This article provides more in-depth information on the Autocoding Mode Guardrail.
Introduction
The Auto Coding Threshold controls how confident the codeit AI should be to apply a suggestion.
This guardrail is available at project-level and at task-level. By default, the tasks will use the project default.
Auto Coding Threshold Explanations
Consider the following verbatim:
"They give good service, but it was expensive and the staff were rude"
Suppose also that the Machine Learning generates the following suggestions, in response:
| Suggestion | Text Suggested | Confidence |
|---|---|---|
| Code 3: Good Service | they give good service | 98% |
| Code 7: Poor Price/Value / Too Expensive | but it was expensive | 91% |
| Code 5: Poor service | the staff were rude | 60% |
For each suggested code, the AI calculates a Confidence measure which indicates how confident the model is that the verbatim is about that particular code.
The Auto Coding Threshold is used by the codeit AI as the minimum Confidence percentage needed to apply each suggestion.
In the case above, the codeit AI is very confident about Code 3 and Code 7 but is not very confident about Code 5 as it has a Confidence of only 60%.
So, using the example above:
If the Auto Coding Threshold is below 60%, the suggestion Code 5 will be applied by the AI.
If the Auto Coding Threshold is above 60%, the suggestion Code 5 will not be aplied by the AI.
We recommend using a Auto Coding Threshold of at least 50%. Lowering the Auto Coding Threshold will result in a higher volume autocoded but with a lower quality.