Call it yourself before a customer does. From your own phone, run the calls your office gets every week. Write down what it said and what it did, and mark each call right or wrong. Fix the rule that caused each wrong answer, then run the whole set again. Repeat after every change, not just before go-live.
Who this is for: owners and office managers of heating and air or plumbing companies. Use it the week before an AI receptionist goes live, or right after it answers a call wrong. The examples use a heating and air company. A plumbing shop runs the same set with its own emergencies.
How we know: the steps follow the rule we hold for every AI receptionist we build: you hear it take your calls before it goes live. The test set below covers emergency service calls, maintenance calls from service agreement customers, job status checks and callers who want a person.
What should you have ready before the first test call?
Decide the answers first and testing goes fast. Have these five things ready.
- Your rules. Your service area, your hours, what counts as an emergency at your shop, what it may say about price and who gets a transferred call at each time of day.
- A test sheet. One row per call, with five columns: the call, what you said, what it said and did, right or wrong, and what to change.
- Your dispatch board open. A booked call should land there, and you want to see it land.
- Two or three phones. Your cell, a coworker's cell and one number that is not in your customer records, so it hears a new caller.
- A second person if you can. One person calls. The other watches the dispatch board and the texts.
Which test calls should you run?
Start with these seven calls. Run each one twice, worded two ways. Customers do not read from a script, so you should not either.
| Test call | What you say | A right answer does this |
|---|---|---|
| Emergency after hours | "It is late, the furnace quit and the house is freezing." | Confirms it is an emergency under your rules. Gets the name, address, callback number and the problem. Follows your after-hours rule, like reaching the on-call tech. Tells the caller what happens next. Never promises an arrival time you did not approve. |
| Outside the service area | Give an address in a town you do not serve. | Checks the address against your area, says plainly you do not cover it and does not book. If you refer those calls, it gives the referral. |
| Asks for a person | "Can I just talk to somebody?" | During hours, transfers to the right phone on the first ask. After hours, takes a message and says when someone will call back. It does not keep asking questions first. |
| Price question | "How much to replace my AC?" | Says only what your price rule allows, then offers to book the visit or the estimate. It never makes up a number. |
| Maintenance plan customer | "I am on your service plan and I am due for my fall tune-up." | Treats the caller as a service agreement customer, books the maintenance call in the slots you hold for plan members and gives the priority your plan promises. |
| Existing job status | "Your tech said the part was ordered. Any word?" | Shares only what your rules let it see and say. If it cannot see the job, it takes the details, sends them to the office and says when someone will call back. It never guesses a date. |
| Wrong number | "Is this the pizza place?" | Says who you are, ends the call kindly and does not create a job or a customer record. |
For a plumbing shop, swap the emergency for "Water is coming through the ceiling" and the price question for "What does a new water heater cost?" Add a safety call too, like a caller who smells gas. A right answer gives your safety instruction first, before it asks for anything else.
Then add the calls your office gets that this list misses. A home warranty work order. A customer calling back about recommended work from the last visit. A customer asking about an unsold estimate.
How do you call it yourself from your own phone?
Call the real number your customers dial, not a test link. That tests the call forwarding along with the AI.
- Call at the times customers call. During hours, at lunch, after close and on a weekend. After-hours rules only show up after hours.
- Play the caller, not the owner. Do not help it. Interrupt. Give the address in pieces. Change your mind halfway through.
- Call from real places. From the truck, from a loud job site, on speaker.
- Use more than one phone. A number it knows and a number it has never heard.
- Have your office person call too. The person who answers your phones knows the odd calls better than anyone.
The call is not done when you hang up. Check what it left for your team. Did the booking land on the dispatch board? Did the caller get the right text? Did the transfer ring the right phone? Read the dispatch ticket the way your dispatcher would. If a tech would show up without the address, the problem or the gate code, mark the call wrong, even if the conversation sounded fine.
How do you mark each answer?
Mark every call on the test sheet the same way.
- Write what you said as the caller.
- Write what it said and did. Use its exact words where they matter.
- Mark it right or wrong. Almost right is wrong.
- For each wrong answer, name the kind: a missing rule, a wrong rule, a broken handoff or bad wording.
- Write the fix as one plain sentence, like "Callers in towns we do not serve hear that we do not cover their area."
The kind of mistake tells you where the fix goes. A missing rule needs a new rule. A broken handoff needs the connection checked, not new words.
How do you fix each wrong answer?
Here are wrong answers to listen for and where each fix goes.
| What went wrong | Where to look first | What to change |
|---|---|---|
| Booked a caller outside your area | The area is vague, like "the metro" | List every town or zip code you serve and what it says to everyone else |
| Treated a real emergency as routine | Emergency is not defined | Write what counts at your shop, like no heat on a cold night or water that will not stop |
| Woke the on-call tech for a small job | The emergency rule is too broad | Name what waits for morning, like a dripping faucet or a noisy fan |
| Made up a price | No price rule | Write exactly what it may say about price and what it says instead |
| Would not transfer, or rang the wrong phone | The transfer rule or phone list is wrong | Name who gets which calls at which hours, then call it again from your phone |
| Booking never reached the dispatch board | The handoff is broken | Check the connection to your field software, then book a test call again |
| Guessed when a part would come in | No rule for what it cannot see | Tell it to take a message and promise a callback by a time you choose |
| Talked too long or asked the same thing twice | Wording | Shorten the greeting and cut every question you do not need |
Fix one thing at a time. Change three rules at once and get a right answer, and you will not know which change did it.
How do you retest after every change?
After any change, run the whole set again, not just the call that went wrong. A fix for one call can break another. Tighten the emergency rule so it stops waking your on-call tech for a dripping faucet, and it can start treating a real leak as routine. Only the full set catches that.
Retest whenever any of these change:
- Your hours, holidays or on-call schedule
- Your service area
- The phone numbers it transfers to
- Your maintenance plan or what plan members get
- What it may say about price
- The greeting or any wording
- The season, before the first cold snap and the first heat wave
Keep the test sheet after go-live. Every time it gets a real call wrong, add that call as a new row. Over time the sheet becomes a record of every kind of call your shop takes. Once it is live, someone should also listen to its real calls on a set schedule. Our guide on what should happen after you buy an AI receptionist covers that part.
How do you know it is ready for real customers?
Let it take real calls when all of these are true.
- Every call on the sheet came out right on the last full run.
- You called it yourself after hours with your most common emergency.
- Bookings land on your dispatch board with what the tech needs.
- Transfers ring the right phone at every time of day you tested.
- The person who answers your phones called it and agrees it is ready.
If any one is false, it is not ready. One more day of testing costs less than a missed emergency on its first night.
How does FlowChainLabs handle testing?
You do not get a tool to set up. We build your AI receptionist around your trade and your rules, run it, and listen to the calls every month. You hear it take your calls before it goes live. We build it into the systems you already run, teach your team to use it and keep improving it.
See how the call side works on our AI receptionist page, or read how we work with heating and air and plumbing companies.
Going live soon, or did yours just answer a call wrong? Schedule a strategy call. Bring your test sheet or a few recordings of the calls it missed.
