Everyone asks us what our AI phone agents can do. The more interesting question is what they do when they can’t help.
We looked at 862 calls where our AI forwarded a customer to a real person. The patterns tell you a lot about what good AI actually looks like in production.
Pattern 1: “Let me talk to a human”
Some callers ask for a person in the first 5 seconds. Our AI doesn’t argue, doesn’t try to prove it can help, doesn’t run through a script. It says okay and connects them. Average call before handoff: 16 seconds.
One of our fencing businesses had 31 of these forwarded calls. Most were under 20 seconds. The caller said “I need a person,” the AI said “connecting you now,” and it was done.
No ego. No friction.
Pattern 2: The diagnostic handoff
These are the best calls. The AI gathers real information, diagnoses what it can, then hands off when it hits its limit.
An HVAC customer described his AC unit “roaring but not blowing air.” Our AI suggested it might be a fan motor or capacitor problem, noted the details, then transferred to a live agent to schedule a tech visit. 92 seconds total.
Another call: a customer needed an Aprilaire 1410 whole house air purifier filter replacement. The AI confirmed the specific filter type, then transferred to scheduling. 111 seconds.
The AI didn’t pretend to book the appointment or guess at pricing. It collected the technical details that a human scheduler would need anyway, then got out of the way.
Pattern 3: The pricing wall
A plumbing customer had an assessment already scheduled but asked about potential charges. Pricing policy is something the AI doesn’t have authority over, so it forwarded to an operator. 96 seconds.
Another caller wanted a detailed quote breakdown for a 40-gallon gas water heater, specifically manual, labor, and unit costs. The AI collected the customer’s name and number, then transferred to someone who could actually price it.
This is the right call. Giving wrong pricing is worse than not giving pricing at all.
Pattern 4: Asking for someone by name
Callers asking for the owner by name, asking for “an owner or operator,” or calling with a request that has nothing to do with the business itself. The AI doesn’t try to redirect these. It recognizes “I need to talk to Francisco” isn’t something a script can handle and forwards immediately.
The warm transfer
For the most complex conversations, the AI does a warm transfer, stays on the line, introduces both parties, then drops off. We’ve seen 36 of these. Average duration: over 2 minutes. These are the calls that need real human judgment.
Of those 36, 23 customers stayed on after the transfer attempt and 13 hung up before it completed. That’s a 64% connection rate on what are inherently the hardest calls.
What this means
The AI isn’t trying to replace humans. It’s trying to make sure the right calls get to the right people, with the right context, at the right time.
When it can handle a call end to end, it does (8,500+ calls completed by the AI). When it can’t, it gets out of the way fast and clean. No hold music. No “I’m sorry, I didn’t understand that.” Just a smooth handoff with the context the human needs to actually help.
That’s not a limitation. That’s the feature.
Niche Team
Product