The work a firm is proud of — research, drafting, argument — happens after engagement. Whether there is an engagement at all is decided earlier, by whoever answers the phone and by how long an inquiry sits before anyone responds. That is intake, and at most firms it gets less design attention than the letterhead.
What intake has to capture, precisely
An intake conversation is not a message pad. Before any substantive discussion, someone has to record the caller's full name and the adverse party's name, because a conflicts check must run before the conversation goes further — hearing a prospective client's confidences and then discovering the firm represents the other side is a problem that arrives fully formed. The date of the incident matters because a limitations period runs from it, and a firm that sat on an inquiry until a deadline passed has converted a missed call into an exposure. Then the matter has to be sorted: contingency work worth evaluating, hourly work the firm handles, or a referral out. Declined matters need a non-engagement letter, in writing, promptly. The caller who believed the firm was handling it is a malpractice claim with a return address.
What the research says about speed
The best available study is old but audited. In 2011, Harvard Business Review published research by Oldroyd, McElheran, and Elkington that tested 2,241 US companies with web inquiries. The average first response took 42 hours. Twenty-three percent never responded at all. Companies responding within an hour were roughly seven times more likely to qualify the lead than those who waited even an hour longer. The honest caveat: the data is fifteen years old and measured business leads, not legal consumers, so the multipliers should not be treated as facts about your practice area. The mechanism has not changed, though. A person who has decided they need a lawyer keeps calling names until one answers.
The inquiry arrives after your staff leaves
People call employment lawyers the evening they were walked out. They call about a wreck from the hospital that night. The moments that generate matters are indifferent to office hours. An intake agent built for this answers every time, captures the conflict-check fields as structured data rather than a voicemail to decipher, records the incident date, sorts the matter type, and offers the earliest consultation slot on the real calendar. It is also built to refuse the one thing it must always refuse: legal advice. The caller gets a scheduled consult. The firm gets a transcript and a completed intake sheet waiting in the queue at 8 a.m.
Why intake notes cannot go into a public chatbot
This stopped being theoretical in February 2026, when Judge Rakoff in the Southern District of New York held that documents a defendant had generated on the consumer tier of a public AI platform — analyzing his own legal exposure — were protected by neither attorney-client privilege nor the work-product doctrine. The platform's terms, which permitted data retention and disclosure, negated any reasonable expectation of confidentiality. Professional guidance points the same direction. ABA Formal Opinion 512, issued July 2024, requires informed client consent before entering information relating to a representation into a self-learning generative AI tool, and says boilerplate engagement-letter language is not enough. Texas Opinion 705, from February 2025, adds that a lawyer must understand the tool, verify its output, and cannot bill for the time it saved.
What a private deployment changes
The architecture we build runs intake and document work on infrastructure the firm controls. There is no third-party platform whose terms of service, retention practices, or litigation become the firm's problem — a live concern, since in 2025 a federal magistrate judge in the New York Times copyright case ordered OpenAI to preserve user conversation logs, including chats users had deleted, while that dispute played out. Records on your own hardware answer to your retention schedule and nobody else's discovery posture. That is an architectural fact, not an ethics opinion. The obligations under your state's rules remain yours, and the design should be reviewed against them before anything runs.
Where this is the wrong tool
An agent cannot decide which cases to take, and automating a bad intake criteria sheet produces bad decisions faster. Conflicts checking is only as good as the names in the conflicts database — an agent queries it; it cannot repair it. Some practices need a human on call regardless: criminal defense does not get to route an arraignment-hour call into a scheduling flow. And every output that touches a client decision — the engagement, the declination, the non-engagement letter — needs a lawyer's review before it leaves the building.
How you would measure it
Pull last month's phone log and web inquiries. Count three things: inquiries that arrived outside staffed hours, the elapsed time from each inquiry to first substantive contact, and signed engagements as a share of total inquiries. Those numbers, taken before and after, are the whole evaluation. No vendor's claim substitutes for them.
Where to start
Mark every call in the last thirty days that hit voicemail, and every web form answered later than the next morning. Put that list next to your average fee per matter and decide whether the gap justifies an hour's conversation.