Skip to main content

Debugging

When something didn't behave the way you expected, describe what you saw and the Copilot investigates the actual configuration and history behind it. It works evidence-first — tracing the real wiring and, for live tickets, the real timeline — rather than guessing.

Where a problem can originate​

A wrong or missing value that reaches the customer can come from any layer of the stack, and the Copilot traces through all of them:

  • The custom API returned bad or empty data.
  • The data point mapped the wrong input or has the wrong expected values.
  • The flow is wired wrong — an objective in the wrong order, a transition that fires too early, a condition that never matches.

Knowing this helps you describe the symptom without having to pre-diagnose it — the Copilot figures out which layer is at fault.

Flow behavior​

Why a flow didn't trigger, why a transition didn't happen, or why the agent responded unexpectedly.

"Why didn't the order-id flow ask for the order id?" "The cancellation flow escalated when it should have auto-resolved — why?" "A returning customer didn't get routed to the warranty flow. What matched instead?"

A specific custom API​

An API call or response that's failing or returning empty.

"get_order_status returns empty — how do I fix it?" "The Freshdesk sync isn't creating tickets. What's wrong with the call?"

A live ticket​

Inspect a real support ticket's lifecycle — SLA state, assignment, timing, or agent activity. This reads the ticket's actual history, so name the ticket by its id.

"Why did ticket 123 breach SLA?" "Summarize the agent activity on ticket 123." "Why is ticket 456 still unassigned?"

Cross-cutting failures​

When a problem plausibly spans both a flow and an API, the Copilot investigates both rather than stopping at the first layer.

"The refund flow is broken end-to-end — what's going on?" "Customers report the tracking link is wrong — trace it from the flow to the API."

Describing a problem well​

  • Say what you saw and what you expected — "it escalated, but it should have auto-resolved" is far more useful than "it's broken."
  • Include the identifiers — the ticket id, the flow name, the API name. They let the Copilot go straight to the evidence.
  • Give the customer's message when the question is about triggering or routing — that's what the matching actually turns on.

After you find the cause​

Once you know the root cause, apply the fix in Build & change flows, then reproduce it as a test so the same failure can't come back unnoticed.