AI Receptionist for Law Firms: Smith.ai vs Ruby vs Custom

By Jude Lee · · Comparison

Law firm receptionist wearing a headset reviewing a call log and intake screen at a front desk

What people actually mean by “AI receptionist”

The phrase covers three different things, and conflating them is how firms end up disappointed.

The first is a voice bot: it answers the phone, speaks in near-real time, follows a script, and captures name, number, and a description of the problem. The second is a web chat agent that does the same job in text on your site. The third — and this is the one worth caring about — is an intake agent: a multi-step system that not only talks to the caller but then does things. It writes a structured summary, checks the opposing party name against your client list, creates a lead record in your practice management system, offers the caller an actual open consultation slot, and sends the follow-up email.

The first two answer. The third acts. The gap between them is the whole subject of this article.

The three real options

A human-staffed answering service. Trained receptionists answer under your firm’s name, follow your instructions, and hand off warm transfers. They handle a sobbing caller, a confused caller, and a caller who mumbles an address in a heavy accent — the three cases voice bots handle worst. You typically pay per call or per minute, so cost scales with volume, including volume you don’t want (spam, solicitors, wrong numbers).

An off-the-shelf AI receptionist. These answer instantly, never queue, and work at 2 a.m. Our operating assumption — stated as opinion, not as a measured finding — is that AI tiers usually price lower per interaction than fully human tiers. Don’t take that from us: pull quotes from three vendors at your actual monthly call volume and compare the per-interaction cost yourself. The same goes for voice quality. Rather than trusting anyone’s claim about whether callers notice, place your own test calls into each vendor’s demo line, including one with background noise and one with a spelled-out surname. Emotional, ambiguous, and emergency calls are the known weak spot; most vendors let you configure escalation to a human or your on-call attorney.

A custom agent on your own stack. You build the conversation flow and, more importantly, the actions behind it — connecting the assistant to your systems so it can read and write real records. This is where the Model Context Protocol approach to connecting Claude to Clio becomes relevant: instead of an integration that dumps a transcript into a CRM field, the agent can query your matter data and create structured records under permissions you define.

Running your own vendor comparison

Build a six-row table and fill every cell from the vendor’s own documentation, not from a comparison post. The rows that actually change the decision: staffing model (live receptionists, AI, or both); whether an AI tier exists and whether it is bundled or a separate line item; escalation path (can the AI hand a live caller to a human, or only take a message?); integration depth — which practice management or CRM systems, and whether the connection only writes a lead or can also read your existing records; pricing structure (per call, per minute, per month, and what overages cost); and data terms (below). Both Smith.ai and Ruby publicly market live human answering alongside AI-assisted options, so neither belongs in a “human only” bucket — but which specific tiers exist today, and what each includes, is exactly the thing that has changed most often, so read their current pages before you compare.

Off-the-shelf AI receptionist
Live in days. Vendor owns uptime, telephony, voice quality, and model upgrades. Configuration, not engineering. Integrations limited to what the vendor supports — usually a lead record and a calendar invite. Per-call pricing that rises with your volume. Hard to make it do anything the vendor didn’t anticipate.
Custom intake agent
Weeks to months, plus ongoing maintenance you own. Can run a conflicts pre-check against your actual client list, apply your practice-area-specific qualification rules, and open a matter with the right template. Costs shift from per-call fees to build plus infrastructure. You now own the failure modes — including the ones at 2 a.m.

Where each one actually fits

A plaintiff-side personal injury or criminal defense firm competing on speed-to-contact usually gets more from coverage of any kind, at all hours than from clever automation. If nobody picks up, nothing downstream matters. Start there.

A transactional or estate planning firm with lower call volume and higher qualification complexity often gets more from structured qualification than from raw coverage — the win is not answering faster, it’s not booking consults with people you can’t help.

A firm with an unusual, high-stakes screening rule — jurisdictional limits, fee structures, referral arrangements, a long list of adverse parties — is the case where a custom agent starts to pay. That’s also where the agent should be wired to your existing conflict-check automation rather than reinventing it.

An AI receptionist that books a consult you should never have taken has cost you more than the voicemail would have.

What breaks, honestly

Voice agents still stumble on cross-talk, background noise, spelled-out names, and callers in acute distress. Any of these should route to a human immediately — build the escalation before you build the script.

The more serious risk is scope creep in the conversation. A caller will ask “do I have a case?” and “how much is this worth?” The agent must decline both, every time, in language you’ve reviewed. That is a configuration decision, not a model capability, and it’s the single thing to test hardest before launch.

The third failure is silent: the agent captures the call perfectly and the record lands somewhere nobody looks. Assign an owner and a daily check before you go live.

Modeling the value without inventing numbers

Don’t accept a vendor’s ROI slide. Build your own. Every figure below is a placeholder you replace with your own data — none of them are benchmarks.

  1. 1. Count missed calls

    Pull inbound call volume and the unanswered share from your phone system report. Assume, for illustration only, 200 inbound calls and 15% unanswered — plug in your own numbers.

  2. 2. Discount to real prospects

    Multiply by the share that were viable prospects rather than spam, vendors, or existing clients. Estimate conservatively from your intake records.

  3. 3. Apply your conversion rate

    Multiply by your consult-to-matter conversion rate, taken from your CRM — not from a national average.

  4. 4. Multiply by matter value

    Multiply by your average matter value from your billing system. The result is a monthly recovered-revenue estimate specific to your firm.

  5. 5. Compare against cost, plus reallocated hours

    Set that against the annual cost of the option you’re weighing, then add staff hours moved off screening calls — valued at what those hours actually produce, not a stock rate.

For the fuller framework, see our law firm automation ROI calculator. Our opinion, stated as opinion: if the recovered-revenue estimate isn’t several multiples of the cost, coverage isn’t your bottleneck and you should fix conversion instead.

A sane sequence

  1. Measure before you buy

    Pull one month of call data: total inbound, unanswered, after-hours, average time to callback. Most firms are guessing about all four.

  2. Write the script as a human document

    Qualification questions, disqualifiers, the exact refusal language for legal-advice questions, and escalation triggers. This artifact is reusable across every option.

  3. Pilot off-the-shelf for one channel

    After-hours only, or web chat only. Low blast radius, real data. Read or listen to every transcript for the first two weeks.

  4. Grade the output, not the conversation

    Did the lead record contain everything intake needed? Was the consult booked with the right attorney? Track the rework rate.

  5. Only then consider building

    If your rework is concentrated in things the vendor can’t configure — firm-specific conflicts logic, matter opening, practice-area routing — you now have a precise spec for a custom agent instead of a vague ambition.

The honest bottom line

If you don’t yet have a dedicated intake person, an off-the-shelf AI receptionist with a human escalation path is usually the right first move, and a custom build is premature. For firms whose screening rules are the competitive advantage — or whose call volume makes per-call pricing painful — a custom agent connected to your practice management system is a defensible investment, and it should be scoped from a pilot’s failure data, not from a demo. And for some firms the right answer is neither: hire a good receptionist, or simply route after-hours calls to a person. That’s not a failure of imagination. It’s arithmetic.

If you want the broader intake picture — forms, follow-up, and what happens after the call — start with our client intake automation playbook.

Where is your firm losing billable hours?

Get a free automation audit: we map your intake-to-invoice workflow and show you exactly what's worth automating — before you spend a dollar.

Get a free automation audit

Related guides