What are AI Filters?
AI Filters allow you to create a question that will be associated with each feedback. An example question could be something like:
Does this feedback describe a software bug
What is the client emotion in this feedback
Is the customer voicing this feedback likely to churn
Our AI will then take your question, and for each individual feedback, attempt to answer that question. So if you have 100 comments, AI Filters would be able to tell you how many it believes describe a software bug, or how many people are going to churn.
Unlike traditional filters (such as background variables like "region" or "product"), AI Filters are not part of the original dataset. They are generated by AI based on open-text comments or full support conversations. This means they allow you to structure information that was previously unstructured.
While we have over 50 questions already developed — ranging from things like:
“Is this feedback mentioning a competitor?”
“Does this feedback complain about one of our premier products?”
“Is this customer at risk of churn?”
Netigate Insights also allows you to create your own question. This question can be tailor-made to fit your specific data and focus on the metrics and signals that matter most to your organization.
For example:
Which customer journey step does this feedback relate to? (Onboarding / Checkout / Support / Returns / Other)
Is this feedback indicating a critical purchase funnel failure?
Did the agent offer an upsell during this interaction?
Does this feedback signal a trust breakdown (e.g., fraud concern, security fear)?
How can I start using AI Filters?
Currently, AI Filters can only be configured by Netigate Insights. This is because creating high-quality AI Filters requires:
Clear definition of the question
Well-defined answer options
Proper context for your industry
Testing and refinement to ensure accuracy
AI Filters often go through iteration to improve precision, especially when used for advanced use cases such as churn risk, escalation detection, or customer effort scoring.
We recommend that you contact your CS Manager or email help@lumoa.me if you want to get started using AI Filters. That way we can:
Understand your specific use case
Align the AI Filter to your business context
Ensure it works correctly with your data
Help refine it if needed
How can I view AI Filters, after they have been created?
AI Filters will be created as a tag associated with your data. This means that, like all other tags, you could use the filters menu to filter to all responses that report a bug, as seen below:
Conversely, you could also create a custom graph widget that would look at this information, and update itself over time. Below you can see custom graphs made for things like:
client emotion throughout the feedback,
whether this customer will churn,
is this customer reporting a bug,
whether this feedback contains a feature request:
In support contact data specifically, AI Filters can analyze the entire conversation, not just summaries. This enables powerful use cases such as:
Did the agent verify the customer’s contact details?
Did the agent clearly summarize next steps?
Was the customer expressing anger or fear at the end of the call?
Did the customer request contract termination?
These can then be used to build quality dashboards or automated escalation workflows.
What kinds of questions can I ask for AI Filters?
Theoretically, you can ask any question you want. However, AI Filters will only be able to look at your customer open text comments when generating a response. This means, even if you create a question around "pricing", the AI might not be able to answer it if there are no comments in your data talking about "pricing".
Additionally, the questions that you ask are defined by generative AI. Meaning making sure that you ask the right question, and formatted in the right way, will lead to better results. While our AI experts can help you format your question for maximum success, its important to think about why you are asking this question - what is the goal behind question? What information are you looking to obtain? Would you consider this question successful if it gave you Answer A, or Answer B?
One way to understand how generative AI works is to imagine an untrained person (or even a teenager/child) who is unfamiliar with your business reading through the conversation and evaluating the response based solely on the answer options provided.
Lets take an example and assume we want to see all feedback talking about a "Price Estimate: Does the customer service agent provide a price estimate during the interaction? (Yes/No/Unclear)"
To improve the accuracy of responses to this prompt, consider the following refinements:
1. Be More Specific About "Price Estimate"
Issue: The term "price estimate" might be interpreted differently—does it mean an exact quote, a rough range, or mentioning pricing at all?
Improvement: Clarify what qualifies as a price estimate.
Example Revision:
"Does the customer service agent provide a specific price or a price range for the requested product or service during the interaction? (Yes/No/Unclear)"
2. Define "During the Interaction"
Issue: What if the agent promises to send a quote later? Or refers the customer to a website?
Improvement: Specify if the price must be stated directly in the conversation or if referring to another source counts.
Example Revision:
"Does the customer service agent state a price or price range verbally during the interaction, without referring the customer to another source? (Yes/No/Unclear)"
3. Consider Adding Contextual Conditions
If the agent is unable to provide a price due to company policy or missing information, should the response still be "No"?
Example Revision for More Context:
"Does the customer service agent provide a price estimate when requested by the customer? (Yes/No/Unclear)"
Final tips
Start with a small number of AI Filters focused on clear business goals.
Test and refine the wording if results seem inconsistent.
Avoid overly broad conceptual questions (e.g., “How could the agent improve long term performance?”).
Use AI Filters to complement topic modeling — not replace it. Topic models help you understand what the feedback is about overall, while AI Filters help you zoom in on specific signals or risk indicators.
If you are unsure how to phrase your AI Filter or want help designing one for your industry, contact your CS Manager or email help@lumoa.me — we are happy to help you build the right setup for your data and use case.
Get in touch
📧 Do you have any questions or comments about using Netigate Insights? Please don't hesitate to email Netigate Insights Support at help@lumoa.me.