6. How to Analyze Interviews with AI
Learn how to use ValueFlow's AI analysis features to automatically analyze each interview transcript.
Preview
Aggregated results from multiple interviews
Under "Aggregated analysis results" you can see the aggregated analysis of all interviews from the same interview agent. This tutorial will show how to configure custom analysis criteria/items.
Individual result
This is the analysis view for a single interview. It shows the same items but only based on one interview and not on an aggregated level.
Creating analysis items
Where to find the configuration of the analysis items: in the configurations tab on the very bottom, under the title "Analysis items".
Different data types to choose from
Each analysis item is simply prompted. You describe your desired analysis in your own words. You can choose the type of output that you desire:
Boolean: The AI will try to evaluate your prompt and output yes or no
Number: The AI will try to evaluate your prompt and output a single number (in a range of your choice)
Categories: The AI will try to evaluate your prompt and output one of your provided categories
Text item: The AI will try to evaluate your prompt and answer in a short (one-line) text answer
Long text: The AI will try to evaluate your prompt and will answer in more detail
Example 1
In this example, an employee interview is analyzed to see whether the interviewee mentions their relationship to their boss. It is a long text field, as we do not have an idea how employees will describe their relationships.
Example 2
For the same employee interview, we have typical areas in mind that employees typically desire improvement for. We choose the category field option and offer the most expected categories as options. The AI tries to check if these options are mentioned in the interview.
Example 3
Here we simply like to learn whether it's an employee from a technical department (engineering or IT) or rather from management areas. We do this because we have the assumption that the needs are different for technical and non-technical employees.
Example 4
Again an example where we offer a scale by choosing categories.
Analysis workflow
Individual result
This is an individual result, where the above items have been analyzed.
Running analysis on existing interviews
The folliwing image shows the situation when the interview is already there but the items are not yet evaluated. It is possible to add analysis items before an interview is conducted or afterwards. If analysis items exist already before an interview is conducted, the analysis is triggered automatically. If we want to add further analyses after interviews have already been collected, then we have to trigger the analysis manually. We do so by clicking on the symbol in the right corner (the magnifier/repeat icon). We can run all analysis items afterwards or only single items.
Variability in AI analysis results
As it is possible that an analysis item prompts the AI to conclude something, it is possible to get slightly different results if an analysis is rerun.
Example prompt in an analysis item: "Based on the interview transcript, how likely would you say is the employee to quit in the next 12 months? Come up with a value from 0 to 1, where 1 means that the employee will leave with 100% certainty."
This is a highly subjective/uncertain prompt, such that when this analysis item would be run multiple times, it would create e.g. 0.6, 0.65, 0.7, 0.68, 0.5 as results.
AI-generated explanations
The AI also comments its own results to make them self-explanatory.
Conclusion
If you have reached this part then you are able to successfully create, run and analyze ValueFlow's interview agents!