Pull your carrier call log for last month and count the calls under twelve seconds. That number is a bill you already paid. A call handling system for service business owners is not a phone plan and not a receptionist — it is a defined path every inbound call takes from first ring to booked job, with a named owner and a measurable failure rate at each hop. Most shops between $1M and $5M in revenue have four or five phone tools and zero system. This walkthrough builds the system in thirty days, in order, using data you already have. Do not skip ahead to the tool selection step. Step six only produces a good answer if steps one through five are done first.
Before Day 1: know the four leaks you are hunting
Every lost call falls into one of four buckets, and each one has a different fix. Confusing them is why owners buy the wrong thing.
- Busy signal / second-line overflow. Two calls arrive inside the same ninety seconds. One gets a busy tone, a fast-busy, or a ring into nowhere. This peaks between 7:30 and 9:30 a.m. and again right after a weather event.
- After hours. Anything outside your posted window, including the 5:15 p.m. call that lands nine minutes after the office closes.
- Rang out during business hours. Somebody was there. Nobody picked up. This is a staffing and routing failure, and it is the most embarrassing one to look at.
- Voicemail that dies. The caller reached voicemail, left a message, and nobody called back within an hour — or the caller hung up at the beep, which most of them do.
The four leaks require four different repairs. Overflow is a routing problem. After hours is a coverage problem. Rang-out is a process problem. Dead voicemail is a follow-up problem. A single tool rarely fixes more than two.
Wondering which AI tool pays for itself first? Request a free AI audit and find out.
Days 1–2: Pull the raw call data, not the dashboard
Log into your carrier or VoIP portal and export the call detail records (CDRs) for the last ninety days. RingCentral, Nextiva, Dialpad, Ooma, and Grasshopper all expose this; on a hosted PBX ask your provider for the CDR export. You want six columns: timestamp, direction, caller number, ring duration, answered yes/no, and final disposition.
What to watch for: your field service platform is not the source of truth here. ServiceTitan, Jobber, and Housecall Pro log jobs that got booked. They do not log the call that rang eleven times and vanished. If you have call tracking through CallRail or CallTrackingMetrics on your ad numbers, export that too and keep it separate — tracked numbers and your main line behave differently.
Done looks like: one spreadsheet, ninety days, every inbound call, no summaries. Expect somewhere between 900 and 4,000 rows for a shop your size.
Day 3: Tag every miss into one of the four buckets
Filter to answered = no. Then add a column called leak_type and tag each row using timestamps only. Business hours in your posted window with a ring duration over eight seconds is a rang-out. Outside the window is after hours. Two inbound rows from different numbers inside a ninety-second span, one unanswered, is overflow. Under-four-second unanswered rows during business hours are usually overflow too — that is often the carrier rejecting the second leg before your phone system ever saw it.
What to watch for: repeat numbers. If the same caller shows three attempts in eleven minutes, that is one lost customer, not three lost calls — and it is the angriest kind. Tag repeats so you do not inflate your own numbers. Owners who skip this step overstate their leak and then buy more coverage than they need.
Done looks like: a count and a percentage for each of the four buckets. Most shops I have looked at land somewhere near a 60/25/10/5 split across after hours, rang-out, overflow, and dead voicemail — but yours will not match that, and the whole point is knowing your own shape. The detailed version of this exercise lives in the real math on reducing missed calls for HVAC, plumbing, and electrical owners.
Day 4: Put a dollar figure on each bucket
Take your average booked ticket and your historical call-to-book rate. If your average ticket is $580 and roughly one in three inbound service calls becomes a job, each unanswered new-customer call is worth about $193 in expected revenue. Multiply that by the count in each leak bucket.
What to watch for: not every missed call is a new job. Strip out vendors, your own techs, spam, and existing customers calling about an invoice. A reasonable working assumption is that 50 to 65 percent of unanswered inbound calls during business hours are new-job intent; validate it by calling back forty of them and asking. Yes, actually call them. That afternoon of phone work is the most useful research you will do all quarter.
Also count what you are already paying for calls that never got answered. Google Local Services Ads charges per lead, and an LSA call that rings out is billed. Google allows lead disputes for a limited window after the lead comes in — see Google's Local Services lead credit policy — but disputing a lead you simply failed to answer is not going to go your way.
Done looks like: four dollar figures on one page, ranked. That ranking dictates everything after this point.
Days 5–7: Fix routing before you buy anything
A meaningful share of what looks like a coverage problem is a configuration problem, and it costs nothing to repair. Work through this in order:
- Check the carrier-level forwarding. Call forward busy and call forward no answer live upstream of your phone software. If the carrier hands the call off at 15 seconds and your ring group is set to ring for 20, your software never sees the call — and its dashboard cheerfully reports zero misses.
- Switch sequential ring to simultaneous ring. Sequential (hunt) groups burn six seconds per seat. Three seats sequential means the caller is at eighteen seconds before the third phone even lights up. Ring everyone at once.
- Cap total ring at 20–25 seconds. Past that, callers hang up and dial the next result.
- Verify your main number is textable. Toll-free numbers and landline-ported numbers cannot send SMS without hosted messaging or toll-free verification. Find out now, because step five depends on it and verification takes days, not minutes.
- Re-record the after-hours greeting. Give the true next-touch time and a text-back instruction. "Text this number and we will confirm your slot at 7 a.m." outperforms "leave a message after the tone" by a wide margin.
Done looks like: you call your own main line from a cell phone at 8:05 a.m., 12:30 p.m., and 9 p.m., and you can describe exactly what happened each time and why.
Days 8–10: Turn on missed-call text-back and register it properly
This is the highest-return single change in the whole sequence, and it is also where most rollouts quietly fail. Any unanswered inbound call triggers an immediate SMS to the caller: who you are, that you saw the call, and one specific next action.
What to watch for — compliance, because carriers enforce this mechanically:
- Register your brand and campaign for A2P 10DLC through The Campaign Registry before you send a single message. Unregistered long-code traffic gets filtered by US carriers. The messages appear to send, your dashboard shows delivered, and nothing arrives. Owners chase this ghost for weeks.
- Respect the FCC's TCPA rules, including the 8 a.m. to 9 p.m. local-time window for the called party. An automatic reply to someone who just dialed you is a different animal from a proactive marketing blast, but if you are queuing outbound re-engagement, the window applies.
- Include opt-out language in the first message. STOP has to work, and it has to work automatically.
- If you plan to record calls for quality review, remember that roughly a dozen states — California, Florida, Illinois, Pennsylvania, Washington, Massachusetts and others — require all-party consent. That means a disclosure line in the greeting, not a checkbox in a settings panel.
Wire the text-back so it drops a note on the customer record in your dispatch platform. A text nobody can see in the CRM is a text nobody follows up on. If your stack does not connect natively, the rundown of which AI tools actually connect to Jobber, Housecall Pro, and ServiceTitan covers what talks to what. The mechanics of the automation itself are laid out on the missed-call text-back page.
Done looks like: you call your line from an unknown cell, do not answer, and a text arrives in under thirty seconds — and the interaction shows up on the customer record.
Days 11–17: Make the coverage decision with your own numbers
Now, and only now, decide who answers. Your Day 4 ranking picks the answer for you.
| Your biggest leak | What usually fits | Why |
|---|---|---|
| Rang-out during business hours | Add or reassign human capacity | The problem is bodies and accountability during peak, not technology. A CSR who books at 45 percent pays for themselves fast at your ticket size. |
| After hours | AI voice answering or a live answering service | Nobody staffs 6 p.m. to 7 a.m. profitably at three to twenty trucks. This is the cleanest AI use case in the trades. |
| Overflow / second line | AI as overflow only, human first | Set the AI to catch call two and three while the CSR handles call one. Callers never hear a busy tone. |
| Dead voicemail | Text-back plus a named callback owner and an SLA | This is a process fix. Buying software for it just adds a place for messages to die. |
Most $1M–$5M shops end up hybrid: humans on the primary during business hours, automation on overflow and nights, escalation to an on-call phone for genuine emergencies. That configuration is not a compromise — it is the correct architecture, because it puts the highest-judgment resource on the highest-judgment calls. Your CSR reading a homeowner's tone on a $9,000 system replacement is doing something no automation does. Answering the 11 p.m. "do you service Round Rock" call is not.
If you are weighing the two coverage options head to head, the comparison of an AI receptionist versus a live answering service for contractors breaks down where each one holds up and where it does not.
Done looks like: a written one-paragraph coverage policy naming who answers in each of four time blocks: peak morning, midday, evening, and overnight/weekend.
Days 18–24: Write the escalation rules, then test them by breaking them
Whoever or whatever answers needs explicit rules, written down. At minimum:
- Emergency keywords that force a live transfer immediately: gas smell, sewage, water actively running, no heat with an infant or elderly occupant in the home, sparking panel, exposed wire. These never go to a queue, an AI, or a message. Ever.
- Service area check before anything else. Booking an out-of-area job costs you a truck roll and a bad review.
- The five fields that must be captured every time: name, callback number, service address, equipment or problem type, and access constraints (gate code, dog, tenant-occupied, HOA).
- A hard handoff rule: anything the system cannot answer in two attempts goes to a human with the transcript attached, not back to the top of the menu.
What to watch for: test with hostile calls, not friendly ones. Have somebody call with a heavy accent, with a dog barking, from a moving truck, mid-sentence, angry about a prior invoice. Systems that pass clean tests fail dirty ones, and every real call is a dirty one. Read every transcript from week one yourself. That is a couple of hours you will not regret.
Done looks like: twenty test calls run, twenty transcripts read, escalation rules revised at least once.
Days 25–30: Re-measure the same four numbers
Pull the CDR export again for the last two weeks and repeat the Day 3 tagging. You are comparing four percentages against four percentages. Nothing else counts as proof.
Track these five going forward, monthly, on one page:
- Answer rate inside twenty seconds, business hours
- Unanswered calls by leak type
- Text-back delivery rate (this is where unregistered A2P shows up as a cliff)
- After-hours calls that became a booked job
- Median time from first ring to a slot on the schedule
That last one is the number that quietly governs your close rate. The reasoning behind it is in the speed-to-lead breakdown on answering leads within five minutes.
What does a working call handling system actually look like?
A working call handling system answers every inbound call within twenty seconds, routes emergencies to a human immediately, texts back anyone it could not reach, and logs every attempt in the same place your dispatch lives. It has one named owner who reviews the four leak numbers monthly. It does not require the owner to carry the phone. If any of those four things is missing, you have phone tools, not a system — and the difference shows up in your booked-job rate, not in your software bill.
The two mistakes that undo all of it
The first is buying coverage before fixing routing. If your carrier is dropping the second concurrent call before your software sees it, adding an answering layer changes nothing and you will conclude, wrongly, that the technology does not work.
The second is treating the phone as a cost center. At a $580 average ticket and a one-in-three close, a shop leaking forty new-customer calls a month is leaving roughly $92,000 a year on the table. That is a full CSR seat, a truck payment, and change. The phone is the front door of the business, and unlike your ad spend, the leak compounds — the homeowner who could not reach you at 6 p.m. on Thursday does not just cost you that job. They found somebody else, and that somebody else now owns their annual maintenance.
Build it in this order. Measure first, route second, automate third, staff to what is left. That sequence is the entire method, and it is the same one we walk owners through at Turnkey AI.
Questions owners actually ask
Frequently asked questions
How do I know if my missed calls are a routing problem or a staffing problem?
Look at ring duration in your carrier's call detail records. Unanswered calls with a ring duration under five seconds during business hours are almost always routing or carrier configuration — the call never properly reached a phone. Unanswered calls that rang fifteen seconds or longer while someone was on shift are a staffing and accountability problem. The two need completely different fixes.
Why did my missed-call text-back stop delivering after the first week?
Almost always unregistered A2P 10DLC. US carriers filter unregistered long-code business traffic, and the filtering often kicks in after initial volume builds. Your dashboard will show messages as sent while nothing arrives on the customer's phone. Register your brand and campaign through The Campaign Registry, and confirm your main number is actually SMS-capable — toll-free and landline-ported numbers need separate verification or hosted messaging.
Should an AI answer calls during business hours or only after hours?
Start with after hours and overflow only. Those are the calls no human is realistically covering, so anything is an improvement and the risk is low. Putting automation ahead of your CSR during peak morning hours is where owners get burned, because that is when high-value calls and upset existing customers come in. Once you have thirty days of clean transcripts, you can test it on overflow during business hours.
Can I record calls for quality review?
It depends on your state. Roughly a dozen states require all-party consent, including California, Florida, Illinois, Pennsylvania, Washington, and Massachusetts. In those markets you need a recording disclosure in the greeting itself before the conversation starts. If you operate across state lines, apply the strictest standard everywhere — it is simpler than maintaining two greetings and it is what most multi-market shops end up doing.
What is a realistic answer rate to target?
Ninety percent or better of inbound calls answered within twenty seconds during business hours, and a booked-job rate on after-hours calls that is at least half your daytime rate. Perfect coverage is not the goal — the goal is that no call ends with nothing happening. A caller who reaches an automated system, gets a text, and gets a callback at 7 a.m. is a caller you kept.