

One orchestrator, deterministic where it counts

A real version is one intake agent coordinating focused services — with deterministic rules, not the model, owning deadlines, permissions, and anything consequential. Built around a single firm’s practice area, conflict rules, and systems.


Intake
web · voice · email
Persistent Intake Orchestrator
owns workflow state across the matter
Minimal Party Capture
party + employer only
Entity Resolution + Conflict Pre-Screen
stop here before sensitive intake
Detailed Fact Intake
Claude — adaptive questions, facts only
Conflict Re-check
as new parties surface
Rules Engine + Sources
deadlines, deterministic
Document Pipeline
parse → timeline
Firm Policy & Risk Gateway
permissions · transitions · risk
Human Review Queue
attorney approves consequential actions
Calendar · CRM · E-sign · Payment
act only after approval
Outcome Verification
confirms each action landed
Audit & Evaluation Store
every step, source, and decision logged
One orchestrator coordinating specialized services — not a swarm of agents. Deterministic rules own deadlines, permissions, workflow transitions, and any consequential action.

Map your intake workflow

We’ll identify where AI, deterministic rules, and attorney approval should each enter your intake process — and what’s worth building first.

Start a conversation →