“I need to change my next visit” is an existing-customer request. If the incoming phone number returns no account, an automated receptionist can answer it with a new-service question.
The customer needs a different appointment, without having to explain the account from the beginning. Good recovery keeps that request intact while the system finds the right record.
In our FieldRoutes build for a US pest-control firm, we traced an AI customer lookup failure to a missing connection: caller-ID details never reached the agent, so appointment lookup received an empty customer ID. Here is the repair and the check that matters afterwards: find the account, then return to the customer's original request.
Why a valid customer can return “not found”
A household may keep the account under one person's number while another person calls. A customer may ring from work, or from a new phone. In each case a search on the incoming number can correctly return nothing, even though the account exists.
The mistake comes in the next decision. If “no match on this number” becomes “new customer”, the agent throws away what the caller has already said. New-lead intake then feels irrelevant because it is irrelevant.
What was supposed to happen, and what did
The design was simple. Before the agent speaks, the caller ID is looked up in FieldRoutes. A match preloads the customer ID, address, city and subscription. For an existing-appointment request, the agent gets the customer's appointments using that customer ID. If nothing matches, it asks for the phone number on the account, runs lookup_customer and carries on with the confirm, reschedule or cancel the caller came for.
During rollout we found the first step was not wired. The preload never reached the agent, so the appointment lookup was handed an empty customer ID. It had nothing to search with. We wired the webhook and added lookup_customer, so the agent could find the account itself from a phone number the caller gives.
A passing test run had missed the real inbound entry point. The repair needed a check through that phone path, with the recovered customer record carried into the appointment action. Testing the lookup by itself would miss the same gap.
Keep the request while the account changes
Illustrative example — not a client transcript.
Caller: I'm calling about our next pest-control visit. We need a different day.
AI receptionist: I'm the AI receptionist. I couldn't find the account using the number you're calling from. Is there another phone number associated with it?
Caller: It is under my partner's number.
AI receptionist: I can use that to look for the account. I'll then follow the account-verification steps before reviewing the visit.
After the account is found, the conversation should go back to changing the visit. The caller should not repeat the reason for calling or answer a new-customer questionnaire.
The second half of the repair is easy to miss. Once lookup_customer finds the account, the appointment lookup has to run again with that account's customer ID, not the empty one the call started with. The first lookup returned nothing. The agent must not treat that as the answer.
Account recovery and permission to change an appointment are separate checks. This build used a phone-number match; our records do not show further verification. The example above adds the verification step we recommend defining with the business before allowing account changes.
- Use what the caller said to recognise that the account may need to be found another way.
- Ask for the account phone number once, without making the person restart their enquiry.
- Run the appointment lookup again with the recovered account, not the empty starting state.
- Resume the original request and confirm only what the tool reports as done.
This build can transfer a caller to a person in office hours and sends a post-call email to the office. Whether a second miss reaches them with the original request intact is exactly what the acceptance check below is for.
How to test the return to the request
“The second search returned a customer” proves a lookup result. It does not prove the agent used it.
Run these three calls.
Unmatched number, existing customer. Call from a number that is not on the account and ask to reschedule. Supply a different, valid account phone. Check that the agent finds the right appointment, keeps the rescheduling request and does not start new-customer intake.
Both numbers miss. Give a second number that is also not on any account. The agent should keep the request for the office and tell the caller it is unresolved. It should not invent a match or list possible accounts.
A genuinely new customer. This caller should still get the new-customer flow. Fixing account recovery should not send every new enquiry hunting for an account that does not exist.
Then check what the tool actually received. This is the practical lesson. Read the customer ID in the appointment lookup's input. If it is empty, the call is broken, however natural the conversation sounded.
Why this matters to the business
Account recovery lets a genuine customer continue an appointment request even when the incoming number does not match. The useful outcome is continuity: find the account, preserve the request and carry on.
Where the account still cannot be verified, give the office a specific task instead of sending the customer through new-lead intake.
Give staff the problem that remains
When recovery fails, the office needs the original request and which lookups were tried, so it can follow its own process for finding and verifying the account.
The caller also needs an honest ending. A reschedule request stays pending until the appointment action succeeds. An account-recovery task is not a completed cancellation or a confirmed new visit.
This is a small part of a pest-control AI receptionist, but it changes the shape of the call. The customer carries on with the reason they phoned, and the system handles the mismatch in the background.
Where to use this check
Prioritise account recovery when existing customers regularly call to manage visits and your system looks them up by phone number. Include the unfamiliar-number case in acceptance testing alongside ordinary new-customer and matched-account calls.