Sample report. Built from the public MultiWOZ 2.2 run of 7 November 2026, not from a customer. A real audit reads your agent’s last week of conversations.

Unknown Ops · audit report

What your agent couldn’t answer

Conversations read
10,437
Requests not met
5,153
Substantive gaps
24
Coverage
49.5%
Guarantee check: met. The bar is five distinct questions asked three or more times. 24 of the 24 substantive gaps clear it.

1 · Fix with one reply ENTRY

Each of these is a fact the owner or a shift manager can supply. Each one becomes a single email: the question, how often it was asked, and a reply box.

#Question customers couldn’t get answeredTimesDomain
1Entertainment venue in the centre106attraction
2Multiple-sports venue outside the east102attraction
3Entrance fee / admission price not in the data74attraction
4College outside the centre / west38attraction
5Theatre / cinema in an area with none36attraction
6African restaurant in the wrong area / price34restaurant
7Swimming pool in an area with none33attraction
8Concert hall outside the centre27attraction
9Architecture outside the centre26attraction
10Nightclub outside the centre / south25attraction
11Vietnamese restaurant in the wrong area / price14restaurant
12Lebanese restaurant in the wrong area / price12restaurant
13Japanese restaurant in the wrong area / price11restaurant
14Turkish restaurant in the wrong area / price11restaurant
15Thai restaurant in the wrong area / price10restaurant
16Boating in an area with none10attraction
17Park in the west10attraction
18Korean restaurant in the wrong area / price9restaurant
19Seafood restaurant in the wrong area / price6restaurant

The email this produces · example for gap “Entrance fee / admission price not in the data”

Asked 74 times, never answered

“What is the entrance fee?” Reply with the answer and we’ll publish it to your agent.

↩ Reply

2 · Needs a tool TOOL

No knowledge entry can fix these. Each needs a live lookup or an action, such as checking availability. They are reported as integration work, never sent to a shift manager as a question.

#What customers couldn’t getTimesDomain
1Booking slot unavailable (room or table)1,065hotel, restaurant
2Hotel search: no match for the requested combination711hotel
3Train: no service at the requested time / route97train
4Restaurant search: common cuisine, no match for price / area78restaurant
5Buy / reserve attraction tickets5attraction, train

3 · Left out, with the reason

38 “No [cuisine] restaurant” rows, 745 occurrences: a dataset artifact from an ontology that lists cuisines with no venues. Excluded from the 24, and listed in full in gap_table.md.

Domains are the dataset’s (Cambridge, UK, booking dialogue), not target customers. What this run does not show is stated on the method page.