Configurable triggers for Knowledge Base Answer Suggestions

1. What is your original issue/pain point you want to solve?

Currently, Knowledge Base Answer Suggestions seem to be generated automatically whenever a ticket is opened.

This can result in unnecessary API requests and related costs, especially when a user only opens a ticket briefly or when a Knowledge Base suggestion is not required at all.

We would like to have more control over when Knowledge Base Answer Suggestions are actually requested.

2. Which are one or two concrete situations where this problem hurts the most?

  • A user opens a ticket only briefly to check some information. Even though no answer is required, a Knowledge Base Answer Suggestion is generated.
  • A user opens an already closed ticket. In this case, an answer suggestion is usually no longer needed, but an API request may still be triggered.

In both situations, AI resources are used without providing a real benefit to the user.

3. Why is it not solvable with the Zammad standard?

Currently, there seems to be no option to define conditions for when Knowledge Base Answer Suggestions should be generated or to trigger them manually.

As a result, we cannot prevent unnecessary requests based on criteria such as the ticket state or the user’s actual need for an answer suggestion.

4. What is your expectation/what do you want to achieve?

We would like to have configurable conditions and trigger options for Knowledge Base Answer Suggestions, similar to the conditions available for ticket summaries.

Ideally, administrators should be able to configure one of the following behaviors:

  • Automatic: Generate suggestions automatically when defined conditions are met.
  • Automatic with delay: Generate suggestions only after the user has viewed the ticket for a configurable number of seconds and the defined conditions are still met.
  • Manual: Generate suggestions only when the user actively clicks a button to request them.

For example, it should be possible to define a condition such as “Ticket state is not Closed” to prevent suggestions from being generated for closed tickets.

This would make the feature more targeted, reduce unnecessary API requests and costs, and give administrators better control over AI usage.

The answers are suggested by a vector search on elasticsearch. An API-call is only generated for creating the embeddings, if they haven’t been cached. Those run on super-cheap models, usually.