AI support for law firms, intake and client comms without giving advice.

Pull the last quarter of inbound at a typical boutique or mid-market firm and the shape is the same across practice areas. Roughly ninety percent of the calls, web forms, and chat sessions are intake. A prospective client describing a situation, asking whether the firm handles their type of matter, asking how much it costs, asking how long it takes, asking whether the partner takes a free consultation. The remaining ten percent is existing client comms (status updates on open matters, scheduling, document requests, billing questions). Almost none of the inbound is a real substantive legal question, because clients who already retained you know to email their attorney directly. The intake queue is not a help desk. It is a qualifying funnel that runs twenty-four hours a day whether the firm is staffed for it or not.

The math on intake is brutal. Out of every hundred prospective client inquiries that hit a small firm, somewhere between fifteen and twenty-five are real qualified leads (right practice area, viable matter, in-jurisdiction, not conflicted, able to pay). The rest is noise. Wrong jurisdiction, wrong practice area, looking for a free consultation on a matter the firm would never take, calling about a matter against an existing client, calling to ask the partner to take their cousin pro bono. A receptionist or an associate spends real time triaging that noise to find the real leads, and the firms that staff intake well convert at three to five percent on qualified inbound. Firms that leave it to the website form and a voicemail box convert at under one percent and the rest of the qualified leads call the firm down the street.

The reason most firms underinvest in intake is that the dollars do not pencil for a full-time hire. A dedicated intake coordinator at sixty thousand fully loaded is a meaningful line item for a fifteen-attorney boutique, and the partners can never quite agree on whether the role should sit under marketing, business development, or the office manager. So intake stays distributed across whoever picks up the phone, the youngest associate handles the after-hours web form, and the firm leaks qualified leads every weekend, every holiday, and every Tuesday afternoon when the office is in court. We wrote about why the founder ends up holding the after-hours queue in The 11 PM Support Queue. The legal version of that problem is the partner holding the Saturday voicemail at 9am Sunday before the wedding she is supposed to attend.

The unauthorized practice of law guidance is not a suggestion. Every state bar in the US and every Law Society in the Commonwealth jurisdictions draws a hard line between collecting facts and giving legal advice, and the line is drawn in the practitioner manual of every bar association that has thought about chatbots since 2023. The AI can describe the firm. The AI can explain the practice areas. The AI can collect the facts of the situation. The AI can run a preliminary conflict check against the matter database. The AI can schedule a consultation with the attorney. The AI cannot tell the prospect whether they have a case. The AI cannot opine on the merits. The AI cannot quote a statute as guidance or suggest a course of action that constitutes legal advice. Every prompt chain in the system is engineered around this constraint before the first reply leaves the queue.

The engineering of UPL safety is the part most generic legal chatbots get wrong. A chatbot trained on a help-desk template will answer the question the prospect asked because that is what help-desk chatbots do. A legal-tuned intake agent answers a different question. The prospect asks "do I have a case for wrongful termination if my employer fired me last week." The generic bot starts pulling the elements of wrongful termination in the prospect jurisdiction. The legal intake agent says, in the firm voice, that wrongful termination is something the firm handles, asks the structured questions the partner needs answered to triage the matter (employer name, date of termination, jurisdiction, whether the prospect signed an arbitration agreement, whether they were a member of a protected class), schedules the consultation, and ends the conversation with the line every UPL-safe agent ends with: the partner will assess the merits during the initial consultation.

The defensibility of this distinction is what makes the system installable at a real firm. Every prompt chain is auditable. Every reply that goes out is logged with the version of the system prompt that generated it. The managing partner can sit down with the bar compliance officer at any time and walk through the actual conversations the system had, the actual questions it answered, and the actual line it drew at the merits assessment. There is no "we hope nobody asks." There is a written prompt policy signed by the partner of record, the same way the hybrid architecture for matter content is documented and signed before any agent touches a privileged communication. UPL guardrails are not an afterthought. They are the first thing engineered into the intake agent before the first prospect ever talks to it.

7am, Tuesday: a prospective client opens the chat widget on the firm website asking about an employment dispute. The agent identifies the practice area (employment law), confirms the jurisdiction (California), collects the structured facts (former employee, terminated three weeks ago, allegations of retaliation), runs a preliminary conflict check against the matter database (no conflict on the named employer), confirms the prospect understands the consultation fee, and offers three open slots on the employment partner calendar. The prospect picks Thursday at 2pm. The booking syncs to the matter intake CRM. The partner opens a one-screen brief on Wednesday afternoon with the facts already collected. Time from first message to scheduled consultation: four minutes.

11am, Wednesday: an existing client of the litigation practice sends an email asking when the response to the motion to dismiss gets filed. The agent reads the email, pulls the matter from the management system, sees the filing deadline (next Tuesday) and the current status (under attorney review), drafts a reply confirming the deadline and noting that the attorney will share the draft for client review by Friday. The reply lands in fifty-two seconds. The associate handling the matter sees the message in her shared inbox and approves it with one click. The client gets a confident answer the same morning instead of waiting for the associate to break from depo prep.

3pm, Wednesday: a prospective client calls the main line asking whether the firm takes a personal injury case from a car accident the prospect was in last weekend. The agent (handling the transcribed call) identifies that personal injury is outside the firm practice areas (firm is transactional and litigation only, no PI), declines courteously, explains that the firm does not handle PI matters but can suggest the prospect contact the state bar referral service, and logs the inquiry with the decline reason. No partner time burned. No false hope offered to the prospect. The intake CRM tracks the decline so the marketing team can see which queries are landing on the firm that should be routed elsewhere.

11pm, Saturday: a worried small business owner opens the web form asking about a contract dispute with a vendor. The agent collects the facts (verbal contract, partial payment dispute, jurisdiction in Delaware), runs the preliminary conflict check (no conflict on the named vendor), confirms the consultation fee, and offers slots on Monday morning. The prospect books a 9:30am Monday consultation. The transactional partner walks into the office Monday with a qualified consultation already on the calendar instead of an empty Monday morning and a Saturday voicemail to triage. The wedding the partner attended Saturday night did not get interrupted.

The economic reality of the legal market is that prospective clients shop around. The wrongful termination matter that lands in the web form on Saturday afternoon is also landing in the web form of three competitor firms in the same metro at roughly the same time. The firm that responds first with a qualified intake interaction wins the consultation on Monday morning. The firms that respond on Tuesday afternoon with a generic auto-reply lose the matter to the firm that already booked the consultation. The intake function is not a customer service function. It is the first stage of a competitive process that decides whether the firm captures the matter or whether the matter ends up on a competitor billing record.

On a fifteen-attorney boutique with an average matter value of fifteen thousand dollars, a three-percent conversion lift on the qualified inbound pipeline is the difference between a flat year and a year that funds the next associate hire. Most firms running the receptionist plus web form configuration convert under one percent on after-hours inquiries. Most firms running the fractional AI Support Department under attorney supervision convert three to five percent. The compounding effect over a year on a firm with two hundred qualified inquiries per quarter is meaningful. The dollars saved on the receptionist shift pattern are not the headline number. The dollars added to the matter book are.

The defensibility argument compounds the conversion argument. A firm with a written intake policy, audited prompt chains, attorney-routed escalation, and matter-management integration is a firm that can pass a managing partner audit, a malpractice insurance review, and a state bar inquiry without flinching. A firm running an ad-hoc receptionist plus voicemail configuration has none of that documentation when something goes sideways. The same hybrid posture that makes the matter content side defensible to the bar makes the intake side defensible to the bar. Read the technical breakdown of the on-device matter rail in Local Agent Setup, and the day-to-day operations layer in AI Ops for Legal.