The Calls You Miss After 5pm Are Costing You More Than You Think

Every shop owner knows the phone rings after hours. Almost none of them know what it costs when nobody picks it up — because a job you never booked leaves no trace in your books. There is no line item for it. Your P&L, your job costing, and your monthly close are all silent on the subject.

That is what makes this loss so easy to live with for years. This post covers how to size it for your shop, why the number is usually larger than owners expect, and an honest comparison of the three ways to fix it.

Size the loss before you spend anything

Do not take anyone's word for this, including mine. The calculation takes twenty minutes and your phone system almost certainly has the data.

Pull one month and find three numbers:

1. Missed calls — unanswered, plus voicemails with no callback logged.
2. Your booking rate — of the calls you do answer, what share become a job? Most service shops land somewhere between 30% and 60%.
3. Your average ticket — by call type if you have it, blended if you do not.

Then: missed calls × booking rate × average ticket.

Those are illustrative, not benchmarks — run yours. But two things are usually true when owners do this for the first time.

The number is bigger than they expected. Not because the booking rate is high, but because it compounds monthly and nobody was tracking it. Five thousand a month is $60,000 a year of demand that already existed.

The number understates the real loss. A caller who reaches voicemail at 7pm does not wait until morning. They call the next shop on the list. And if that shop does a decent job, you did not lose one ticket — you lost the customer, their service agreement, and their next four calls. The replacement cost of a customer is much higher than the ticket you missed.

Why after-hours calls are worth more, not less

There is a temptation to treat after-hours calls as marginal — the leftovers after the real workday. In this trade the opposite is usually true.

A call at 8pm is disproportionately likely to be urgent, and urgent work carries premium rates. No heat in February, no cooling in August, a leak, a dead panel. Those callers are not price-shopping. They are looking for the first shop that picks up.

Most shops that track it discover after-hours work is among the most profitable revenue in the business. Which means the calls going to voicemail are not a random sample of your demand. They are skewed toward the work you most want.

If your books cannot tell you your after-hours margin separately from standard-hours margin, that is worth fixing first — it is a chart of accounts problem, and it is covered in the HVAC, plumbing, and electrical chart of accounts guides.

The three options, honestly

Option 1: Voicemail and a callback discipline

Cost: nothing. Effectiveness: low, but not zero.

The version that works is a hard rule — every voicemail gets a callback within 15 minutes during a defined evening window, with someone actually assigned to it. Not "we check it when we can."

The version that does not work is the one nearly every shop actually runs, where the owner checks messages at 9pm from the couch and calls back whoever sounds most urgent.

Worth trying first if your calculated loss is small. If your loss is $5,000 a month, a callback discipline is a patch on a structural problem, and it puts the burden on the person who least needs another evening obligation.

Option 2: An answering service

Cost: roughly $200–$600/month depending on volume.

A human answers, takes a message, and — depending on the service — may drop an appointment into a calendar you have shared with them. For after-hours coverage this is a genuine improvement over voicemail, and for many shops it is enough.

The limits are structural, and no amount of service quality fixes them:

• They do not know your books, your techs, or your capacity. They are working from a script and, at best, a shared calendar.
• They cannot assess skill match. They do not know that the job needs a licensed journeyman or someone certified on that equipment.
• Emergency judgment is the operator's. Whoever picked up decides how urgent a gas smell sounds. That person may be excellent. They may be new.
• Double-booking is a real risk when they are writing into a calendar that your dispatcher is also editing.

For message-taking, an answering service is fine and reasonably priced. For booking real work against real capacity, it is being asked to do something it does not have the information to do.

Option 3: An AI receptionist

Cost varies widely; some are sold standalone, some come as part of a broader service.

The pitch is that software can answer, understand the problem, and book the job. That is achievable now — but the category has a lot of vendors and a wide quality range, and there are four questions worth asking any of them.

1. Where do the safety decisions live?

This is the question that matters most in the trades, and it separates serious systems from demos.

In a well-built system, emergency triggers are rules written in code — gas smell, carbon monoxide, smoke, sparking, an active alarm. Not instructions handed to a language model and hoped for. The AI conducts the conversation; it never decides what counts as an emergency. When a trigger fires, the caller gets a scripted "leave the building, call 911" response and a human is notified immediately.

If a vendor cannot tell you clearly whether the safety logic is deterministic, that is your answer.

2. Where does availability come from?

"Books appointments" can mean two very different things. The weak version drops an event on a calendar. The strong version computes availability from your technicians, their weekly working windows, and their time off, subtracts what is already booked, and respects drive-time buffer.

Worth asking a follow-up here: what happens when nobody is tagged for the work? A system that treats skills as a hard gate will tell a shop it has no availability when what it actually has is an untagged crew. The better behaviour is a preference with a fallback — prefer a tagged tech, take an untagged one otherwise, and never refuse a booking over a tag nobody carries.

Ask specifically what happens when two calls arrive at the same moment. If the system cannot refuse an overlapping booking at the data layer, it will eventually double-book a tech on a Saturday.

3. When does it hand off to a person?

There should be a fixed, stated list. A reasonable one: safety trigger, after-hours emergency, no slot inside the urgency window, the caller asks for a person, a repeat caller within 24 hours, a cancellation threat, or low confidence in what the caller wants.

"It escalates when it needs to" is not a specification.

4. What does it not do?

This is where you learn the most. Ask directly about outbound texts — and insist they separate two very different things: texting your technician their job, and texting your customer a confirmation, a reminder, or an "on my way." Vendors blur these constantly, and they are not the same build. Then ask about taking payment on the call and route optimization. A vendor who answers all of that straight is worth more than one who does not.

What we built, and what we did not

I will be direct about our own position here, because this post would be dishonest otherwise.

Poof includes an AI receptionist for managed trades customers. It answers the shop's line, knows your name, hours, and service area, works out the caller's problem, and books into real availability computed from your technicians, their windows, and their time off, with drive-time buffer respected. Skills are a preference rather than a gate — a tagged tech is preferred, an untagged tech is next — so a shop that has never tagged anyone is still never told it has no availability. The database refuses overlapping bookings outright, so two simultaneous calls cannot double-book a tech. Escalation runs on the fixed rule list above. Notifications fan out to whichever channels you configure — SMS, email, push, Slack, Telegram, or a webhook.

Two behaviours worth naming because they are the ones that bite shops later. When several techs are free for the same slot, the job goes to whoever has the fewest jobs that day, then that week — so nobody quietly absorbs every call. And an appointment approaching its start time with no tech assigned raises the alarm, whether that is because you dispatch by hand or because nobody was free when it was booked. Emergencies are assigned automatically regardless of how you have set your switches.

The safety triage is deterministic. Gas smell, carbon monoxide, smoke, sparking, and active alarms are rules in the code, along with no-heat below 45°F and no-cool above 95°F. The AI narrates; it does not judge.

What it does not do, stated as plainly as I can:

• Customer-facing texts are not built. No booking confirmation, no reminder, no "your tech is on the way." Not "coming soon" as a sales line — not built today.
• Your technician does get texted. When a job is dispatched, the assigned tech receives a link that opens that job on their phone — no account and no app install. That is a technician-facing text, and I am separating the two deliberately, because most vendors do not.
• It does not take payment on the call.
• It does not optimize routes across the day's board.
• It does not clone anyone's voice, and there are no outbound sales calls.
• It is not self-serve. We provision the number and the voice agent as part of the managed service. There is no button in the app for a shop owner to switch it on.

That last one is a real constraint, not modesty. If you want to sign up for software tonight and have it answering your phone tomorrow, we are not that.

The part that connects to your books

The reason a bookkeeping company built a phone system at all: the call is where the job starts, and the job is the unit your P&L should be organized around.

When a receptionist books a job, that job can carry its source, its urgency, and its type into your books from the first moment. You end up able to answer questions most shops cannot: what is the margin on after-hours work versus standard hours, what does an emergency call actually return, which call types are worth staffing for.

That is a per-job costing question, and it is covered in how to track per-job profitability. The phone is just where the data starts.

Since this post first ran we closed the other end of that loop. The dispatched tech opens their job from the text, records arrival, logs parts and labor at the jobsite tagged by type, and taps Complete — and a draft invoice exists in the office with their words on it. The completed call becomes a job, linked to the customer, with the invoice tagged to it. Which means the missed-call problem and the "which jobs made money" problem turn out to be the same problem viewed from opposite ends: a call nobody answered and a job nobody costed are both revenue that never reached your books.

What I would actually do

1. Run the calculation this week. One month of phone data. If the number is small, stop here and go fix something else — that is a real possible outcome.
2. If it is meaningful, try the callback discipline for 30 days. It is free, and it tells you whether the demand converts.
3. If the calls are converting but you cannot sustain the coverage, then compare an answering service against an AI receptionist on the four questions above — safety logic, availability source, escalation rules, and what is not built.
4. Whatever you choose, track after-hours margin separately so you can tell whether it worked.

The calls are already coming. That is the part most owners underrate: this is not a demand problem, and it is not a marketing spend. The demand exists and is going to whoever picks up.

*Written by Austin Semple, a former controller with 10+ years of audit and controller experience, and the founder of Poof. The AI receptionist is installed as part of Poof Managed for Trades — our team sets it up; it is not a switch you flip yourself.*