Separate conversation fallback from system recovery
A conversation fallback handles a request the AI should not resolve alone. It may transfer to staff or capture information for review, depending on the configured workflow. System recovery addresses a broader interruption, such as a failed phone route or unavailable integration. These require different owners and actions.
Write the triggers in plain language. Examples include an unresolved guest request, repeated failed transfers, an unavailable ordering connection or unclear payment state. Do not optimize for a containment percentage at the expense of an appropriate handoff.
Verify a destination that can actually receive calls
Test the staff endpoint from the intended route. Check ring behavior, hours, unanswered behavior and whether the call carries useful context. Confirm that the destination will not forward back to the AI and create a loop. Keep the provider-specific configuration with the phone administrator.
If staff are unavailable, choose a policy the restaurant can actually fulfill. Capturing a message is different from an immediate response. The caller should receive an accurate expectation. Loman documents several escalation options, but availability depends on the route and staff configuration.
Prepare a restoration runbook
Before launch, save the current route using the provider's supported method. Write the exact restoration steps, the authorized operator and the verification call. Do not store passwords in the runbook. Rehearse the change with an approved test arrangement where possible.
For a number port, restoration may not be a simple settings change. Use a provider-approved service-continuity plan and identify what can be done during a delay. The rollout planner treats a missing port plan as a blocker rather than assuming instant reversibility.
Reconcile orders and reservations after an interruption
Restoring the phone route does not resolve every transaction in flight. Check whether an order, payment or booking was created before manually recreating it. Confirm with the responsible system owner and use the provider's approved process for incomplete or duplicate records.
Keep a short operational log: time of interruption, affected route, suspected scope, restoration time and unresolved transactions. Collect only the data needed to resolve the issue. Ask the restaurant's responsible team about access, retention and incident-handling requirements.
Define stop and resume conditions
Set conservative stop conditions for the pilot, such as an unresolvable payment ambiguity, repeated ticket failures or a broken staff route. The shift lead should know whom to contact and who can restore service. Resume only after the issue is understood and the relevant test passes again.
Loman reports 93% containment. Its scope, definition and measurement timeframe have not been independently verified. It should not be used as a reason to suppress necessary staff escalation. Use the launch tests to prove the recovery path, then record the result in the plan.
Customer evidence for your rollout
Little Italy kept a person available for reservations, questions and callers asking for human help. Use that deliberate division of work to rehearse a clear, useful fallback.
Results and operating context from Loman’s published customer case studies.
Little Italy
- Over 24%
- reduction in restaurant labor costsReported since adding Loman; comparison period not specified.
- 3 days
- Toast-connected phone setupLittle Italy’s launch experience.
- 1 retained phone person
- questions, reservations and human requestsThe hybrid workflow described after adoption.
Rehearse an explicit request for a person and a reservation question. Little Italy retained one phone person for those needs; confirm who receives the request, what information travels and what the caller is told.
Read the Little Italy case study ↗Watch Little Italy on YouTube ↗
A fallback drill staff can explain
Illustrative drill: a caller asks for a person while the receiving line is busy. Observe the configured next step, the caller’s expectation and where the unresolved request appears. Then verify that the destination cannot forward back into the same AI route.
Use Loman’s plan and transfer descriptions to identify the intended scope. The proof is the configured receiving endpoint and the staff response, not the presence of a transfer feature on a page.