AI receptionist for heating and cooling companies

Balaay is an AI receptionist for businesses that run on calls. Balaay is an AI receptionist for businesses that run on calls. It learns a business from that business’s own website, answers its phone, holds the conversation, takes the next step where it can, and records what it actually did.

“My AC stopped working and I need someone today.”

Knows already: Your service area, your opening hours, and the services your website lists. It read those from your own site during setup.

Asks: The address the system is at, what the system is doing or not doing, and roughly what kind of system it is if they know — so the engineer brings the right parts.

May do: Capture the job with the address and the fault, mark it as a cooling failure, and hold a slot in Balaay’s own schedule if you have given it one.

Will not invent: Whether a technician is free today. It does not see your engineers’ diaries, so it does not promise a same-day visit.

Escalates: If the caller describes extreme heat with a vulnerable person, it flags the call for a person rather than deciding how urgent that is.

Verified: If it says it took the job, the record shows the address, the fault and the time it was captured. If a step failed, the call is marked as needing attention rather than booked.

Left for a person: Deciding who goes, when, and whether today is possible.

“My furnace won’t start.”

Knows already: That heating faults in cold weather are on your urgent list, because that is how the trade profile is configured.

Asks: The address, what happens when they try to start it, and whether there is any heat at all in the property.

May do: Capture the fault, flag it as no-heat, and record it as urgent so it is not sitting in a queue behind a maintenance enquiry.

Will not invent: A diagnosis. It will not tell a caller the ignitor has failed or what the repair costs.

Escalates: No heat in freezing weather is flagged for a person. Balaay does not grade how much of an emergency that is.

Verified: The urgency flag and the captured details are both on the record, so nobody has to take the receptionist’s word for what was said.

Left for a person: The judgement of whether this is tonight or tomorrow morning.

“My boiler is making a banging noise.”

Knows already: The services you list, and whether boilers are among them.

Asks: The address, when the noise happens, and whether there is heat and hot water.

May do: Capture the symptom in the caller’s own words rather than translating it into a fault code it cannot verify.

Will not invent: What the banging is. Kettling, a failing pump and trapped air are not distinguishable over the phone, and guessing sends the wrong parts.

Escalates: If the caller also reports a smell or a leak, the safety path takes over and the booking becomes secondary.

Verified: The description is recorded verbatim, so the engineer reads what the customer said and not a summary of it.

Left for a person: Diagnosis, and deciding what to bring.

“I smell gas.”

Knows already: That this is a life-safety matter and not a booking.

Asks: Nothing that delays them.

May do: Tell the caller to leave the property and contact the emergency gas service or emergency services now, and flag the call immediately.

Will not invent: Reassurance. It does not suggest the smell might be nothing, and it does not offer to book an engineer instead.

Escalates: Immediately, and the call is recorded as a safety escalation rather than a lost booking.

Verified: That the safety message was given and the call was flagged.

Left for a person: Everything after that. This is the one call where the correct outcome is no appointment.

“Do you cover my postcode?”

Knows already: The service area stated on your website.

Asks: The postcode or the town.

May do: Answer from what your site says, and take the details anyway if the answer is unclear so you can decide.

Will not invent: Coverage. If your site does not say, it says it will check rather than guessing yes.

Escalates: Not usually. An out-of-area caller is captured as a lead with a note, not turned away silently.

Verified: What it told the caller about coverage is on the record, so a wrong answer is findable.

Left for a person: Deciding whether to stretch for a job just outside the area.

“How much is a callout?”

Knows already: Only what your website publishes.

Asks: Whether they would like a call back with a price.

May do: State a price only if your site publishes one, and otherwise offer a callback.

Will not invent: A number. This is the single most common place an AI receptionist damages a business, because a quoted price the business does not honour is worse than no answer.

Escalates: A pricing question with a booking intent is captured so somebody can call back with a real figure.

Verified: That it declined rather than estimated. A refusal is recorded as an outcome in its own right, not as a failure.

Left for a person: The price.

“Can someone come out tonight?”

Knows already: Your published opening hours.

Asks: The fault and the address, before anything about timing.

May do: Say what your hours are, capture the job, and flag it if the fault is on the urgent list.

Will not invent: An out-of-hours visit. It has no on-call rota, so it does not commit an engineer to a night call.

Escalates: An urgent fault outside hours is flagged for a person immediately.

Verified: The time of the call and the flag are both recorded, which is what an argument about response time later turns on.

Left for a person: Whether anyone is going out tonight.

“What brands do you service?”

Knows already: The brands your website names.

Asks: Which system they have.

May do: Answer from your site, and capture the enquiry either way.

Will not invent: A brand list. If your site does not name brands, it does not produce one.

Escalates: No.

Verified: What it claimed about brands, so an incorrect answer traces back to the page it came from.

Left for a person: Confirming an unusual system is one you will take on.