Template: Rules of Engagement for Your Legal Intake Bot

Two of the biggest names in legal AI launched "agentic front doors" for in-house teams on back-to-back days. Legal IT Insider reported on October 13 that Legora's Intake is live in Slack, where employees tag Legora to turn a request into tracked work, sometimes with a first draft built from the team's playbooks. Teams, email and MCP are planned. On October 14, Artificial Lawyer reported that Harvey's Intake begins early access this month. It routes requests to the right agent or lawyer, resolves routine ones automatically and runs a first-pass contract review against the team's playbook. Harvey says it designed the product with in-house teams at Carvana, GE Aerospace and Dentsu.

Whichever tool you pick (or already have), an intake bot that answers the business on legal's behalf needs rules before it goes live. Here's a ready-to-use framework. Copy it, fill it in, and get sign-off from your GC before turning anything on.

Part 1: The Routing Matrix

Sort your top 20 request types into three lanes. Use real data from the last 90 days of email and chat requests, not guesses.

• Green lane (bot answers alone): Policy lookups, signature authority questions, "which NDA template do I use," status checks on existing requests. Rule: the answer must come from an approved source, and the bot must cite it.
• Yellow lane (bot drafts, lawyer approves): Standard NDAs, low-value vendor agreements on your paper, routine marketing review. Rule: nothing goes back to the requester until a named lawyer signs off.
• Red lane (straight to a human): Litigation threats, regulator contact, employee complaints, anything involving a potential breach, M&A, or anything the requester labels urgent and sensitive. Rule: the bot acknowledges receipt and routes it. It does not summarize, advise or draft.

Part 2: The Playbook Readiness Checklist

An intake bot drafting from a bad playbook is just a faster way to be wrong. Check every box before Yellow lane goes live:

• ☐ Each Yellow lane document type has one approved template.
• ☐ Fallback positions and walk-away points are written down, with a one-line reason for each.
• ☐ Approval thresholds (value, risk, jurisdiction) are explicit numbers, not vibes.
• ☐ Every playbook has a named owner and a review date.
• ☐ Superseded templates are archived where the bot can't reach them.

Part 3: The Escalation and Privilege Rules

1. Trigger words: List terms that force Red lane routing (e.g., "subpoena," "lawsuit," "regulator," "breach," "harassment," "whistleblower").
2. Confidence floor: If the tool offers a confidence or uncertainty signal, set a threshold below which requests go to a human.
3. Channel hygiene: Decide whether sensitive requests may be raised in shared chat channels at all. If not, the bot should move the conversation to a private channel or form.
4. Privilege posture: Work with your litigation and privacy teams on how bot conversations are labeled, retained and treated for privilege. Don't assume a chat with an AI tool is privileged. Courts are still working through these questions, and the answer may depend on the facts.
5. Retention: Align bot logs with your records retention schedule and your matter management system.

Part 4: The Requester Disclosure (Copy and Paste)

"You're chatting with Legal's automated intake assistant. It can answer common questions from approved Legal sources and route your request to the right person. It does not replace advice from a lawyer on your specific situation. For urgent or sensitive matters (litigation, regulators, investigations, or employee concerns), say so and you'll be routed to a lawyer directly."

Part 5: The 30/60/90 Scorecard

Track these from day one:

• Deflection rate: Share of requests resolved in Green lane without a lawyer.
• Override rate: Share of Yellow lane drafts that lawyers materially rewrote. If it's high, fix the playbook before you blame the bot.
• Misroutes: Count of Red lane matters that started in Green or Yellow. Your target is zero, and every one needs a root-cause review.
• Time to first response and time to close, compared with your pre-launch baseline.
• Requester satisfaction: One question, one click, after each closed request.

Go / No-Go

Green: Routing matrix signed off, playbooks ready, zero misroutes in a two-week pilot. Expand.
Yellow: Override rate above your threshold, or gaps in the playbook. Keep piloting Green lane only.
Red: Any Red lane misroute you can't explain. Pause and fix the triggers.

Compare legal intake and workflow tools and contract lifecycle management platforms on CorporateLegal.tech to see who supports these controls.

This article is general information, not legal advice.