You spot a conversation where Arbyn handed things over to your team, or gave an answer that reads a little unsure, and you want to know why. This is the walkthrough: where to look, what it usually means, and the fastest way to close the gap so the next customer who asks the same thing gets a real answer straight from Arbyn.

When Arbyn genuinely doesn't have grounded information to answer a question, it doesn't guess. It says so and hands the conversation to your team. Every one of those moments is recorded, not lost, so you can go back, see exactly what the customer asked, and decide what to add so it's covered next time.

Knowledge → Knowledge gaps is the fastest path to "what has Arbyn actually struggled with." Individual conversations also show their own reasoning on the conversation's own timeline in your Inbox, useful when you're looking at one specific case rather than the pattern across all of them.

Start with the question itself, not a guess. Open Knowledge → Knowledge gaps. Every row is a real conversation Arbyn escalated because it couldn't find a grounded answer, not a hypothetical or a worst case. You'll see the actual question the customer asked, what Arbyn understood they wanted, and how confident it was right before it gave up and handed the conversation over. Reading the real sentence a real customer typed is almost always more useful than trying to reconstruct what might have gone wrong.

There are really only two reasons. Either nothing in Arbyn's knowledge covers that topic at all, or something exists but it's written in a way that didn't clearly match the question. The Missing knowledge count on that same page tells you which one you're looking at: rows in that bucket are ones where nothing you've added came close, meaning a brand-new document or FAQ entry is what's needed. Rows outside it usually mean something does exist somewhere in your setup, it just wasn't specific or clear enough to win a confident match, which points you toward editing what's already there rather than writing something new.

Check whether it's actually a knowledge problem first. Not every escalation is Arbyn "not knowing" something. A conversation can also be handed to your team because of a guardrail you've set (an action mode that requires your approval), a request that needs a person regardless of what Arbyn knows (managing a subscription on the customer's account, for instance), or simply because the customer asked for a human. The Knowledge gaps page only ever lists the first kind, questions Arbyn genuinely lacked grounded information for, so if a conversation you're curious about doesn't show up there, the reason it escalated is almost certainly something other than a knowledge gap. Check Settings → Guardrails for that conversation's action type before assuming your knowledge is thin.

Fix it where the answer actually lives. If the topic is something you'd normally explain in a written policy (returns, warranty, shipping), add or update it on Knowledge → Documents as a Policy. If it's a specific question-and-answer pair customers ask in their own words, an FAQ entry usually matches better than a policy does, because the question is stored and matched separately from the answer, so a customer's rough paraphrase still finds it. If the honest answer is "it depends on your account," that's usually a sign the topic belongs in your written policy AND that particular customer still needs a person, both can be true at once. Either way, the fix takes effect immediately, no delay, no reprocessing, and the very next matching question can be answered from it.

Your own words always win. If you're wondering whether Arbyn already "knows" something because it's common industry knowledge, general platform behavior is covered by Arbyn's built-in knowledge automatically, no setup needed. But the moment you write your own policy on the same topic, your version is what Arbyn uses. So if a customer got a generic-sounding answer instead of your actual policy, the real fix is usually to write your own version down, not to assume Arbyn already had it and got it wrong.

Confirm the fix actually closes the gap. After you add or edit a document, the original gap row on the Knowledge gaps page stays there until that specific conversation reaches a resolved state, it doesn't disappear just because you added new content. That's by design: the record is honest about what happened at the time, not retroactively rewritten. What matters going forward is the next customer who asks something similar, and you can confirm the fix is working by watching whether the same kind of question stops generating new gaps.

A customer got an answer that felt generic instead of matching our actual policy. What happened? Most likely nothing of yours covers that topic yet, so Arbyn answered from its general, all-stores knowledge instead. Add your own policy or FAQ entry on the same topic and it takes over immediately.

I fixed the gap. Why does the old row still show as open? It records what happened in that specific past conversation honestly. It closes when that conversation itself reaches a resolved state, not automatically the moment you edit a document. What matters is whether the same question stops generating new gaps going forward.

Should I write a Policy, a Custom document, or an FAQ? A Policy is for something you'd put on a dedicated policy page (returns, warranty, shipping). An FAQ is best for a specific question-and-answer pair, since the question and answer are matched separately, which helps with a customer's own phrasing. A Custom document is for anything else. None of the three gets any advantage over the others when Arbyn searches, so pick whichever fits how you naturally organize the content.

Where do I see why one specific conversation escalated, without digging through the whole gaps list? Open that conversation in your Inbox. Its own timeline shows what Arbyn looked up and why it handed things over, useful when you already know which conversation you care about rather than scanning for a pattern.