AI Topics¶
Function description¶
Daktela AI Topics is a tool designed to extract key information and generate insightful outputs or fill activity statuses based on conversations between you and your customers using prompts that you can define. By leveraging artificial intelligence, it simplifies the process of analysing communication data . Set up your inquiries about activities in the AI functions within the queues grid. After an activity is finished, an AI topics are generated based on the conversation and are store in activity detail.
Info
Note: AI Topics can only process text, so to use it with calls, you need to ensure the calls are recorded and Smart Call Transcriptis enabled.
Licensing and Setup¶
Licensing¶
AI Topics consumes 15 Power Pack tokens per question per activity analysed. Administrators can manage Power Pack licence packages in the Lincensing module under the Add-ons Tab . AI topics are generated for every activity, if enabled on a queue.

Setup and Configuration¶
- Enter AI Functions from the queue grid.
- Enable Use AI Topics checkbox.
- Don't forget to select the AI Language for the generated output.
-
Click the Edit topics button to access the configuration page, where you can design questions about user activities.
-
You can now Add new topics in the right panel, where all your unused AI topics are stored.
-
When you are creating a new topic, choose a questionType , fill in the Title and Description - those will be used in GPT prompt. If the type has Options or Statuses , you can define them bellow. The option descriptions are also used for the GPT prompt.
- After you have finished creating new AI Topics, you can drag and drop them into the form.

- You can then set Conditional Evaluation for each question in the form. If you need a question to be evaluated based on the result of another question, you can set this by clicking on the gear icon .

Explanation of Question Types¶
| Question Types | Description |
|---|---|
| Text | An open-ended question where the answer will be a text-based analysis. This answer will be displayed in the activity details. |
| Selectbox | A closed question where the answer will be a single option from those you define below. This answer will be displayed in the activity details. |
| Multi-selectbox | A closed question where the answer will be one or more of the options you define below. This answer will be displayed in the activity details. |
| Activity status | A closed question where the answer will be a single status from those you define below. The status will be recorded in the activity. |
| Activity multi-status | A closed question where the answer will be one or more statuses from those you define below. The statuses will be recorded in the activity. |
On-demand AI Topics¶
You can manually trigger the generation of AI Topics for a specific activity. This can be done in the activity details using the On-demand AI button. If it’s a call without a generated Smart Call Transcript, the system will generate both the transcript and the AI Topics. Before confirming, you will see how many tokens (and potentially how many minutes for the transcript) will be consumed. The On-demand AI button is enabled only if AI Topics are activated for the queue associated with the activity. This feature is also subject to user accesses permissions. This feature can be useful in two scenarios:
- The activity is from an earlier date when AI Topics were not enabled.
- You have updated your AI Topics setup, but an older activity still uses the previous configuration.

Recommendations for Writing AI Topics Prompts¶
Tip 1: Be clear and specific¶
A clear and unambiguous query helps AI generate accurate and focused results.
- Use concise language and simple sentence structures.
- Break complex requests into smaller, logical parts.
- Don't mix multiple concepts or questions into a single query.
Tip 2: Ask one thing at a time¶
Limiting a request to a single specific goal prevents ambiguity and increases AI accuracy.
- Split multiple requests into separate queries.
- Focus each query on one clear output.
Tip 3: Use English for better comprehension¶
English is the most common language in training data, increasing the likelihood of correct query interpretation.
- Write in standard English with good grammar.
- Use clear vocabulary and avoid idiomatic expressions.
Tip 4: Provide examples¶
Examples help AI better understand what you want and what to avoid.
- Include an example of the desired output.
- Include an example of what to avoid.
Tip 5: Account for speech-to-text errors¶
Transcriptions may contain errors, so flexibility in your query can help capture the correct meaning.
- Provide approximate words or phrases that can be accepted as alternatives.
- Consider phonetic variations or common transcription errors.
Tip 6: Focus on specific parts of the conversation¶
Identifying a specific part of the dialogue helps AI focus on the most relevant content.
- Specify the exact section (e.g., "Look at the first 30 seconds").
- Highlight a specific event or phase (e.g., "After identity verification").
Tip 7: Iterate and refine¶
Prompt quality improves over time as you learn what works and what doesn't.
- Adjust wording based on previous results.
- Track changes and their impact on output quality.
Tip 8: Don't be afraid of A/B testing¶
Experimenting with different prompt versions can reveal the format that delivers the best results.
- Test multiple query structures and compare results.
- Record outcomes and identify the most successful format.
Tip 9: Use contextual details wisely¶
Providing relevant information can help AI better understand the situation.
- Include necessary context (e.g., speaker roles, call purpose).
- Don't overload the prompt with unnecessary details.