Skip to main content

Documentation Index

Fetch the complete documentation index at: https://docs.enneo.ai/llms.txt

Use this file to discover all available pages before exploring further.

AI post-processing is suitable for structured evaluations of completed tickets. It evaluates processed tickets based on configured, typed questions. The generated results are stored in the system as AI Insights and can be used in the ticket and in exports.

​
What is AI post-processing?

AI post-processing is a downstream classification of tickets. After a ticket is completed, AI answers previously defined questions based on the ticket conversation and available metadata. Each question has a professional name, description, and data type. This prevents free, unstructured summaries and instead creates evaluable values such as yes-no answers, categories, numbers, or short texts. The configuration is done in the channel-specific settings under the point AI post-processing.

​
What is the AI post-processing used for?

The function is particularly useful when information is technically relevant but cannot reliably be represented by status, tags, or fixed process fields. This primarily affects assessments that only arise from the course or result of processing. Examples include:
  • whether the concern was actually fully resolved
  • why a process was handed over to an employee
  • whether automation has failed due to missing data, missing approval, or technical limitations
  • whether the original customer intention has changed during the conversation
  • whether a process was formally closed, although clarification was still needed
Such information is relevant for evaluations but is often not stable to model over operational fields. The AI post-processing supplement this structure subsequently, without changing the editing process itself.

​
What are AI Insights?

AI Insights are the stored results of AI post-processing. They are stored on the ticket and are also available in the data export. The export of AI Insights can be found under Advanced Settings → Data Exports. In data export, the AI Insights are output as structured information per ticket. These include:
  • the answered question
  • the typed value
  • the confidence
The value depends on the data type of the question. For a Boolean question, the result is, for example, true or false. For an Enum question, the result is one of the defined categories. The confidence describes how confident the AI is in its answer based on the available information. A low confidence does not necessarily mean that the result is wrong. It indicates that the evidence in the ticket can be weak, indirect, or contradictory. The brief justification is saved in the ticket context but is not output as a bulk entry in the data export, as it may contain free text and can quote conversation contents.

​
Importance of question types

The type of question determines what kind of result the AI can deliver and how well this result can be evaluated later. Boolean questions
  • For clear yes / no classifications
  • Suitable if it should be clearly decided whether something applies
  • Example: “Has the concern been resolved?”
Enum questions
  • For predefined categories
  • Especially suitable for reporting, filtering, aggregation, and comparisons
  • Example: “What is the reason?”
Text questions
  • For brief qualitative classifications or summaries
  • Less stable for reporting, as free text is harder to compare
Number fields
  • Only use if the number can be clearly derived from the conversation or the metadata
  • Not suitable if the AI would have to estimate or interpret

​
Best practices for questions

Questions should be technically unambiguous, narrowly limited, and answerable from the ticket conversation. The description is crucial in this context: It controls which evidence the AI should consider and when no reliable result is available. A good question not only describes the desired result but also the demarcation to similar cases. Example: Name: Processing result
Type: Enums
Options: completely_resolved, partially resolved, unresolved, unclear
Description: Evaluate whether the customer's request in the ticket was finally resolved. Use "unclear" if no reliable statement can be made from the conversation.
Questions that examine several aspects simultaneously or assume information not contained in the ticket are less suitable.

​
Dealing with empty results

An AI Insight may remain empty if the AI does not find a sufficient basis for an answer. This is technically desired and should be considered in evaluations. If an indeterminate result is analytically relevant, it should be explicitly modeled. An unclear or unquantifiable category can be used for Enum questions.

​
Implications of changes

Changes to questions affect future evaluations. Historical results remain preserved. If a question is deleted, it will no longer be used for new tickets. However, previously generated AI Insights continue to be available for audit and export purposes. When changing the question, description, data type, or options, it should be taken into account that old and new results may not always be technically directly comparable. For stable time series, questions, and option values should be as consistent as possible.

​
Recommendations for productive use

AI post-processing should start with a few, clearly defined questions. A small number of precise questions usually provide more stable results than many broad or overlapping questions. Before using in reports, new questions should be checked using real tickets. It is particularly important whether the answer can actually be derived from the conversation and whether the categories are sharply differentiated enough. AI post-processing makes sense if completed tickets need to be evaluated in a structured way. It should not be used as a substitute for binding process logic, compliance decisions, or manual technical reviews.