Workforce Engagement Management

I wonder if anyone has any experience in AI scoring who can please offer advice on this? We have the ability to 'train' our transcription to identify words in transcription and improve how it transcribes. Can anyone tell me if this is picked up in AI scoring?

In a very simple example, if a local dialect uses a slang word for 'hello', and we ask AI to confirm if the agent has been greeted, would it understand its meaning based on us improving the transcription? I suspect the answer is no, and AI runs independently, but it would be good if someone is able to confirm for me please? 

I am currently thinking the best way to include this would be to use Evaluation Assist and add the 'new' word to a topic, rather than use AI, but all input is welcome!

Thanks, Heather

#AIScoring(VirtualSupervisor)
#QualityEvaluations
#SpeechandTextAnalytics

------------------------------
Heather Henderson
------------------------------

Hi Heather,

Great question – and your understanding is essentially correct.

AI Scoring operates on the finalized transcript, not in parallel with the transcription process itself. In other words, the flow is:

1. Audio → Transcription (ASR)
   This is where Dictionary Management comes into play. Custom words, slang, or local dialect terms help the ASR engine recognize and correctly transcribe what was actually said.

2. Transcript → AI Scoring evaluation
   AI Scoring then analyzes the text output of the transcription, looking for patterns, intents, phrases, and semantic meaning based on what is present in the transcript.

So Dictionary Management and AI Scoring act at different stages and for different purposes:

• Dictionary Management improves how accurately speech is converted to text.

• AI Scoring evaluates what is written in that text.

If the slang word for "hello" is correctly transcribed thanks to dictionary tuning, AI Scoring can pick it up only if the model is able to semantically associate that term with a greeting. However, adding a word to the dictionary does not teach AI Scoring its meaning by itself - it only ensures the word appears correctly in the transcript.

Because of that, your thinking makes sense:

• For very specific or localized expressions, Evaluation Assist with topic configuration (or explicit phrasing/logic) is often more deterministic and reliable.

• AI Scoring works best when the language patterns are more broadly recognizable or when enough contextual signals exist in the transcript.

In practice, many teams end up using:

• Dictionary Management to improve transcription accuracy

• AI Scoring for broader behavioral or semantic checks

• Evaluation Assist / topics for very domain- or dialect-specific validations

Hope this helps clarify the separation and how they work together.

------------------------------
Mateus Nunes
Tech Leader Of CX at Solve4ME
Brazil
------------------------------