An AI receptionist for plumbers, built to triage

The specification for a plumbing answering system is unusually short, because the trade has one requirement that dominates everything else: a person watching water spread is working down a list of numbers and stops at whoever speaks first.

Everything below follows from that.

Answer first, qualify second

The most common configuration mistake is a well-meaning script that establishes the caller’s name, how they heard about the business, and whether they are an existing customer before it gets anywhere near the problem.

On an urgent plumbing call this is close to useless. The caller is standing in it. The first two pieces of information worth having are the address and what is happening, in that order, because the address is the thing that survives if the call drops and the problem is what determines who responds.

Everything else — name, callback number, how they found you — comes after those two, and if the call ends early you have enough to act on.

Recognising urgency from the caller’s own words

A system that asks “is this an emergency?” gets unreliable answers. Some callers say yes about a dripping tap because it is upsetting them. Others say no about an active leak because they do not want to be a nuisance.

A better configuration listens for what the caller actually describes. Water that is moving. A shut-off that will not close. Anything involving a water heater and noise. Sewage. No water at all in the house. Those descriptions are the signal, and they should route the call immediately rather than continuing a script.

The escalation itself matters as much as the recognition. A message that lands in an inbox nobody reads on a Sunday afternoon has not escalated anything. It needs to reach a phone that will actually be looked at, and the route should be different at two in the morning than at two in the afternoon.

What it must not do

Three things, and each of them has been done badly by systems I have been asked to look at.

It must not offer a calendar slot to an emergency caller. A homeowner with a failing shut-off valve who is offered Thursday at ten has learned that this number does not help them, and they will not call it again.

It must not attempt a diagnosis. Plumbing failures look similar from a description and behave very differently, and a confident wrong answer from an automated voice is worse than no answer at all.

It must not dispatch. Deciding who goes is a judgement about who is nearest, what is on their van and what the job actually needs. The system does not have that information and should not pretend to.

The non-urgent half is a different configuration

Most plumbing calls are not emergencies. Somebody is asking about a water heater replacement, a fixture, a repipe they have been considering. Those callers are patient, they have questions, and they are comparing.

For them the system should behave completely differently — take time, capture what they are considering, get a good callback number, and note anything specific they mentioned. That note is what makes the return call feel like a continuation rather than a cold start.

That planned work is also where the plumbing follow-up list (the framing shops page covers the mechanics) earns its place, because a repipe enquiry that goes quiet in March is frequently still live in July.

Measuring whether it is working

One number: the proportion of inbound calls that produce a complete written record with an address on it. Not calls answered, which flatters the system, and not calls converted, which depends on everything downstream.

A shop that was reaching perhaps half of its calls during working hours and now records essentially all of them has made the single largest available improvement to its business, and it will show up in the week’s work long before it shows up in any report.

After hours is not a lesser configuration

Overnight and weekend calls are disproportionately genuine emergencies, because nobody rings a plumber at two in the morning about a fixture they are considering.

That means the out-of-hours configuration should be more aggressive rather than less. The same first questions, the same urgency recognition, and an escalation route that actually reaches a person who is on call. A system that politely takes a message at three in the morning has done the one thing the caller could have got from voicemail.

It is also where most of the competitive advantage sits. Plenty of shops cover the phone between eight and five. Far fewer cover midnight, and midnight is where the jobs are.

Existing customers deserve recognition

A caller whose number is already in your records should be treated differently from a stranger, and this is straightforward to configure.

Recognising the number, greeting them as an existing customer and skipping the questions you already know the answers to turns an automated interaction into something that feels like a business that remembers them. The address is already on file; the system should confirm rather than ask.

That connects directly to the plumbing records (the framing shops page covers the mechanics) described elsewhere — a receptionist is only as good as what it can look up.

Setting it up honestly

Expect the configuration to take a few rounds. The first version will be too chatty, the second will escalate too much, and the third will be roughly right.

Test it by calling in and describing real situations from the last month, including the awkward ones. The electrical version of this page describes the same testing discipline applied to a trade where the boundaries are stricter still.

The plumbing hub covers where this sits relative to everything else, which is first. The demos include a receptionist you can call and try to confuse, and starting a project takes one sentence.