How to sell AI systems to personal injury law firms
Personal injury firms buy AI intake faster than most trades, because one missed call is a lost case. But a phone agent at a law firm is not a receptionist. Here are the intake rules, the bar rules, the pricing, and the builds you should refuse.
Lumina14 min read
Personal injury firms are the easiest legal vertical to sell an AI intake system into, and the one where a sloppy build can get your client in front of a disciplinary board. The sale is easy because a PI firm already knows what a missed intake costs: they price cases, they track signed versus lost, and they can tell you which referral sources go cold when nobody picks up. The risk is that a phone agent at a law firm is not a receptionist. It is collecting facts from someone who may become a prospective client the moment they start talking, and the confidentiality duty attaches whether or not the firm ever takes the case.
So the pitch is not "AI answers your phone." The pitch is "no intake call goes to voicemail, every caller gets screened against your criteria, and the agent is built to gather facts without giving advice." That last clause is the whole sale. Build it any other way and you are selling a liability.
Here is how to run the deal: what the agent has to do, which rules govern it, what to charge, how the demo goes, and where this is the wrong product.
Why do personal injury firms buy AI intake faster than other trades?
Three reasons, and only one of them is about technology.
First, case value. A plumbing company loses a service call when the phone rings out. A PI firm loses a case. Do not use someone else's average case value here, because the partner across the table knows theirs. Ask for it. The arithmetic does the selling.
Second, the calls arrive at the worst hours. Accidents do not respect office hours, and the person calling at 11pm has usually just been handed somebody else's business card. Speed to contact is the edge in this practice area and the firms know it. The Lumina law firm pack is written around that case: intake at 11pm for the accident that happened at 10.
Third, the profession is past the "is AI real" argument. Thomson Reuters surveyed more than 1,500 respondents across 27 countries for its 2026 AI in Professional Services Report and found organization-wide AI use had almost doubled to 40%, up from 22% the year before, with agentic tools appearing in the data for the first time at 15% adoption and another 53% planning or considering. Clio's 2025 Legal Trends Report, based on 1,702 US legal professionals, found 36% saying AI had a positive influence on revenues, rising to 69% among wide adopters.
You are not walking into a room where you have to explain that AI works. You are walking into a room where the partner has already tried ChatGPT, formed an opinion, and is worried about the wrong things.
What does the intake agent actually have to do at a PI firm?
Write this list down before your first call, because a generic receptionist demo will lose to a specific one every time.
Answer on the first ring, any hour. Not a callback promise. Answered.
Screen against the firm's criteria, in the firm's order. Most PI firms have a short set of gates: what happened, when, where, was there an injury, was there treatment, is there insurance, has the caller signed elsewhere, and is the statute of limitations close. That last gate matters. A two-year-old accident in a one-year state is a decline, and the agent should route it differently from a fresh case.
Run a conflicts pass on names. The agent collects the caller's name and the other party's name, checks them against whatever list the firm keeps, and flags a hit instead of booking a consult.
Gather facts without advising. No opinion on whether the case is good. No number. No "you're probably entitled to." The agent takes the facts, says a lawyer will review them, and books the slot.
Route by urgency. A new injury with a police report and a hospital visit is not the same lead as a general question about a fee agreement. One pages the intake attorney, one lands in a queue.
Text the caller a confirmation. The follow-up text is often what stops the caller from working down the list of billboards. Log the intake into whatever the firm already uses, not a spreadsheet you built.
That is the build. Almost all of it is configuration rather than engineering, which is why it sells at a wholesale build price instead of a development retainer.
Which rules govern an AI intake agent at a law firm?
This is the part almost nobody selling into legal has read, and it is where you beat competitors who all promised "fully automated intake."
The caller becomes a prospective client, and confidentiality attaches immediately
ABA Model Rule 1.18, which most states have adopted in substance and which Louisiana reproduces in full, defines a prospective client as "a person who consults with a lawyer about the possibility of forming a client-lawyer relationship with respect to a matter." Paragraph (b) then says that "even when no client-lawyer relationship ensues, a lawyer who has learned information from a prospective client shall not use or reveal that information" except as the former-client rule permits. Comment [3] adds that the duty exists "regardless of how brief the initial conference may be."
Comment [2] is the one that should change how you build the intake script. A consultation is likely to have occurred where a lawyer, in person or through advertising, "specifically requests or invites the submission of information about a potential representation" without "clear and reasonably understandable warnings and cautionary statements that limit the lawyer's obligations." An AI agent that opens with "tell me what happened" is doing exactly that, at scale, for every caller.
Two build consequences. The agent needs a short cautionary statement at the top of the call, in the firm's words, cleared by the firm. And Comment [4] says a lawyer should "limit the initial consultation to only such information as reasonably appears necessary" for deciding whether to take the matter. That is an argument for a tighter script, not a longer one. Collecting more than the firm needs creates disqualification exposure on cases the firm never even took.
The firm stays responsible for what the agent says
The ABA issued its first formal opinion on generative AI on July 29, 2024. Per the ABA's own announcement, Formal Opinion 512 holds that the rules on competence, informed consent, confidentiality and fees principally apply, that lawyers should understand "the benefits and risks associated" with the technologies they use, and that the confidentiality duty covers information relating to the representation "regardless of its source, unless the client gives informed consent." Other rules extend similar protection to prospective clients' information, which lands right back on Rule 1.18.
California went further. The State Bar's 2026 Practical Guidance for the Use of Generative Artificial Intelligence in the Practice of Law replaced the 2023 version specifically to address agentic AI, and it names "autonomously facilitating client intake" as an example. The operative line: lawyers "must not deploy agentic systems in a manner that allows the system to make substantive legal determinations, communicate legal advice, prepare and file pleadings, or otherwise act in a representative capacity" without meaningful lawyer supervision and review. It also says the greater the system's autonomy, the greater the lawyer's obligation to build oversight, and that a lawyer "must not deploy an agentic AI system in a manner that permits autonomous external transmission of client information" without safeguards and human review. That guidance binds California lawyers; if your client is elsewhere, the direction of travel is still worth showing them.
None of that prohibits what you are selling. It prohibits the version where the agent evaluates the case. The difference is a scripting decision you make on day one.
Inbound is fine. Outbound to accident victims is a different rule entirely
Know this one cold, because a PI firm will eventually ask you to build it. Model Rule 7.3(b) bars soliciting professional employment by live person-to-person contact where a significant motive is pecuniary gain. Comment [2] defines that contact as "in-person, face-to-face, live telephone and other real-time visual or auditory person-to-person communications" where the person is "subject to a direct personal encounter without time for reflection." An AI voice agent dialing crash reports is a live telephone communication with someone the firm knows needs legal services. Comment [6] singles out contact with people "especially vulnerable to coercion or duress" as ordinarily not appropriate.
The inbound side is explicitly clear. Comment [1] says a communication is not a solicitation "if it is in response to a request for information or is automatically generated in response to electronic searches." Someone calling the firm's number is requesting information. Comment [2] also carves out "chat rooms, text messages or other written communications that recipients may easily disregard," which is why a website chat agent sits on safer ground than a dialer.
So the answer to "can your AI call the accident report list for us" is no, and you should say no in the room. Then sell them the version that is allowed. Refusing one feature is the most credible thing you will do on that call. For the disclosure side of the same question, we wrote up whether you have to tell callers it's an AI separately.
How do you price an AI intake build for a law firm?
Higher than a trade build, for a reason you can defend. Your cost through Lumina is the same as any other vertical. A Top Tier buildout with voice runs $1,150 to $1,400 setup and $315 to $460 a month, plus $0.35 a minute on voice, agreed on a 30-minute call before anything gets built. Line-by-line, the same work prices at $1,730 to $2,280 setup. The pricing page has the full tier table.
What you charge is your call. Three things justify a higher retail number in this vertical than in HVAC:
Their unit economics are different. One signed case usually covers a year of the system. Ask for their number and let them do the multiplication out loud, in the room.
Intake staffing is expensive and leaky. Many small PI firms pay an answering service that takes a message and nothing else, or a paralegal fielding calls between real work. You are replacing a cost, not adding one.
The compliance work is real. The cautionary statement, the conflicts flag, the statute-of-limitations gate, the escalation rules. That is judgment, and it is worth being paid for.
Skip the separate build fee on top. The reason the no-build-fee structure shortens the sale is that the partner is not signing a capital expenditure, they are adding a line item. We laid out the full version of that argument in how much to charge for an AI buildout.
One thing to raise before they ask: California's guidance says subscription fees for general-purpose AI tools are ordinarily overhead to be absorbed in the lawyer's fee, while costs "specifically incurred for the client's matter" can be charged through with disclosure and no markup absent informed written consent. Your system is firm infrastructure. Tell them it is overhead so they do not go looking for a way to bill it to a client file.
How does the demo call go?
Differently than with a trade, in one specific way: the partner will test the agent's judgment, not its voice quality.
Build the demo on their firm name and practice area before the call. Open the live demo, hand them the phone, and stop talking. They are going to try to break it, which is what you want. Three attempts are near certain:
"Do I have a case?" The agent declines and says a lawyer will review. If it hedges toward an answer, you lose the room. Test this yourself twenty times before the call.
"How much is my case worth?" Same. No number, ever.
A fact pattern that is out of scope. A workers' comp question at a car accident firm, or a case already signed elsewhere. They are checking whether the agent recognizes a decline and routes it politely.
When it handles those three correctly, the conversation moves to their intake criteria, which is where you want it. Our demo call playbook covers the general structure, and what's inside a vertical pack explains why the law firm pack sounds sober instead of chirpy.
One note on the demo itself. The page states that the AI is a demo and that bookings inside it are simulated. With a lawyer, that disclosure helps you. It is the first evidence that you build things that say what they are.
Where this is the wrong sale
The honest part, because the version of this post without it is worthless to you.
Firms with no intake criteria. If the partner cannot tell you in five minutes what they accept and reject, there is nothing to configure. The agent will collect facts nobody screens and you will get blamed for bad leads. Walk, or sell them the website and chat tier first and revisit.
High-volume advertising firms. A firm running heavy TV or paid search with a staffed intake department has already bought something, usually custom and wired into their case management system in ways a productized build will not match. You are competing against a sunk investment. There are far more three-lawyer firms than billboard shops.
Practice areas where the first call is advice. Some legal work does not separate cleanly into intake and advice. Criminal defense at 2am is the obvious case. The agent can take a message and page a human, which is useful, but it is a smaller product than what you are pitching and you should price it as one.
Anyone who wants the agent to make the call on cases. If the partner's actual goal is to stop reviewing intakes, say no. Both the California guidance and plain professional judgment put that decision with a lawyer. A client who wants the thing the rules prohibit is a client who will blame you when it goes wrong.
The platform has its own limit worth stating plainly. The $49 plan includes 100 live voice minutes a month, 500 chat messages, and 50 website audits, with a five-minute cap on any single call and one published demo at a time. That is roughly forty demo calls. Plenty for prospecting, not a production system. Client voice minutes on a delivered build are separate and billed per minute. The FAQ covers what happens at the cap.
What actually gets delivered
Once they sign, you order the build through the fulfillment platform and onboarding collects the inputs. For a law firm that list is specific: practice areas accepted and declined, the intake question order, the cautionary statement the firm wants read, who gets paged and when, the conflicts list, hours and holiday coverage, the calendar, and the case management system the intake lands in. Get the cautionary statement approved by the partner in writing. It is the one line in the build with their bar card behind it.
Everything the client sees carries your branding. The verticals list includes packs for personal injury and family law firms alongside the other hundred trades.
FAQ
Can an AI intake agent give legal advice?
No, and you should build it so it cannot. California's 2026 guidance says lawyers must not deploy agentic systems that communicate legal advice or make substantive legal determinations without meaningful lawyer supervision and review. The agent's job is facts in, consult booked, lawyer decides.
Does a caller who talks to the AI become a prospective client?
Treat it as yes. Model Rule 1.18 attaches a confidentiality duty to information learned from a prospective client even when no engagement follows, and the comments say that duty holds however brief the conversation was. Build the script to collect only what the firm needs to decide on the matter, and put a cautionary statement at the top.
Can the agent make outbound calls to accident victims?
No. Model Rule 7.3(b) bars soliciting employment by live person-to-person contact for pecuniary gain, and the comments define live telephone contact as exactly that. Inbound calls are fine, because a communication made in response to a request for information is not a solicitation.
What if my client is not in California?
The California guidance binds California lawyers. Rule 1.18 and Rule 7.3 are ABA model rules that most states have adopted in some form, with real variation, so the answer for a specific firm is theirs to confirm with their own bar. Your job is to raise the question, not to answer it for them. You are not their lawyer.
How long does the build take?
Your demo is live in about ten minutes. The client build runs on the timeline agreed on the 30-minute pricing call, and the thing that decides it is how fast the firm returns their intake criteria and approves the cautionary statement. Firms that assign one person to that finish quickly. Firms that route it through a partners' meeting do not.
Start here
Pick one PI or family law firm in your market with two to five lawyers and no visible after-hours intake. Call their main number at 8pm and listen to what happens. That is your opening line.
Then build the demo on their firm name, call it yourself, and ask the three questions the partner will ask. If it declines all three cleanly, send the link. You can start the agency workspace for $49 a month, and the platform price goes up on October 1, so joining before then locks the current rate.
