Part of Fin Channel Communication Standards
How a 74% missing-introduction rate traced back to two independent root causes — a workflow bug and a guidance gap
A live conversation, whether with a person or an AI agent, usually starts with some version of an introduction: who you're talking to, what they can help with, and what happens if they can't. Fin wasn't doing any of that. The question was how bad the gap actually was, and why it was happening, since a channel that's supposed to feel like a real conversation loses that feeling immediately if it opens with zero acknowledgment of who's on the other end.
The audit combined three passes at different depths: a random sample of 250 conversations from a full month's population of 3,564 chat conversations, scored on opening behavior; a deep manual read of 10 conversations, scored turn by turn against five specific criteria; and a full review of the underlying configuration, every active guidance rule and every procedure, to check whether anything elsewhere in the system was compensating for the gap.
The results were about as unambiguous as a finding gets. 74% of conversations had no opening acknowledgment of any kind. Of the 10 conversations reviewed in full manual depth, all 10 failed all 5 criteria: zero introductions, zero AI disclosures, zero capability framing, zero escalation expectations, generic tone across the board.
The root cause had two separate, independent layers, and both had to be found before either could be fixed. First, the automated workflow responsible for greeting new conversations only actually sends its greeting message when a customer clicks one of its preset reply buttons. The moment a customer types freely into the message box instead, which is what happened in every single one of the 10 manually reviewed conversations, the greeting step gets skipped entirely and Fin's first message is pure content with no introduction. Second, and this is what made the gap total rather than partial, the guidance rule that defines Fin's identity and behavior never actually instructs it to introduce itself. It tells Fin what it is, not to say what it is to the customer. So even when the workflow-level greeting failed to fire, there was no fallback instruction telling Fin to introduce itself anyway.
The audit is complete and the two-layer root cause is confirmed directly against the underlying configuration, not inferred from the conversation data alone. Four concrete recommendations came out of it:
Whether these have since been implemented isn't confirmed as of this audit.
A 100% failure rate sounds like it should have an obvious, single cause, but this one didn't: it took confirming both a workflow bug and a guidance gap independently before either fix would have been sufficient on its own. Fixing only the workflow would leave every future channel or routing path that bypasses it with the same silent gap. Fixing only the guidance would leave the majority of conversations, the ones that go through the broken workflow path, still un-introduced. Finding a compounding failure like this requires checking the actual configuration behind the numbers, not just describing the numbers themselves.