How to Review Top Gaps on the Supervisor Overview

Every AI handover carries a reason, but reading them one at a time doesn’t tell you which gap is actually worth fixing. Top gaps does that grouping for you: it sits on the Supervisor Overview tab, directly below the reason timeline, and rolls up recent AI-handover events into two ranked lists — the things the agent couldn’t do, and the things it couldn’t answer. Instead of scanning a feed of individual handovers, you’re looking at the patterns already sorted by how often they’re happening, with the raw customer quotes one click away.

This is the place to notice that a dozen customers this week all asked for something no capability handles yet, or that the same unanswered question keeps surfacing in three different languages — the kind of pattern worth turning into a new capability or a new KB article.

The two lists: Missing abilities vs Missing answers

Missing abilities covers requests the agent recognized but had no capability to act on — anything from a lookup it couldn’t perform to a change it wasn’t allowed to make. Missing answers covers questions the agent had no knowledge to answer from. Both lists are ranked independently, highest conversation count first, so the top row in each is your biggest opportunity in that category.

The count next to each row is distinct conversations, not raw handover events — if the same conversation triggered a gap more than once, it’s still counted once.

Rows group by meaning, not exact wording

A row isn’t one literal phrase — it’s a cluster of customer requests that mean the same thing, gathered across however the customer happened to phrase it and whatever language they used. The label on the row is a representative phrasing; the count is how many distinct conversations matched that meaning. This is why a single row can be worth investigating even if you never see two customers type the exact same sentence.

Each list shows its top rows first, with a Show more toggle underneath naming how many additional rows are available if the list is longer than what’s shown by default. Only a bounded number of top rows per list carry expandable examples — rows beyond that point are still counted and ranked, but won’t have quotes to expand.

Expanding a row for example quotes

Click a row to expand it. You’ll see a small handful of raw example quotes pulled from real conversations that matched that gap — shown in the customer’s original text direction, so right-to-left languages render correctly. Under each quote is its language, channel, and how long ago it came in, giving you enough context to judge whether this is a one-off or a genuine pattern worth acting on.

The New pill

A row that was first seen within the last 7 days carries a New pill next to its label. Hovering it confirms what it means: first seen in the last 7 days. This is what lets you catch an emerging gap before it’s had time to climb the ranking on volume alone — a request that just started showing up is flagged the same way a long-running high-count gap is, just for a different reason.

Drilling into “See all”

Inside an expanded row, a See all N conversations link opens the deep-dive handover feed already filtered to that one gap group — you don’t need to rebuild the filter yourself. Opening the drill clears any other filter you had active, so what you land on is exactly the conversations behind that row, nothing else layered in. The feed also collapses repeated handovers from the same conversation into a single card, matching what the panel already told you about conversation counts, and it’s capped at 200 rows with a note if the group is larger than that.

The gap filter chip only shows up on the Handovers tab of the deep dive, since that’s the only place a gap-group drill makes sense.

What if gap grouping isn’t turned on for this tenant?

If meaning-based gap grouping is disabled tenant-wide, a gap drill deliberately returns no rows rather than quietly falling back to the full, unfiltered feed under a chip that still claims to be filtered — there’s no equivalent exact-match filter to fall back to, so an empty result is the honest one. If you land on an empty drill and expected conversations, that’s the signal to check whether gap grouping is enabled for the tenant rather than assuming the gap disappeared.

If the section won’t load

Top gaps shows a loading skeleton while it fetches, and if the request fails, it renders an explicit message rather than disappearing — an absent section would otherwise read as “no gaps today,” which isn’t the same thing as “this failed to load.” If you see the error line, reload the page before concluding there’s nothing to review.

Who can see it

Top gaps is available to the supervisor role directly — it doesn’t require a separate analytics-dashboard permission, so a role scoped to supervising handovers can use it without needing broader dashboard access.