HVAC Busy Season Checklist: Cope With the Summer AC Surge Using AI

Coding Liquids tutorial cover featuring Sagnik Bhattacharya for HVAC Busy Season Checklist: Cope With the Summer AC Surge Using AI.
Coding Liquids tutorial cover featuring Sagnik Bhattacharya for HVAC Busy Season Checklist: Cope With the Summer AC Surge Using AI.

Cope with the HVAC busy season by deciding in advance who gets answered, who gets booked and who waits: put an AI receptionist on overflow and after-hours calls, sort callers into four triage tiers, text back missed calls, move tune-ups out of peak weeks and stock last summer's most-used parts. Start eight weeks before the first heatwave.

AI cannot add technicians, so crew capacity stays the real limit. The plan's job is to protect urgent no-cool calls, keep the backlog honest and stop your office drowning in calls it cannot answer. The checklist below runs in time order, and each item says why it is there and how you will know it is done.

Follow me on Instagram@sagnikteaches

Eight weeks out: write the rules while it is quiet

Rules written in the middle of a heatwave are written badly. Do these six things while the phones are still calm.

Connect on LinkedInSagnik Bhattacharya
  1. Set the daily capacity number. Vans times realistic calls per van after travel. Why: a receptionist, human or AI, needs a number to say yes or no against. Done when: one figure is written down and your dispatcher agrees with it.
  2. Hold a tier-1 reserve. Block a share of each day's slots, say a fifth, for the most urgent calls, and release unused ones at a set hour. Why: first come, first served lets tune-ups eat your emergency capacity. Done when: the calendar shows the block every morning.
  3. Write the four triage tiers (next section). Why: an assistant can only apply rules you have written. Done when: five sample calls are sorted the same way by you and by the assistant.
  4. Decide what the AI may do. Book tiers 3 and 4, take messages for tiers 1 and 2, and never diagnose, quote prices or promise arrival times. Why: scarce same-day slots are a judgement call. Done when: the receptionist's instructions say so in those words.
  5. Settle the fees and say them first. If you charge a priority or after-hours fee, the call and the text-back both state it before dispatch. Why: surprise fees create complaints. Done when: a test call plays the fee line.
  6. Name the on-call owner and a backup, with the two phone numbers that escalations will ring. Why: the best escalation rule fails if nobody answers. Done when: the rota for the next eight weeks exists.

Here is why the numbers matter. Take an illustrative six-van firm: six vans at six calls a day is 36 slots. In the first week of a heatwave, say 85 requests arrive a day. That is a gap of 49 a day, and after five days a backlog of 245 jobs. A tier-1 reserve of a fifth is seven slots, which leaves 29 for everything else. No software closes that gap; rules decide who falls into it. Without rules, the gap is filled by whoever rang first, and the caller with an older relative and no cooling waits behind a routine tune-up.

Subscribe on YouTube@sagnikteaches

Four triage tiers an answering agent can apply

Tiers work when each is defined by what the caller says, not by what you suspect is wrong. An assistant, like a new call handler, cannot diagnose. It can listen for phrases and ask a fixed set of questions. Here are four tiers as a starting point; adjust the promises to what your crews can deliver.

TierThe caller saysWhat the agent doesWho acts
1. Safety firstA gas smell, smoke, a burning smell, sparks or a sounding carbon monoxide alarm; or no cooling in a home with a baby, an older person or someone with a health conditionFor gas, smoke, burning or an alarm: tells the caller to leave the building and call the emergency number, then alerts the on-call owner by call and text at once. For a vulnerable person with no cooling: takes the details and alerts the ownerOn-call owner, immediately
2. Same dayThe system has stopped cooling or is leaking water indoors; no safety words; nobody vulnerableTakes the intake details and promises only a callback with a time windowDispatcher, from the tier-2 queue
3. SoonWeak cooling, new noises, thermostat faults, minor drips outsideTakes the details and books the next open slotAgent or dispatcher
4. PlannedTune-ups, replacement quotes, membership questionsBooks a shoulder-week slot or routes to salesAgent, then sales

The intake details are the same every time: name, mobile number, address, the problem in the caller's own words, system age if known, anyone vulnerable at home, access notes such as a gate or pets, and the best callback time. Put the rules into the receptionist's instructions in plain words.

You answer calls for [business], an air-conditioning company. You never diagnose, quote
prices or promise an arrival time.
Collect: name, mobile, address, the problem in the caller's own words, system age if known,
whether anyone at home is a baby, an older person or has a health condition, access notes,
best callback time.
Assign a tier:
TIER 1: gas smell, smoke, burning smell, sparks, carbon monoxide alarm, or a vulnerable
person at home with no cooling.
TIER 2: not cooling or leaking indoors, with no TIER 1 signs.
TIER 3: weak cooling, noises, thermostat faults, minor drips outside.
TIER 4: tune-ups, replacement quotes, membership questions.
For gas, smoke, burning or an alarm say: "Please leave the building now and call the
emergency number. We will follow up straight after." Then alert [on-call number] by call and text.
If unsure between two tiers, choose the more urgent one.
For TIER 1 and 2 take a message only. For TIER 3 and 4 offer slots from OPEN SLOTS.
OPEN SLOTS: [list]

Have whoever handles safety in your business approve the wording for gas, smoke and alarms against your insurer's or trade body's guidance before it goes live. Then test it. Here is how an assistant might sort six illustrative calls, and what to change.

CallAssistant's tierCorrected tierWhy it changed
"AC has stopped, the house is boiling and my mother, who is 84, lives with us."21A vulnerable person with no cooling is tier 1 under the rule; the assistant gave it the routine tier
"There is a burning smell from the outdoor unit."1, but it said a technician would arrive within two hours1The tier was right; the arrival promise was not, and the leave-the-building line was missing
"It is cooling, but not very cold."33No change
"Can I book my yearly tune-up?"44No change; offered a shoulder-week slot
"Water is dripping from the indoor unit onto the ceiling."32Water indoors can damage property and may reach electrics; ask whether it is near sockets or lights, and treat yes as tier 1
"I want a quote for a new system for a four-bedroom house."44Routed to sales; no price given

Two of six needed changing, and both were errors that the "choose the more urgent one" rule exists to catch. Add the vulnerable-person trigger and the water question to the prompt, then run the six calls again. When all six land where you would put them, add ten more from your own call history, because those are the odd ones a generic list misses.

The Complete Excel Course with AI Integration thumbnail

The Complete Excel Course with AI Integration

From formulas and pivot tables to Power Query, DAX, VBA and AI with Copilot, ChatGPT and Claude — hands-on from the first lesson.

By Sagnik Bhattacharya and Virginia Thorn, Primetime Emmy Award winner

Buy now

Overflow answering: choose the trigger, the hours and the hand-off

There are three coverage modes: the AI answers every call, only after hours, or only when your lines are busy or unanswered. All calls, after hours or overflow for a plumbing company compares them for a neighbouring trade, and the logic carries over. For a surge, overflow plus after-hours is the usual fit: your people keep the calls they can reach, and nothing rings out.

The trigger is normally set outside the AI product. Jobber's Receptionist page, for example, says you set up forwarding with your current phone provider, to kick in after hours or after a set number of rings. If the AI product has its own answering delay as well, the two add up, so time the whole wait from a mobile before the surge. Start at four rings and adjust after listening to real calls: long enough for a free dispatcher to pick up, short enough that the caller does not give up.

Then test the hand-off. Jobber describes an escalation policy: you list trigger terms such as "emergency" or "urgent", and a matching conversation either transfers the call to your escalation contact or sends you a text alert with a snippet of the conversation. It also handles several calls at once and turns each message it takes into a follow-up task. Smith.ai's AI Receptionist page lists 24/7 call answering and transfers to its live receptionists when a caller needs a person. Whatever you choose, run the tier-1 phrases through it, because a keyword list without "smoke" or "burning" misses the calls that matter most.

Every message the AI takes becomes a callback for somebody, so decide who owns the queue before the surge starts. Fixed callback batches work better than a constant trickle: tier 1 immediately, tier 2 at set times such as mid-morning, early afternoon and late afternoon, tier 3 and 4 by text. One person owns the queue per shift. Without that, you have swapped a ringing phone for a growing list.

Here is what overflow could cost in a month with 400 overflow calls, using list prices as of October 2026. Vendors define a unit differently (a conversation, a call), and a caller who rings three times can use three, so check each definition before you rely on the sum.

OptionPricing basisCost for 400 calls
Jobber Receptionist add-on$29 a month with 30 conversations, then $0.79 each; unlimited on Plus$321.30
Smith.ai AI Receptionist, Pro at the 300-call tierAbout $500 a month for 300 calls ($1.67 a call), then $2.17 a callAbout $717
Smith.ai AI Receptionist, Free25 calls free, then $3 a call$1,125
Smith.ai human-answered plan, Pro$2,100 a month for 300 calls, then $8.50 a call$2,950

Smith.ai's Pro plan starts at $150 for 75 calls with extra calls at $2.50; on that entry tier, 400 calls would cost $962.50. Pick the tier nearest your real peak-month count rather than the entry price.

If you run Google's pay-per-lead ads, check how calls are billed before the surge starts. Google's lead-billing help says a missed call during business hours is charged once the caller has stayed on the line for more than 20 seconds, and that voicemail or an automated greeting counts as a connected call. Business hours now sync from your Google Business Profile, so keep them accurate before the heat arrives.

Text-back that buys an hour, not a promise

A missed-call text-back is an automatic text to a caller you did not reach. It cannot fix the missed call, but it can keep the caller from ringing your competitor, provided it says something useful. Quo, for example, offers auto-replies to texts, missed calls and voicemails, and sends one auto-reply per contact every three hours, so a caller who rings four times in an hour does not get four texts. Its pricing page lists auto-replies on every plan. Missed call text back for a service business covers the mechanics.

The wording is where the damage happens. These are illustrative drafts.

1. Sorry we missed you! A technician will be there ASAP.
2. Thanks for calling [business]. We are taking a very high number of calls. Reply with your address and what is happening and we will text back within [x] hours. If you smell gas or smoke, or a carbon monoxide alarm is sounding, leave the building and call the emergency number.
3. Thanks for calling [business]. For tune-ups and quotes, book here: [link]. For no cooling, reply with your address.

Version 1 promises something nobody can deliver, and "ASAP" in a heatwave is a complaint waiting to happen. Version 2 is the one to use: it asks for exactly the details the tiers need, it says when to expect an answer, and it carries the safety line. The [x] must be a number your crews can meet. Version 3 separates planned work from urgent calls, but test whether links are delivered, since carriers can block business texts that contain them. Add a stop line if your provider requires one.

Measure the text-back by replies, not by sends. If 60 missed calls got the text and 24 replied with an address, that is 40 percent, and each reply is a job you would otherwise have lost. An assistant can read the replies and assign tiers, using the same rules as the receptionist.

Deferring tune-ups without annoying members

Tune-ups are the most movable work you have. Spring is the time to sell them and the peak weeks are the time to protect capacity. Three rules keep members on side: they keep their priority for breakdowns, nobody's booked visit moves without asking, and no new tune-up promotion runs in the peak weeks. The spring campaign that fills the shoulder weeks is laid out in a worked spring and autumn tune-up campaign for HVAC.

Do the sum first. Say 120 tune-ups are booked inside the six peak weeks, at an assumed hour each. Moving them frees 120 technician-hours, which is about three technician-weeks at 40 hours. That is real capacity, and it costs you nothing but a message.

Ask the assistant for three versions, one per situation, and give it the shoulder dates and what you are willing to offer. Sample drafts follow.

Member with a booked visit: Hi [first name], your tune-up is booked for [date]. We are in our busiest weeks and would like to keep that day free for urgent no-cool visits. Would you rather keep it or move to [shoulder weeks]? Reply 1 to keep, 2 to move.
Member not yet booked: Hi [first name], your tune-up is due. We are holding tune-ups for [shoulder weeks] so urgent visits come first. Reply with a week that suits you and we will confirm. Your plan benefits for breakdown visits are unchanged.
One-off customer: Thanks for asking about a tune-up. Our next openings are in [shoulder weeks]. Reply with a week and we will book you. If your system has stopped cooling, tell us now and we will treat it as urgent.

The first drafts needed three edits. One said "we are too busy for tune-ups right now", which sounds like a brush-off, so the reason became "to keep urgent visits moving". One promised "a free filter credit", marked ADDED, and deleted because nobody had agreed it. And "your plan benefits are unchanged" stays only if it is true under your written terms, so check them before sending.

Parts prep before the first heatwave

Last summer's invoices are the best forecast you have. Export the parts lines, count by part per week, find the four busiest weeks and set reorder quantities from the peak week. Let the assistant do the grouping and let a sheet do the sums.

Below is an export of parts used on service invoices from last June to August:
date, job number, part description, quantity.
1. Group descriptions that are the same part written differently, and list the groups so I can
   check them. Do NOT merge parts with different ratings, sizes or models.
2. Count quantity per part per week.
3. Name the four busiest weeks overall.
Do not forecast. Output two tables: the group mapping, and the weekly counts.

[paste export]

A typical first-run error is merging two capacitors with different ratings into one line because their descriptions look alike. Different ratings are different parts, which is why the prompt forbids it. Always read the group mapping before you accept the counts.

My rule of thumb for the reorder quantity, which is a judgement call and not an industry standard: cover the supplier's lead time plus one week at the peak-week rate, with 20 percent on top for a hotter summer, then subtract what you hold. In a sheet that is =ROUNDUP(B2*(C2+1)*1.2,0)-D2. The invented numbers work out like this.

PartPeak-week useLead time (weeks)On handCover targetTo order
Run capacitors221155338
Contactors141103424
Condenser fan motors6232219
Condensate float switches9142218
Thermostats11182719
Blower motors3311514

Notice what the table does not say: nothing here claims which parts fail most in summer. Your own invoices say that, for your customers' equipment. Before you place orders, ring the supplier to confirm lead times and any limits on quantities in a hot week, and ask for a delivery slot before the first heatwave. Then top up the vans; van stock management for HVAC and electrical vans covers reorder levels per van.

The 8 am check: queue, slots and an honest wait time

In season, the dispatcher's day starts with ten minutes that matter more than any software setting. Pull the overnight messages and call summaries, sort them into the four tiers, review the oldest open job in each, count today's slots, decide what rolls to tomorrow and update the wait time you quote to customers. Picture one morning at the 36-slot firm, with a seven-slot tier-1 reserve and every number invented.

QueueOpenOldestAction this morning
Tier 121.5 hoursBoth assigned before 8:30; owner confirms. Five reserve slots stay free until 2 pm
Tier 22420 hoursAll 24 scheduled today, oldest first
Tier 3383 daysFive slots now and five more when the reserve releases at 2 pm; at that pace the queue clears in about four days, so the wait text says four days
Tier 418Not applicableHeld for shoulder weeks; batch message goes at 10:00

The slots add up: two for tier 1, 24 for tier 2 and ten for tier 3 make 36. The assistant can prepare the sheet from the overnight messages if you stop it improvising.

Here are overnight call summaries and texts, and my tier rules. For each tier give the count and
the oldest item, and list any vulnerable-person flag first. Do not change a tier I have already
set, never lower a tier because the caller says "not urgent", and do not suggest times.
RULES: [paste the tier rules]
MESSAGES: [paste]

The "not urgent" instruction exists because of an illustrative failure: the assistant moved a night message from tier 1 to tier 2 because the caller had written that it was not urgent, when the same message mentioned a burning smell. A person decides when a tier goes down.

Then update the wait you quote in every place a customer hears it: the voicemail greeting, the text-back wording and any website banner. Quoting a wait you can meet is cheaper than apologising for one you cannot. Agree overtime and on-call arrangements with the crew before the season, not on the day, and if you plan to bring in extra crew, settle the conditions (licensing, insurance, who owns the customer relationship) before the first heatwave. Callers whose problem sounds like a real emergency are the subject of setting up an AI answering service that never misses an emergency, and training staff and the AI on the same words is covered in the HVAC call script for CSR staff and AI.

A scoreboard of five numbers, and when to change course

Check five numbers every morning, each with a trigger you set in advance. The readings here are illustrative.

NumberIllustrative readingChange course when
Share of calls answered by a person or the AI96 in 100It falls below your target for two days: shorten the ring count before overflow
Missed calls with a text-back inside five minutes91 in 100It drops after any phone-system change: recheck the auto-reply settings
Tier-1 jobs assigned within your target time2 of 2Any miss: review the escalation path the same day
Oldest open tier-2 job20 hoursIt passes your limit, say 24 hours: move a tier-3 crew onto tier 2
Quoted wait for tier 3About four daysIt passes what you are willing to promise: stop new tier-3 bookings and switch to waitlist wording

When the surge ends, keep the numbers. Log calls by hour, the backlog by day, texts sent and answered, parts used and what customers complained about, then ask the assistant for a one-page review: what the plan handled, where it broke, and what to change before next year. Put next year's three-weeks-before date in the calendar while the memory is fresh. The first heatwave always arrives sooner than the plan expects.

Questions before the first heatwave

Will callers accept an AI answering?

Many will if it is quick, clear about what happens next and hands urgent calls to a person. Say at the start that an AI is answering, and check the rules on recording calls where you work before you switch recording on. Test it with your own staff first and listen to the recordings before it takes a real caller.

Should I charge more during a heatwave?

That is a pricing decision for you. If you charge a priority, after-hours or emergency fee, state it at the start of every call and text before anyone is dispatched, and write it into the receptionist's script. Check the consumer rules where you work, because disclosure requirements differ.

Can the AI book appointments by itself?

Some can. Jobber's Receptionist can book during the call using the online booking settings you have already configured. I would let it book tier 3 and tier 4 calls only and take messages for tiers 1 and 2, so a person always decides who gets the scarce same-day slots.

When should I switch it all on?

About three weeks before the first forecast heatwave: one week to configure and test, one to listen to real calls, one to fix what you hear. Look at last year's first hot week and put a date in the calendar now, not when the weather turns.

Further reads

Sources: Jobber Receptionist help article and product page; Smith.ai pricing pages for AI Receptionist and human-answered plans; Quo pricing page (auto-replies); Google Ads Help on pay-per-lead goals and Google's lead-billing help (missed calls, voicemail and automated greetings). Call volumes, parts counts and costs in the examples are illustrative.

Want your surge plan tested before the heat?

On a 1:1 call we can walk through your call flow, capacity and parts list, decide which parts of the surge plan AI can run safely, and leave you with rules your dispatcher can follow.

Book a 1:1 call with me