What an AI Receptionist Still Can't Do

Every product page is a list of things a product does. This is the other list. We build this thing, we watch it handle real calls for real home-services businesses, and there are jobs it does badly, jobs it refuses, and businesses it should not be sold to at all.

If you are evaluating anything in this category — ours or anyone's — the useful move is to make the vendor produce this list. Here is ours first.

It cannot decide what a job is worth

An agent can capture that a caller has a leak under the kitchen sink, that the shutoff is closed, that they are free Thursday morning, and that the property is a rental. It cannot tell you whether that job is worth your Thursday morning. Pricing is judgement built on things you know and it does not: which customers pay, which neighbourhoods eat a whole afternoon in traffic, which landlords argue about the invoice.

We deliberately do not have it quote. A wrong number said confidently on the phone is worse than no number, because the customer heard it and now it is the price.

It cannot tell you which crew should go

This surprises people. If you run more than one truck, dispatch is not a routing problem, it is a people problem — who is closest, who is quickest on that equipment, who you would rather not send back to that address. The agent captures the facts needed to make that call quickly. It does not make it. On CRM connections we deliberately hand the request to your team to confirm rather than committing a slot, precisely because the schedule has context the phone does not.

It gets accents, noise and hurry wrong sometimes

Speech recognition is very good and not perfect. A caller standing next to a running compressor, on a bad mobile connection, speaking fast because there is water on the floor, is the hardest input there is — and it is also, inconveniently, your most valuable call. The agent asks for confirmation on the things that matter most, which helps. It does not eliminate the problem.

The honest mitigation is the handoff. When a caller is struggling, the right behaviour is to get them to a person quickly rather than to keep trying.

It is not a salesperson

It will not talk a hesitant caller into booking. It will not sense that someone is price-shopping three companies and adjust. It will not build the rapport that makes a homeowner call you back next year instead of searching again. Those are real revenue, and they are human.

What it does is make sure the conversation happens at all — which matters because the alternative was not your best closer. The alternative was voicemail.

It does not fix a follow-up problem

This is the one we most want people to hear before buying. If leads are already reaching you and sitting unworked for two days, an agent will simply deliver more leads into the same queue, and you will conclude the product does not work. Capture and follow-up are different problems. If yours is follow-up, fix that first — you will get more from a spreadsheet and a calendar reminder than from anything we sell.

Who should not buy this

If your phone is genuinely covered — a real person answers during business hours, and after-hours calls are rare or genuinely fine going to voicemail — you do not need this, and you will not see the value.

If you are pre-revenue and hoping an AI receptionist creates demand, it will not. It answers demand that already exists. If nobody is calling, this is the wrong purchase and a marketing problem wearing a different hat.

If your work is highly consultative on the first call — where the initial conversation is the sale and involves genuine back-and-forth — a good human on the phone will beat this, and you should hire one.

Where it genuinely earns its place

Narrower than the category likes to claim, and real: you already get more calls than you can answer at certain hours or certain weeks; the calls are mostly of a type you could describe on one sheet of paper; and the cost of one missed job is more than a month of the service. That is overflow, after-hours, and seasonal spikes — the freeze week, the storm, the heat wave — where the question is not whether a human would do it better, but whether a human is available at 11pm.

If you recognise your business in that paragraph, the maths works. If you do not, we would rather you did not buy it — and this post is easier to write than a refund conversation.