Influencing Intent learning in Smart Assist
We recently activated Intelligent Triage + Smart Assist EAP. Based on the documentation, I understand changing the Intent on a ticket does not train the machine learning model, but I'm wondering if there is any other way to influence the way it's categorizing tickets.
For example, one customer wrote "How much does it cost to ship to Australia?" — Smart Assist labeled this as "Price is different than expected," when it should be categorized as a request for shipping costs. I want to train the AI that when customers mention Australia, Canada, Hawaii, etc., alongside "quote" or "cost," the intent is more likely about shipping rates, rather than product cost.
Is Smart Assist training something ZenDesk is still developing, or are there additional configuration settings I'm not seeing?
Hi Harper Dane!
Thank you for the feedback! The intelligent triage models are continually evolving so while you cannot provide feedback to/train the model at this time, we continue to strive to make it more accurate through future updates. There are early ideas to allow for training and feedback to influence the model(s), but that is a longer term target without a definitive ETA.
For the exampled you provided, what was the Intent confidence level on the ticket where the intent was "Price is different than expected" rather than about shipping costs?
Hey there Jake Bantz — in that particular example, I believe the intent confidence was "low."
However, I've seen other examples of customers requesting shipping costs, and the ticket being categorized as a request for quote on product (I can't recall the exact name, but that's essentially the intent).
Those are much more common, and the intent confidence is higher on them. It seems as if when the customer mentions cost or price—no matter if they're also including a shipping address, saying "international," or mentioning "shipping" or "freight"—the current system heavily favors product-related categorization over shipping.
Hope this helps!
Super helpful feedback. Thanks for bringing it up!
Bitte melden Sie sich an, um einen Kommentar zu hinterlassen.