The AI Receptionist Decision for Law Firms: Smith.ai vs. Ruby vs. Rosie vs. Building It Yourself

The decision rule

If your firm takes fewer than about 60 new-matter calls a month and your intake is genuinely simple — practice area, contact details, calendar a consult — buy an AI-only plan and spend the savings on follow-up. If your calls are emotionally loaded, high-value, or arrive from people in crisis, keep humans on the first thirty seconds and let AI handle overflow and after-hours.

Almost every firm gets this backwards. They buy the human service for everything because intake feels too important to automate, then discover that 40% of the calls were vendors, wrong numbers, and existing clients asking for a case status.

What each option costs

Prices below are as published on each vendor's own pricing page, read in the first week of September 2026. Verify before you sign — this category re-prices often.

The single most useful number in that table is the gap between Smith.ai's two products. Same company, same intake workflow, same Clio integration — $2.00 a call when AI handles it, $10.00 a call when a person does. That is a 5x spread, and it exists because the vendor knows most intake calls do not need a person.

The billing-unit trap

Ruby bills per minute. Smith.ai bills per call. This is not a detail; for a law firm it is the whole cost model.

Legal intake calls run long. A personal injury screening call with a caller who is upset and telling the story out of order is routinely 10 to 15 minutes. On Ruby's Standard plan — $395 for 100 minutes — eight of those calls consume your entire monthly allowance. On Smith.ai's live plans you pay one flat call charge no matter how long the caller talks.

Flip it for a firm whose calls are mostly short. A transactional practice fielding 90-second "are you taking new clients" calls burns very little of Ruby's minute pool and would be overpaying badly on a per-call plan.

So: long calls favor per-call pricing, short calls favor per-minute pricing. Pull your phone records, compute your actual average handle time, and price both models against it before you talk to anybody's sales team.

What makes legal intake different from every other vertical

Generic AI receptionist reviews will not tell you any of this.

Conflicts. An AI that books a consultation before anyone has run a conflicts check has created a problem, not solved one. The workable pattern is: the agent captures adverse party names as structured fields and books a tentative slot that intake confirms after clearing conflicts. If a vendor's demo shows the agent confirming appointments outright, that is the demo they built for dentists.

Unauthorized practice of law. The agent must not answer "do I have a case," "how much is my case worth," or "how long do I have to file." Those are the questions callers most want answered, so the model will be strongly pulled toward answering them. Test this deliberately and adversarially — ask a live agent those three questions in the trial and see what it says.

Statute of limitations urgency. Some intake calls are time-critical in a way a booking agent has no concept of. If your practice area has hard deadlines, you need an escalation trigger on date language, not a generic "book them next Tuesday."

Privilege and confidentiality. Anything the caller says is likely privileged from the first sentence. That makes the recording, transcript, and retention policy a professional-responsibility question, not an IT question. Ask where transcripts live, how long, and who at the vendor can read them.

Who each option is genuinely wrong for

Smith.ai AI Receptionist is wrong for you if a meaningful share of your inbound is distressed callers. The AI plan handles logistics well; it does not read the room. It is also wrong if you are not on Clio, Salesforce, or HubSpot — the native integration is most of what you are paying for, and Zapier-only firms can get the same outcome cheaper elsewhere.

Ruby is wrong for you if your average call is long. The per-minute pool makes your bill a function of how talkative your callers are, which is the one variable you control least. It is also worth noting Ruby does not publish overage rates on its pricing page — get them in writing before you sign.

Rosie is wrong for you if you need the intake record to land inside Clio as a structured matter. It is a well-priced generalist that reaches your systems through Zapier, which means field mapping is your job and it breaks silently when something changes upstream.

Building on Vapi or Retell is wrong for you unless you have an actual reason the packaged products cannot serve — a case management system nobody integrates with, or a multi-office routing problem. At single-firm volume the platform cost is not the constraint; the maintenance is. See the build-versus-buy arithmetic for the numbers.

The recommendation

• Solo and small firms, simple intake: Smith.ai AI Receptionist at $150/mo. The Clio Grow write-back is the reason to pick it over Rosie, and it is worth the $100 difference the moment it saves you one re-keyed intake.
• PI, family, criminal — high emotional load: Live answering during business hours, AI for after-hours and overflow. Smith.ai's Starter at $300/mo plus their AI tier is a defensible stack; so is Ruby if your calls are short.
• High-volume intake shops (200+ calls/mo): Do the per-call math on Smith.ai Pro against a built agent. This is the one firm shape where building starts to make sense, because your volume amortizes the maintenance.
• Anybody on a $49 budget: Rosie, and accept that you own the Zapier plumbing.

What breaks in month three

The failure mode is not the AI mishandling a call. It is drift, and it is quiet.

The calendar fills with consultations that should never have been booked. Week one you read every transcript. Week three you stop. By month three the agent books anyone who asks, because nothing in the system distinguishes "wants to hire a lawyer" from "wants free advice," and nobody ever told it where that line sits in your practice. The tell is your consult-to-retainer rate falling while your consult count rises — which looks like success on the dashboard.

The intake script goes stale. You add a practice area, drop another, change your fee structure. The agent does not know. No packaged vendor prompts you to update it, and the model will confidently answer with last quarter's facts.

Nobody owns the escalation queue. Calls the agent could not handle route somewhere — an inbox, a voicemail, a Slack channel — and that destination is somebody's job for exactly as long as they remember it is.

All three have the same fix, and it is one recurring calendar entry: thirty minutes a month, listen to five random calls and every escalation. That habit is worth more than the difference between any two vendors on this page.

The objection you will hear

Pitch AI-only intake to a receptionist service's rep and you will get some version of: "Your clients are in crisis. They need a human."

It is a fair point and an incomplete one. It is true for the caller who was in an accident this morning. It is not true for the vendor call, the wrong number, the existing client asking when their hearing is, or the person calling at 9pm on a Sunday who currently reaches voicemail. Those are most of your volume, and the honest comparison is not AI against a human — it is AI against the voicemail those calls actually get today.

The reasonable answer is to route on intent rather than pick a side. Humans on the calls that need them, AI on the rest, voicemail on none.

What to ask on the demo

Send these before the call so they have to prepare:

• Show the agent running an intake for my practice area, including a caller who asks whether they have a case.
• Show me the record it creates in Clio — the actual fields, not a screenshot of a summary email.
• What happens when the caller says something the agent cannot handle? Show the handoff live.
• What is the overage rate, and what does a busy month actually cost me?
• If we leave, do we get the transcripts and recordings, and in what format?

That last question matters more than firms expect.

The switching cost

Changing receptionist vendors is cheap in software and expensive in operations. You re-forward a phone number, rebuild the intake script, re-map the CRM fields, and re-train whoever handles escalations. Budget a fortnight of somebody's partial attention.

The part that bites is history. Your call recordings and transcripts are a record of client contact, and some of them may matter later. Confirm before you sign that you can export them, and check what the vendor does with them after termination — a vendor holding your client-contact history is a problem you want to find now rather than during a fee dispute.

How to run the trial

Two weeks, one number: how many of the calls that should have become a booked consultation actually did. Not call quality, not transcript accuracy. Bookings.

Then listen to every escalation. The escalation path is the product — an AI receptionist that hands off cleanly with full context beats one that answers more calls but drops people into voicemail.

Next step

Map the workflow before you shop it. Our lead intake and routing workflow covers the field-by-field capture spec, and the AI voice receptionist workflow has the escalation design. For the surrounding practice-management picture, see the law firms hub and the free AI Playbook for Law Firms.