I am conducting a vendor evaluation for our new conversational AI project and I am having a lot of trouble with the Genesys Dialog Engine. It seems to fail completely when customers use regional dialects or non-standard phrasing. I have added dozens of utterances but the confidence score still drops below the threshold for basic requests. Is the NLU engine just not mature enough for complex interactions?
In my experience building custom agent desktops, the success of the NLU engine depends heavily on the quality of the training data. You should consider utilizing the ‘Intent Miner’ tool within the platform. This utility analyzes your historical interaction transcripts to identify the actual phrasing used by your customers.
This provides a much more accurate training set than manually inventing utterances based on theoretical requests.
Intent Miner is fine, but it does not solve the core issue of how slow the platform is to train. I run a small shop and I do not have time to spend weeks tuning an NLU engine that ignores half the inputs. If you are struggling this much, you might be better off using the Google Dialogflow integration.
It is much more robust with dialects and you can manage the training in a much more friendly interface. It is an extra cost but it saves so much time.