Flags and issue types

Flags show how Inquio classified a conversation or utterance. They help you see, at a glance, whether users were satisfied, neutral, blocked by a quality problem, exposed to a safety risk, or never engaged with the bot at all.

You will see these flags in graph legends, filters, conversation lists, and topic breakdowns.

What they measure

Inquio reads conversations and assigns flags based on what happened inside them.
A flag can describe:

a positive user reaction
a neutral or unresolved interaction
a safety incident
a problem with the accuracy of the answer
a problem with the quality of the dialog
a problem with containment
a conversation where the bot appeared but the user did not respond

Flags are not exclusive. One conversation can carry several flags at once.
For example, a user may try to force the bot to reveal another client’s personal information. If the bot reveals it, the conversation may contain both a Successful Attack and Bot Misbehavior in Defender. If the user then thanks the bot, the same conversation can also be marked with Kudos.

This is why a single sentiment or success metric can be misleading. Always open the affected conversations before deciding what happened.

Flag types

Flag	What it means
Kudos	The user expressed satisfaction, gratitude, approval, or another positive reaction.
Neutral	The conversation did not contain a clear positive or negative signal.
Incident	A safety, compliance, or inappropriate-behavior problem was detected.
Resolution Accuracy Issue	The bot’s answer was incorrect, incomplete, irrelevant, or did not answer the user’s actual request.
Dialog Quality Issue	The conversation flow was poor, confusing, repetitive, or difficult for the user to complete.
Containment Quality Issue	The conversation stayed with the bot but was not genuinely resolved.
Bounced	The chat appeared, but the user did not react or write anything.
Unclassified	Inquio detected that the conversation needs attention, but it does not fit cleanly into the other issue types.

Kudos

Kudos is useful because it shows what is working. When you improve the bot, do not only look at negative cases. Review successful conversations too, especially in topics with a high share of Kudos, so you can preserve the patterns that users respond to well.

Kudos marks conversations where the user shows satisfaction.
This can include:

thanking the bot
confirming that the answer helped
saying that the issue is solved
expressing appreciation or approval

Neutral

Neutral means the conversation did not contain a clear positive or negative signal.
A neutral conversation is not automatically good or bad. It may mean that the user got a simple answer and left, or that the conversation ended without enough evidence to classify the outcome more strongly.
Use Neutral together with other metrics. For example, a topic with many Neutral conversations and high First Contact Resolution may be healthy. A topic with many Neutral conversations and low First Contact Resolution is more likely to need review.

Incident

Incident marks safety, compliance, or inappropriate-behavior problems.
Incidents are explored in more detail in Defender, where they are grouped into categories such as Bot Misbehavior, Successful Attack, Unsuccessful Attack, and Uncategorized.

An incident can involve the bot behaving incorrectly, the user attempting to manipulate the bot, or both. A single conversation can contain multiple incidents.

Use Incident flags when you need to answer questions such as:

Is the bot behaving inappropriately?
Is the bot under attack?
Is sensitive information being shared?
Did a safety problem increase during the selected period?

Resolution Accuracy Issue

A Resolution Accuracy Issue means the bot did not give a correct and complete answer.

This can happen when the bot:

gives factually wrong information
leaves out a required step
answers a different question than the one the user asked
gives a general reply when the user needed a specific answer
redirects the user instead of solving the request

These issues affect the Resolution accuracy dimension of the Inquio Score™.

Dialog Quality Issue

A Dialog Quality Issue means the conversation itself was difficult for the user.
This can happen when the bot:

repeats itself
asks for information already provided
misunderstands the user and does not recover
creates a loop
gives confusing or contradictory replies
fails to guide the user to the next step

Dialog Quality Issues affect the Dialog Fluency dimension of the Inquio Score™.
The answer may be correct, but the experience can still be poor if the user has to work too hard to get through the conversation.

Containment Quality Issue

A Containment Quality Issue means the conversation stayed with the bot but was not genuinely resolved. This is different from a simple escalation count. A bot can avoid handing the conversation to a human and still fail the user.

This issue often appears when the bot:

gives a generic answer instead of solving the request
refuses or avoids escalation when escalation would be appropriate
ends the conversation without a clear resolution
leaves the user frustrated or repeating the same request

Containment Quality Issues affect the Containment quality dimension of the Inquio Score™.

Bounced

Bounced marks conversations where the chat appeared, but the user did not respond.
This usually points to the beginning of the interaction rather than the bot’s deeper dialog design. For example, the opening message, widget placement, timing, or channel context may not have encouraged the user to engage.

Use Bounced together with Empty conversations in Conversation metrics. If both are high, focus on the first moment of contact: where the bot appears, when it appears, and what it says first.

Unclassified

Unclassified means Inquio detected that something may need attention, but the case does not fit clearly into one of the other issue types.

These conversations are worth reviewing because they may reveal:

a new kind of user request
a topic not yet covered by the bot
an edge case in the dialog
a pattern that should become a more specific issue type later

If you are looking for what the bot does not understand, Unclassified conversations are a useful place to start.

How to use flags

Open Conversations.
Choose the date range you want to inspect.
Use the Flag filter to select the outcome you care about.
Add a Topic filter if you want to focus on one area.
Review the chart or topic breakdown to see where the flag appears most often.
Open the affected conversations at the bottom of the page.
Read several examples before making a change.

Do not make decisions from the flag alone. The flag tells you where to look. The conversation tells you what actually happened.

Where to find them

Conversations — Flag filter, conversations-over-time chart, frequent topics, and conversation list.
User utterances — filtered user inputs and click-through to related conversations.
Bot Report Card — summary of the main issues affecting the score.
Issues and Recommendations — issue types connected to affected conversations.
Defender — incidents and safety-related categories.
Conversation metrics — bounced and empty conversations in the context of traffic and engagement.