Use this article to understand why Fin AI Agent answered, escalated, or ran a procedure the way it did in a conversation, to find out why a conversation was or wasn't assigned to a teammate, and to fix the root cause. Fin Operator, Intercom's AI assistant for teammates, reads full transcripts — including Fin's thoughts, traces, and the content sources it used — then diagnoses what happened and helps you resolve the root cause.
Note: Fin Operator is included as part of the Pro add-on. You need the Pro add-on to use Fin Operator for conversation debugging.
What can you investigate with Fin Operator?
With Fin Operator, you can investigate these questions about a conversation:
Why a specific answer was given — which content sources were used, how Fin reasoned about them, and whether the right information was available.
Why Fin behaved a certain way — trace the decision path that led to a specific response, escalation, or resolution — including which content, guidance rules, and procedures influenced the outcome.
Why a procedure step executed a certain way — see exactly which steps ran, in what order, with what data, and why a step was skipped or took an unexpected branch.
Why a conversation was or wasn't assigned to a teammate — for balanced assignment teams, see what blocked assignment while a conversation waited in the queue, when that changed, and which setting to adjust. Fin Operator can also explain why balanced assignment picked a specific teammate, including their limits, status, skills, and who was next in line.
Tip: Fin Operator can analyze Fin Voice (phone) conversations (including their transcripts, reasoning traces, and outcomes). Monitors also support phone calls, so you can score Fin Voice calls in a Monitor. See How to review phone calls in Monitors.
Note: Fin Operator can explain assignments made by balanced assignment or the "Pull conversation" button. Reasons a conversation waited unassigned are kept for 30 days, and explanations of why a teammate was assigned are kept for 90 days. Learn more in the Balanced assignment deep dive.
What example prompts can you use for conversation debugging?
Use these example prompts with Fin Operator to debug a conversation, a procedure, or an assignment:
"Why did Fin give the wrong answer in this conversation?" (paste the conversation link or ID)
"This conversation was escalated — why?"
"Why did step 3 get skipped in this procedure?"
"Why was this conversation stuck in the queue for 4 hours?" (paste the conversation link or ID)
"Why was this conversation assigned to John?"
How does Fin Operator investigate and fix a conversation issue?
When you ask Fin Operator to debug a conversation, it pulls the data it needs, cross-references your Fin configuration, and proposes fixes in one flow, in these four steps:
Fetches the full conversation — including Fin's reasoning traces, the content sources it referenced, and any tool calls it made.
Looks up your Fin configuration — checks the relevant content, guidance rules, and procedure definitions to understand what Fin was working with.
Identifies the root cause — pinpoints whether the issue is in your content, guidance, procedures, or a combination.
Proposes a fix — depending on the root cause, suggests changes to your knowledge base, procedures, or guidance rules. You review and approve in the same conversation.
When you ask Fin Operator about a procedure issue, it fetches the execution trace showing exactly which steps ran, in what order, and with what data — then cross-references against the procedure definition to pinpoint the root cause.
What tips help you debug conversations with Fin Operator?
Use these tips to get faster, more accurate diagnoses when you debug conversations with Fin Operator:
Share specific conversation links or IDs for targeted debugging. The more specific you are, the faster Fin Operator can diagnose the issue.
Describe the symptom, not the investigation for pattern analysis. "Customers asking about returns are getting escalated" is better than "Look at escalation triggers for return-related conversations." Let Fin Operator determine the right investigation path.
Start with a single conversation if you're unsure what's going wrong. Once Fin Operator identifies a pattern, you can ask it to check whether the same issue appears across more conversations.
Fin Operator verifies claims against actual data. If you say "step 3 was skipped," it checks whether that actually happened before building an explanation.
Need more help? Get support from our Community Forum
Find answers and get help from Intercom Support and Community Experts