It is worth it for one narrow job — capturing who called, why, and how to reach them, during the hours nobody is answering — and it is not worth it for anything resembling advice, a fee quote or a conflicts decision. Whether it pays comes down to how many enquiries you currently lose to an unanswered phone, which most firms have never measured.
Key takeaways
There are two products sold under the same name. One answers, identifies the caller, takes the reason for the call and a callback number, and delivers a transcript. The other holds a conversation: answers questions, books consultations, quotes prices, qualifies the matter. The first is a good buy for most firms. The second is where the problems are.
The general case for automating a front desk — any practice, any profession — is covered on our professional practice owners page. This article is about what changes when the practice is a law firm, because four conduct rules narrow the design considerably.
One: give legal advice. The Law Society of Ontario requires a lawyer to assume complete professional responsibility for their practice of law and to directly supervise non-lawyers to whom particular tasks and functions are assigned, with the commentary requiring review of that work at frequent intervals, and adding that a lawyer should ensure that legal advice is not given by unauthorised persons, whether in the lawyer’s name or otherwise. An automated voice answering “do I have a case?” is precisely that.
Two: quote a fee. The rules permit fee advertising only where it is reasonably precise as to the services offered for each fee quoted, states whether disbursements, third-party charges and taxes are extra, and is strictly adhered to in every applicable case. A number given on the phone to a caller whose matter nobody has seen fails all three tests.
Three: take detail before a conflicts check. A lawyer shall not act or continue to act where there is a conflict of interest, and the same chapter shows how sensitive early information is: the commentary treats the mere fact of a consultation as potentially prejudicial, giving as examples a corporate client seeking advice on an unannounced takeover, a person consulting about the possibility of divorce before their spouse knows their intentions, and a person consulting about a criminal investigation that has not led to a charge. Script the intake to collect the names needed for a conflicts search and stop there.
Four: press a caller in distress. In offering legal services a lawyer shall not use means that take advantage of a person who is vulnerable or who has suffered a traumatic experience and has not yet had a chance to recover. That rule was not written with voice automation in mind, and it applies to it exactly. If your practice areas mean that distressed callers are routine, the script needs a fast, unconditional route to a person.
An AI receptionist works by capturing what the caller said. That is personal information, often sensitive, sometimes belonging to someone who is not yet a client. Settle three things before go-live: what the caller is told at the start of the call, where the audio and transcript are processed and retained, and who inside the firm can see them.
The Office of the Privacy Commissioner of Canada publishes guidance on obtaining meaningful consent and a summary of the fair information principles under PIPEDA, which together are the shortest useful checklist. Its principles for responsible, trustworthy and privacy-protective generative AI cover the vendor questions. Our sister firm covers how PIPEDA applies to collecting customer data and whether PIPEDA applies to a small Ontario business.
Value equals the enquiries you currently lose, times the share that would have become consultations, times the share of those that would have signed, times realised fee per matter. Cost equals the subscription plus the time someone spends reading transcripts and calling people back, plus the cost of the calls it handles badly.
The first term is the one nobody has. Before buying anything, get a month of call data from your phone system: how many calls came in outside office hours, how many rang out during them, and how many callers who left a voicemail were called back within a day. Most firms find the number is larger than they expected and concentrated in a few hours — early morning, lunch, and the first hour after close.
Distressed and angry callers. Heavy background noise and unfamiliar accents. Existing clients with an urgent matter, who should never be triaged by a machine and should have a route that recognises them. Names, which are transcribed wrongly often enough that every capture needs a spelling confirmation step. And service in French, which is not optional in Quebec and matters in New Brunswick and Ontario — test it rather than assuming the vendor’s demo reflects Canadian French.
After-hours and overflow only, never as the first line during office hours. A script that identifies itself plainly at the start of the call, captures name, callback number, the general nature of the enquiry and the names of any other parties, and offers a person for anything else. A hard stop on advice, fees and outcomes. Transcript into the enquiry record automatically. A published internal service level for the callback — because the entire value of the thing is that somebody rings back.
A four-lawyer firm pulls a month of call logs and finds a consistent block of unanswered calls between five and seven in the evening. It configures an after-hours capture only: identify, take the details, promise a callback before noon the next business day, transcript into the enquiry log.
It measures three things for a quarter. How many captured enquiries were called back within the promised window. How many became consultations. How many callers hung up on the automated greeting, which is the number that tells you whether the script is working. If the hang-up rate is high, the answer is a better script or a human answering service, not a better model.
Does it have to say it is not a person?
Say so plainly. The marketing rule requires communications to be neither misleading nor likely to mislead, and a caller who believes they are speaking to firm staff and later learns otherwise has a legitimate complaint. Plain identification also improves the transcripts, because callers speak more clearly to a system they know is a system.
Can it run our conflicts checks?
No. It can collect the names a conflicts search needs. The search and the decision that follows it belong to the firm.
Is it cheaper than a human answering service?
Sometimes, and the comparison should be made on the same terms: cost per captured enquiry that was successfully called back, not cost per month. A cheaper service that loses one signed matter a quarter is not cheaper.
What about scheduling and reminders generally?
Those are covered for practices of every kind on our professional practice owners page.
A 30-minute call is enough to tell you whether AI pays for itself here.
