Workforce Engagement Management

We're removing the Transcription Confidence Threshold setting from Speech and Text Analytics and setting the threshold to 0 for all organizations.

What this means:
✅ Complete transcripts with every transcribed word preserved
✅ No more ellipses (…) replacing low-confidence transcript segments
✅ Better visibility into every customer interaction
✅ The transcription confidence threshold is no longer configurable

If you've previously configured a custom transcription confidence threshold, there's nothing you need to do. Genesys Cloud will remove the setting automatically, and all transcripts will include every transcribed word.

As transcription accuracy has continued to improve, preserving the full transcript now delivers more value than filtering low-confidence segments. It also helps improve downstream AI capabilities like sentiment analysis, topic detection, interaction summaries, and more.

Hi Liore,

When is this targeted to be released?

Hi Peter, 

This is being rolled out this week, in accordance with our release schedule :)

Excellent thanks!

A few questions:

• What is going to happen in low confidence scenarios? will that be shown in the UI as "low confidence"? It would be important to distinguish low versus high items, either through a color or through a "hover-over".
• Will the confidence scores still show in the raw API results?
• What was the data behind making this decision? Based off what you say, the transcript has gotten so much better, but it would be interesting to understand the specific results that led to this, so that we can show our business "hey look, It's improved X percent"

Appreciate the help!

Pete

------------------------------
Peter Stoltenberg
Engineer
------------------------------

Hi Peter, 

Great questions. 

At the moment, we aren't planning to highlight low-confidence words in the UI.

There are no changes to the API at the current time. Confidence scores will continue to be returned as they are today.

This decision was based on a combination of factors:

• Customer feedback requesting complete transcripts.
• Our decision to already default the threshold to 0 for new organizations (which has already been implemented).
• Ongoing improvements in transcription quality, validated through our internal WER benchmarking, making confidence-based filtering less valuable than preserving the full transcript.

Hope this helps!

Hi Liore,

We've seen scenarios where Zoom verification was completed, but phrases like "I can see you there" or "I'll send that through now" weren't always reflected in the transcript, which impacted topic detection.

Having full transcripts available should help ensure these key steps are captured, and hopefully improve detection accuracy in cases like this.

Hope this will out out soon.