Feedback for AutoSummary

​

Feature Release

This is the announcement for an upcoming ASAPP feature. Your ASAPP account team will provide a target release date and can direct you to more detailed information as needed.

​

Overview

AutoSummary now supports model retraining using agent feedback. The feedback endpoint receives free-text paragraph summaries submitted by agents, and uses the difference between the automatically generated summary and the final submission to improve the model over time.

​

Use and Impact

In agent workflows that expose an automatically generated summary for final review, agents may make edits to incorporate uncaptured nuances or correct details where needed. Rather than losing this valuable edit data, AutoSummary can now incorporate it into a feedback loop to improve the quality of the model.

The feedback endpoint can also be used to accept edited summaries that are reviewed in quality assurance workflows.

Standardizing the format and mechanism for receiving feedback facilitates ongoing model retraining that leads to improvements in…

1. Error reduction
2. Coverage of key topics

​

How It Works

A summary identifier field is being added to free-text summary responses. This new field should be used in requests to the feedback endpoint.

Free-Text Summary ID

The GET endpoint for free-text summaries will be updated with an additional response field called summaryId . This field must be included in requests to the feedback endpoint to identify the summary for which you are providing feedback.

Feedback Request Format

The request body for the new feedback POST endpoint requires the following:

• Conversation identifier
• Summary identifier (as mentioned above)
• Final free-text summary submitted by the agent

​

FAQs

1. Is it mandatory or optional to use the AutoSummary feedback endpoint? For implementations of AutoSummary that give agents the ability to edit summaries, ASAPP strongly recommends using the feedback endpoint to improve summaries over time. Anecdotal feedback from the field is considerably less useful for making model improvements. For implementations of AutoSummary that do not give agents the ability to edit summaries, there is no need to use the feedback endpoint.
2. Will the feedback endpoint accept free-text summaries that are not edited? Yes. For simplicity, ASAPP recommends sending all summaries that agents or other end-users have reviewed and submitted, even if the reviewed summaries are identical to the automatically generated summaries. AutoSummary will detect edits and use them for model training.
3. How does automatic model retraining work? AutoSummary continuously collects feedback from each conversation, identifying which submitted summaries contain edits from the initially generated summary. These edits represent signals that AutoSummary models can use to adjust outputted text to better represent key events in the conversation.