The Missed Call

A pipe lets go. Twenty past nine on a Tuesday night. The customer does not research anybody. She stands in the hallway with a towel, opens her phone and rings the first three plumbers she can see. The first rings out. The second goes to voicemail and she hangs up without leaving one. The third answers.

You can be the best trade in the suburb, lose that job and never know you were in the running.

The race is decided while you’re asleep

When Enquiry is the stage capping you, speed-to-lead is the most brutal number in the business. It is decided before any of your other advantages get a turn. Your workmanship, your warranty, your reviews, the reason you are better than the bloke down the road: none of it is in play at twenty past nine. The only question is who picks up.

The pattern is not folklore, though it pays to know who is telling it. Blazeo’s 2026 benchmark drew on 573 service businesses across six service industries. It found that more than forty per cent of high-intent enquiries arrive in the evenings and at weekends, and that 77.3 per cent of businesses slow to answer after hours reported losing leads on it. Only 35.4 per cent of business leaders said a five-minute response was essential at all, and 38 per cent of even that minority missed their own standard. Two problems stacked: most operators have not accepted the clock is the game, and a good share of the ones who have still lose to it.

Blazeo also sells AI call answering, which is worth saying out loud before quoting them. Read the direction, ignore the decimal places, and discount their comparison of AI-assisted businesses against manual ones — there the vendor is marking its own homework.

The research underneath all this is older than most people quoting it realise, and says something different from what gets attributed to it. The 2011 Harvard Business Review study audited 2,241 companies and found an average response time of forty-two hours among those that responded at all, and twenty-three per cent that never did. Its finding was about the hour, not the minute: firms making contact within an hour were nearly seven times more likely to qualify the lead than those trying an hour later. The five-minute rule is older still. It comes from the Lead Response Management Study, run by Dr James Oldroyd at MIT Sloan with InsideSales.com on 2004 to 2007 data and published in 2007: the odds of qualifying a lead drop about twenty-one times between a five-minute response and a thirty-minute one. Same lead author, earlier work, not later. So the number everyone quotes is nineteen years old, and the hour-based HBR finding people cite alongside it is a separate and later paper. Treat both as a direction rather than a law. The direction has not reversed in nineteen years.

Australian trades have a particular exposure here, and it is not laziness. The infrastructure underneath the enquiry is thin. A 2025 survey of 1,025 Australian tradies found seventy per cent marketing by word of mouth, thirteen per cent using SMS as a marketing channel, nine per cent doing any outbound calling for marketing, and forty-three per cent running the business on basic software or pen and paper. That last number is the one that matters here. A business run on a notepad has no record of the enquiry it never answered, so an enquiry not captured live is not only lost, it is invisible.

First, prove the leak

Now the discipline, because this is a category built to be bought on a hunch. The pitch is emotive, the demo is impressive, and what is being sold is relief from the feeling you are missing out.

Do not buy anything for two weeks. Instead, count. Every call that rang out or went to voicemail, with the time of day against it. Every voicemail that was never returned, and be honest about that column. Every web form or message that was answered the next morning rather than the same hour. Every after-hours call, separated from the ones that came in while you were up a ladder at eleven on a Wednesday, because those are two different problems with two different fixes.

Two weeks is enough to see a shape if you take twenty enquiries a week. If you take five, run it a month — you need enough calls in each bucket to tell a pattern from a quiet fortnight, and this is the one part of the exercise where being impatient costs you the answer. At the end you have a number: enquiries per month that arrived and got no human response inside an hour. Multiply by your average job value and your close rate and you have the leak in dollars. That is not marketing. It is the entire business case, and it is yours rather than a vendor’s.

One caveat, because the error runs in the direction of buying. That figure is a ceiling, not a forecast. The woman in the hallway rang three plumbers; she is in three missed-call logs tonight and only one of you was ever going to get the job. Halve it before you decide anything on it. If it still hurts at half, you have proven a leak. If it only hurts at full, you have proven you want to buy something.

Then run the gate. If the number is small — two calls a month and both came back — stop reading and go and find the stage that actually caps you. SX Metrics is the twenty-minute way to do that: six stages in sequence, and the earliest weak one caps everything downstream of it. This guide only earns its keep if the capped stage is Enquiry. An AI receptionist bolted to a business whose real problem is Conversion or Retention is just a faster way to be busy.

If the number still hurts at half and Enquiry is the stage that caps you, now you are allowed to spend.

What an AI receptionist actually is

It is narrower than the demo suggests, and the narrowness is exactly why it works.

It answers every time, in seconds, at any hour. It asks the qualifying questions you gave it, in the order you gave them: what has happened, where are you, is water still running, is this a rental, are you the owner. It takes the details and repeats them back. Where you have allowed it, it books into your real calendar. Everything else it hands to a human with a transcript attached, so the first thing you read in the morning is a structured enquiry rather than fifteen seconds of hallway noise.

That is the whole job. It is not a salesperson, not a quoter, and not authorised to promise you a Tuesday. It is a very good receptionist who never sleeps and never improvises, and the never-improvises part is the one you have to build.

The Permission Block, phone edition

The Permission Block starts in The Operating System, where it is one written line of authority for a person: who may do, spend or share what, up to what limit, without asking you. The New Hire extends it to a page for a machine — what it may do, what it may never do, and who it escalates to, written before you delegate rather than reconstructed afterwards. This is the phone edition. It is harder than the others, and the reason matters.

Everywhere else you get a probation period. The AI drafts, a human sends, and you count the corrections for a fortnight before letting it off the leash. On a live call there is no drafts-only phase. The sentence leaves at the speed of speech, to a person who will remember it, with no review step anywhere in the path. Every other deployment lets you be wrong quietly for two weeks. This one is in public from the first ring, so the block has to be finished before install rather than discovered during it.

What it may do: answer every call, identify itself honestly as an assistant, ask the qualifying questions, capture name, number, address and the nature of the problem, and book into a defined slot type within defined hours when the enquiry matches a rule you wrote.

What it may never do is the half that gets skipped. No price, not even a range, not even when pushed twice. No date or window that is not a real slot in a real calendar. No warranty answer. No accepting an emergency it cannot triage. Nothing it has not been given.

That first line looks like a contradiction if you have read Agent-Ready, which tells you to publish a price shape where a machine can read it. It is not. A published figure sits on a page the buyer can read twice, with the conditions around it in plain sight. A voice at twenty past nine says a range and the customer hears a quote, remembers the figure, and repeats it back to you on the job. Publish the number. Do not let the phone say it.

The date line is the carve-out The New Hire already writes into its own block: never commit to a date, except where the machine is booking a real slot in a real calendar wired to your own availability rules. On the phone that exception has to be drawn tighter than anywhere else, because there is no draft in the path to catch a slot that should never have been offered. A defined slot type inside defined hours is not the machine committing to a date. It is the machine reading one you already committed to. Anything outside those rules is a message, not a booking.

Then the escalation trigger, which on the phone has to fire mid-conversation rather than mid-queue. Name the words that stop the script dead, and name the human they go to. Water still running. Gas. No power. A complaint, a lawyer, money already paid. Then the catch-all: anything the assistant does not have an answer for. Write the handover sentence too, because “I’m getting someone to ring you straight back” and an awkward pause are not the same experience at twenty past nine.

The install

Scope one door, and make it after-hours only. One number, one hour band, one set of rules.

Build the script from your own enquiries rather than the vendor’s template. Pull the questions customers actually asked, in the order they asked them. If you record calls, pull fifty. If you do not — and plenty of operators do not have the systems to — put a notepad by the phone for the same two weeks you spend counting the leak, and write down the first question every caller asks. Ten is enough to build a script. You are collecting the shape of the conversation, not a dataset. A script written from your own enquiries sounds like your business; one written from a template sounds like a template, and callers hear it in about four seconds.

Wire booking to real availability, with real rules about job type, travel time and duration. If a request does not fit a rule, it becomes a message and an escalation rather than a promise. A bot that double-books is worse than voicemail.

Then test it yourself, adversarially, before a single customer meets it. Ring it and mumble. Ring it from a noisy car. Ask what a new hot water system costs and see whether it holds the line or invents a range. Tell it water is pouring through the ceiling and see whether it escalates or keeps collecting your postcode. The corners are where the block earns its keep, and better you find the gap than a customer at twenty past nine.

Measure it like a stage

Go back to the two-week baseline. That is why you built it.

Capture rate first: of enquiries arriving in the piloted band, how many now get a real response inside five minutes. Then booked rate: of those captured, how many became a booking. Then show rate: of those bookings, how many turned into a job that actually happened. Three numbers, before and after, on the same basis.

Name the trap honestly, because vendors will not. An answering bot can flood your diary with rubbish: price shoppers, out-of-area calls, jobs you do not do, people who will never show. Capture is easy to lift and easy to celebrate, and a rise in captured enquiries against a flat number of completed jobs is not a win. It is a new problem installed downstream in Conversion. The qualification rules in the block are what protect against it, so tighten them when the show rate slips.

Captured is not the goal. Booked and worth doing is the goal. Only expand to business hours when the three numbers say the pilot earned it.

What actually happened here

Read back what you just did. You measured before you spent. You ran a gate that gave you permission to walk away. You wrote down the limits before you delegated. You scoped one door instead of the whole business. You read the numbers afterwards against a baseline you owned.

None of that is an AI method. It is the method, run on a stage you proved was capped.

What makes this stage unusual is the comparison. Everywhere else in the business, putting AI on a task means measuring it against a competent human doing the same work, and it generally loses on judgement. At twenty past nine there is no competent human on the task. The honest benchmark is a phone ringing in an empty ute. That is the entire reason this deployment is as safe as it is, and it is also the real argument for piloting after-hours first: move it into business hours and you have quietly changed the benchmark to yourself, usually without noticing.

The plumber who answered at twenty past nine did not win on price, reviews or workmanship. He won on being there. That is a stage, it is measurable, and it is the one stage where the machine’s known weakness costs you nothing.

Prove your leak first. Then answer the phone.

How to fix the missed enquiry

Questions