An AI-answered call might be a pest-control request, an invoice question, a sales pitch, or a quick hangup. All four appear in the phone logs. Only some represent new business.
That distinction came through in BrandLyft’s discussion of roughly 2,600 Voice AI calls from a 30-day review. Shawn and Javi described what the calls looked like across locations of a service-business franchise, including pest and critter-control requests. Their point was broader than making sure a ringing phone gets answered.
AI voice lead handling becomes useful when a business separates new inquiries from existing-customer needs. Some calls still require a person. The episode offers observations from real transcripts, but it also exposes some important limits on what call counts can prove.
What the 2,600-call review actually measured
In the episode, BrandLyft describes an internal review of roughly 2,600 Voice AI calls from a recent 30-day period. The calls came from multiple locations of a service franchise. The speakers said they reviewed the transcripts to classify what callers wanted and understand the results of those conversations.
This was not presented as a controlled comparison with human receptionists. Nor did the locations in this particular review have their invoice or final-sale outcomes connected to the call analysis. That distinction matters. A transcript can show a request for service and the information collected. It cannot, by itself, prove the customer eventually booked, paid, or stayed with the business.
The team described a larger call history, but these observations concern the 2,600-call review. Mixing those populations would make the findings harder to interpret.
Almost half the calls were classified as new service opportunities
Javi reported that nearly half the reviewed calls fell into a new-service-opportunity category. That is a meaningful observation, especially for an operation using AI as overflow coverage. It also leaves the rest of the phone traffic to account for.
The speakers described solicitations and robocalls, callers seeking a service the business did not provide, existing customers with billing or account questions, and hangups. They gave the example of people asking a pest or critter-control company to deal with stray dogs or cats, which may fall outside that company’s work. Those requests still take time to sort out when a person answers.
The episode does not give verified counts for the other categories. The practical lesson is not that every business will get the same mix. It is that a single “AI calls answered” number hides several different jobs: identifying a possible new customer, filtering an irrelevant call, and helping an existing customer reach the right team.
A business evaluating AI Voice should decide how each of those outcomes will be labeled. Otherwise, a dashboard can make screening a solicitation look the same as receiving an estimate request.
The after-hours finding deserves its own denominator
The clearest time-of-day finding came from Javi: he said 47% of the identified service opportunities arrived outside business hours. That percentage referred to the service-opportunity group, not necessarily to all 2,600 calls. It was reported from the internal analysis discussed in the recording.
For an owner who assumes most useful calls arrive while the office is staffed, the distinction is worth investigating. Evening and weekend calls can still contain work the company wants to quote. Answering one, however, is not the same as winning it.
The speakers used urgent animal and property problems to explain why a caller may not want to leave voicemail. Those were examples in the discussion, not a verified count of emergency jobs. The review did not establish how many after-hours callers would have gone to a competitor or how many later became paying customers.
A better question is who sees an after-hours request and which cases need human attention. What does the caller hear about the next response?
The phone setup was designed to let office staff answer first
Shawn described a common arrangement across the franchise locations. The customer called the business’s existing published number. During staffed hours, the office usually had three or four rings to answer. When nobody picked up, the phone system forwarded the call to the AI agent. Nights and weekends could follow a different route.
That is not the same as replacing the receptionist. The human team still gets first access to the call when the office is open. AI handles the overflow or coverage gap under the business’s chosen phone rules.
There was also a caution in the presentation: Javi mentioned one location that had a phone problem rather than an AI problem. He did not give enough detail to diagnose that incident, so it would be wrong to assign a technical cause. But it makes the first test obvious. Call the actual customer-facing number during office hours and again after closing. Confirm where the call rings, when forwarding occurs, and whether the AI answers the calls intended for it.
HighLevel documents inbound Voice AI call-flow settings, but an outside office phone system still needs its own forwarding configuration checked. A successful AI test-call screen does not prove the customer’s published number reaches that agent.
Some calls are useful precisely because they are not new leads
The discussion returned several times to screening. Sales pitches about a Google listing, loan offers, and requests outside a pest-control company’s scope could take minutes of staff time. An AI agent may be able to establish what the person wants and provide an appropriate response without tying up the front desk.
The speakers offered possible time-saving examples, not a measured total of staff hours saved in the 2,600-call set. There is no reason to turn those illustrations into a labor-savings percentage.
Existing-customer calls are a different case. Shawn recalled a commercial customer contacting a service location with a question about work already performed. The call was received and the team could follow up. That account supports the value of collecting the request; it does not prove the customer was retained or that revenue was protected.
A billing question needs account follow-up, not a new-lead sales sequence. A wrong-service request needs a clear answer, not a made-up booking. Getting the classification right is part of serving callers, not merely filtering them out.
The summary email is a starting point for human takeover
In the configuration Shawn described, staff could receive an emailed summary after an AI call. The summary helped them see who called and what the person wanted. A team member could then return the call with enough context to begin the conversation.
That is different from a completed handoff. The email tells somebody about a request; it does not show whether that person called back. Nor does the episode supply a measured live-transfer success rate or a count of staff-resolved cases. It discusses options for routing callers to a person, depending on the business, without reporting how often those options were used.
For a current HighLevel implementation, Voice AI settings support post-call notifications, workflows, contact updates, and Call Transfer to a human number. Those are configurable capabilities, not proof that every BrandLyft location used all of them in this review.
A summary needs somewhere to go: a responsible employee, a callback task or escalation, and a visible final status. When a caller asks for someone who can make a service, pricing, or urgent-response decision, that decision still belongs to qualified staff.
Call volume did not answer the conversion question
Javi argued in the episode that capturing more calls may increase the pool of service opportunities without necessarily changing a location’s sales effectiveness. He also described differences in how well locations converted incoming inquiries. Yet the same discussion acknowledged that downstream invoice conversion was not connected for the reviewed locations.
Those statements are useful as hypotheses to examine, not verified before-and-after conversion findings. A call classified as a service opportunity is not a booked estimate. A booked estimate is not a completed job. To measure the commercial result, a business needs reliable records beyond the voice transcript.
Another detail complicates a universal success story. The speakers mentioned two locations that discontinued an earlier implementation. They pointed to issues with earlier versions of the technology and low incoming call volume. The episode does not establish a single confirmed cause for either cancellation.
That is a better buying lesson than a blanket claim that every office needs AI Voice. First examine how many calls the location actually receives, when they arrive, which calls its staff miss, and who will work the resulting requests. AI cannot manufacture demand for a location that rarely gets calls.
What to test before adding another location
The recording was a review of call patterns, not a published acceptance-test report. The following checks are recommendations drawn from the problems and operating choices Shawn and Javi discussed. They are not tests the episode claims BrandLyft completed on all 2,600 calls.
- Office-first routing: Call the public number while staff are available, let it ring unanswered, and confirm the intended AI fallback.
- After-hours coverage: Call outside normal hours with a legitimate service request. Check the information collected and the promised next step.
- Non-sales requests: Try a solicitation, an unsupported service, and an existing customer’s account question. Review the resulting classifications and replies.
- Human assistance: Ask for a person or raise an issue that needs staff judgment. Verify the configured transfer or callback path.
- Post-call ownership: Confirm the right employee receives a usable summary, knows the next action, and can mark it completed.
- Outcome tracking: Check whether the call became a callback, estimate, or closed request. Keep those states separate from an AI-answered count.
HighLevel’s current documentation distinguishes a transfer to another AI agent from a transfer to a human. That difference belongs in the test plan. A caller who needs a qualified person has not received human help merely because the system switched to another voice agent.
The best next question is not whether AI can answer
Across the review, the strongest pattern was the range of calls landing in the same answering channel. Some looked like new service opportunities. Others were existing-customer needs, unwanted pitches, requests outside the business’s scope, or calls that ended quickly. Nearly half of the identified new opportunities were reported outside business hours.
The episode makes a reasonable case for examining overflow and after-hours coverage. It does not show that AI replaced the front office, that every caller accepted it, or that a certain percentage became paying jobs. Those answers require different evidence.
If you’re evaluating this for one location or a franchise network, start with the call route and the person who works what comes out of it. BrandLyft’s AI Voice Solutions is the relevant starting point for the answering setup. Where call summaries, staff tasks, and the CRM’s sales path also need work, a Revenue System Build may be the wider fit.
Want to find out where AI Voice belongs in your call flow? Book a discovery call to review your current phone coverage, existing staff process, and what should happen after the agent answers.
