The missed-call text-back workflow an HVAC company actually needs.

A missed-call auto-reply does one useful thing: it tells a caller they have not been ignored, quickly, while they are still deciding who to call next. That is worth having. It is also the entire extent of what the feature does on its own — it does not create an owner, classify urgency, decide who calls back, or notice when the customer replies.

Treated as a complete solution, it produces a familiar failure: the caller receives a friendly message, replies to it, and nobody answers. That is worse than silence, because the business has now demonstrated automation without follow-through. The workflow below is the part that has to exist around the message.

For businesses running GoHighLevel, the built-in setting is documented under the voice configuration for the number, as Voicemail & Missed Call Text Back. HighLevel's documentation describes it as automatically sending a text message to a caller when their inbound call is missed, after the incoming-call timeout period, which is adjustable within a short range.

Three documented behaviours matter more than the feature description. It does not deduplicate: HighLevel states that the feature triggers a message for every missed call, even if the same caller tries several times within a brief timeframe. It respects contact preferences: DND settings continue to apply. And personalisation is conditional, because merge fields such as the contact's name only resolve if the system already recognises that contact from a prior interaction.

That first behaviour is the one to design around. An anxious customer with no heat in February will call three times in five minutes. Without a deliberate decision, they receive three identical texts, which reads as a malfunctioning business at precisely the moment the business is trying to look responsive. HighLevel's own documentation points at the fix — building the response as a workflow with a wait step or a tag filter rather than relying on the standalone setting.

Platform behaviour changes. Everything stated here about GoHighLevel is taken from HighLevel's own support documentation, cited at the end of this article, and should be re-checked against it before a build rather than taken from any agency description, including this one.

Each state below names who holds it and what moves it on. The owner column is the point of the exercise: a state whose owner is nobody is where missed calls stop being recoverable, and it is usually one specific state rather than a general lack of process.

1. 01

   Call missed

   Owner

   The phone system

   Advances when

   The inbound call ends without being answered, after the configured ring timeout.

   Ends when

   The caller reaches a person. An answered call does not enter this workflow at all.

2. 02

   Acknowledgement sent

   Owner

   The workflow

   Advances when

   An automatic reply is delivered, where the channel is appropriate and preferences allow it.

   Ends when

   The contact has messaging DND set, in which case no message is sent and the call moves straight to callback.

3. 03

   Repeat-call check

   Owner

   The workflow

   Advances when

   A second or third call from the same number within a short window is recognised as the same event rather than three.

   Ends when

   No further acknowledgement is sent for the same event; the record is escalated instead, because repeat calling is an urgency signal.

4. 04

   Record created or matched

   Owner

   The workflow

   Advances when

   The caller is matched to an existing contact, or a new one is created with the number and the source attached.

   Ends when

   The record exists. Everything after this point depends on it, which is why this state is not optional.

5. 05

   Classified

   Owner

   The workflow, then a person

   Advances when

   The call is provisionally classed — no heat or no cooling, deferrable repair, replacement enquiry, maintenance, or unknown.

   Ends when

   A person confirms or corrects the class. Automation may propose the classification; it should not be the last word on an emergency.

6. 06

   Owner assigned

   Owner

   One named person

   Advances when

   The assignment rule resolves, using classification rather than availability, with a defined fallback.

   Ends when

   This state does not end until a person holds the record. If the rule can resolve to nobody, this is the state where work is lost.

7. 07

   Awaiting reply

   Owner

   The assigned owner

   Advances when

   The customer replies, or the callback window expires — whichever happens first.

   Ends when

   Any inbound reply. The moment a customer responds they are talking to the business, and no sequence should continue over that.

8. 08

   Callback attempted

   Owner

   The assigned owner

   Advances when

   The business calls back within the window the acknowledgement implied.

   Ends when

   Contact is made, or the agreed number of attempts is exhausted and a reason is recorded.

9. 09

   Booked or quoted

   Owner

   The assigned owner

   Advances when

   A visit is scheduled, or an estimate is issued and enters its own follow-up states.

   Ends when

   The opportunity moves into the estimate lifecycle, which has its own owner and stopping rules.

10. 10

    Stopped

    Owner

    The assigned owner

    Advances when

    A stopping condition fires: the customer accepts, declines, asks for time, opts out, or is already a live job.

    Ends when

    Automation ceases immediately. A deferral records a date and an owner on that date, or it is not a deferral.

11. 11

    Outcome recorded

    Owner

    The assigned owner

    Advances when

    Won, lost or deferred is written to the record with a reason specific enough to change something.

    Ends when

    The workflow is complete. Without this state the pipeline fills with calls nobody can act on and the reporting describes a business that does not exist.

The generic version of this workflow treats every missed call the same way. HVAC cannot afford to. A household without heat in February is making a decision within the hour, and often within minutes; a customer asking about a spring maintenance plan is not. Sending both the same acknowledgement and putting both in the same queue means the second one is over-served and the first is under-served, on the day when under-serving is most expensive.

Classification therefore has to happen early and cheaply. Some of it can be inferred — the number called, the time of day, the phrasing of the reply, whether the caller is an existing maintenance customer. The rest requires a person, quickly. A useful design has automation propose a class and route accordingly, while making it trivial for a person to correct it, and never letting an unclassified emergency sit in a general queue because the inference was not confident.

The repeat-call signal deserves its own mention here. In most businesses, three calls in ten minutes is a data-quality nuisance. In HVAC it is the strongest urgency signal available, and it arrives before anyone has spoken to the customer. Treating a repeat call as an escalation rather than as a duplicate to be suppressed is the difference between a deduplication rule and an operating rule.

Because the platform feature fires per missed call rather than per caller, the deduplication has to be designed. The mechanism is not complicated — a workflow that checks whether this contact has already been acknowledged within a defined window, and branches rather than sending again. HighLevel's documentation recommends exactly this approach, using a wait step or tag filters, rather than relying on the standalone setting.

What needs deciding is the behaviour, not the mechanism. Within the window, a second call should escalate the existing record rather than create a second one: raise its priority, notify the assigned owner directly, and leave the customer with one conversation rather than three. Outside the window — the next day, or the following week — a fresh acknowledgement is correct, because it is a new event.

There is a second duplicate to guard against, which is quieter. A caller who also submits a web form, or messages the business on another channel, can arrive as two records and two sequences. The result is the same customer receiving parallel conversations from what is supposed to be one business. Matching on the phone number and the email at the point of capture, rather than after the fact, is what prevents it.

The stopping conditions are the most important part of the build and the part most often added last. Any inbound reply ends the sequence: the customer is now talking to the business, and a scheduled message arriving over that conversation is a visible failure. Acceptance, decline, a request for time and an opt-out all end it too, each writing a different outcome to the record.

Beyond stopping, there is handoff. Deciding whether a fifteen-year-old system is worth repairing, explaining why one quote is cheaper than another, judging whether a call is an emergency when the customer is not sure, and handling a household that is cold and unhappy are human work. What the system should do is reach those moments faster and with better information — the call history, the classification, the equipment record, the previous quote — so the person picking it up is not starting from nothing.

It is worth naming the thing that has to be true for any of this to work: someone has to call back. A well-designed acknowledgement that is never followed by a human contact converts worse than no acknowledgement at all, because it sets an expectation the business then fails. The workflow is not the message. The workflow is the guarantee behind it.

One distinction from HighLevel's own trigger documentation is worth carrying into any platform. Its Call Details trigger separates Call Status — a system-recorded outcome such as busy, voicemail, no-answer or completed — from Custom Disposition, which is a post-call classification a person selects afterwards, such as follow up, qualified or requested callback.

The distinction matters because the two fail differently. A system status is always present and never interpreted; it tells you the call was not answered but nothing about what it was. A disposition carries the meaning, and is only as reliable as the discipline of the person selecting it. Building an urgent escalation on a disposition means the escalation depends on somebody remembering to classify a call they did not answer, which is the assumption most likely to fail on a busy day.

The practical rule: automate on the system-recorded fact, route to a human for the meaning, and record the meaning as a state on the opportunity rather than leaving it only in a call log. That way the pipeline can be read without listening to anything.

Sending SMS from a standard ten-digit US number requires A2P 10DLC registration, which HighLevel documents as a Brand identifying the business and a Campaign describing the messages it will send. Submission runs through the platform, but the documentation is explicit that carriers and their registration partners decide the outcome. It can be rejected and it takes time, so it belongs at the start of a project.

Consent has a home in the platform too. DND can be set globally or per channel, HighLevel documents that a contact replying with an opt-out keyword such as STOP or UNSUBSCRIBE enables DND, and a contact marked DND is excluded from receiving communications on the channels their DND settings name. That is a floor, not a strategy. A missed-call reply to someone who called the business moments earlier sits comfortably within most people's expectations; a follow-up sequence sent weeks later to a number captured elsewhere may not.

The requirements themselves depend on the business, its jurisdiction, its consent basis and its messaging provider. Confirm them with qualified counsel and with the provider before an automated sequence carries live traffic. Nothing here is legal advice, and no claim is made about conversion, revenue, booking rates or response-time returns — Aevrion publishes no HVAC client results, because none exist to publish.

1. An automatic reply buys attention. A named owner and an actual callback are what convert it, and without them the message makes the business look worse rather than better.
2. HighLevel documents that Missed Call Text Back fires on every missed call, including repeat calls from the same person in a short window. Design the duplicate behaviour deliberately.
3. In HVAC, a repeat call is the strongest urgency signal available before anyone has spoken to the customer. Escalate it rather than suppressing it as a duplicate.
4. Every state in the workflow needs an owner. The state whose owner is nobody is where recoverable calls stop being recoverable.
5. Any inbound reply ends the sequence. A scheduled message arriving over a live conversation is a visible failure.
6. Automate on the system-recorded call status; route to a person for the meaning; record the meaning as a state on the opportunity, not only in the call log.
7. US SMS needs A2P 10DLC Brand and Campaign registration, and HighLevel documents that carriers decide the outcome. Start it early.
8. Deferred without a date and an owner is not deferred. It is lost, with a friendlier label.