I think we all spend time using Use Cases to help train support or instructional bots—because that is obviously needed; however, Use Cases in a Sales bot system, needs to be considered as well. Recently, we experienced some frustration with the performance of a Sales based Artificial Intelligence (AI) Chatbot. After some trial and error, we hypothesized that during the training of the system, through the Convolutional Neural Network (CNN), and the continuing learning cycles, the Large Language Model used did not have sufficient, non-typical Customer Use Cases introduced. Therefore, customers were led most of the time to the more expensive product. Which in this case was cost prohibitive.
Just a little advice for us all. Sometimes when we are designing these awesome toolsets, we need to think more about the end customer that may not be typical. Non typical use cases and prompts can help the bot discern the customer’s intentions. In the case with my annoying interactions with this “Salesbot”, it became obvious that this Chatbot system focused and biased on selling the product with the most features rather than the necessary feature set for the customer.
In my case, there were multiple licensing tiers and strategies. With each SKU, came a very defined set of features and functions. However, this CNN skewed to the higher, feature rich SKUs. Most likely a majority of orders would be with these licenses. These are licenses that have more comprehensive features and functions. However, there are many customers that have specific needs and there exist SKUs to accommodate their respective needs. The SKUs we were looking for are specialized subset of features and functions of the large software licenses. The cost delta between the full licenses are 3.8 times the cost of the subset licenses.
Here’s the rub. There are a few cybersecurity related features that are needed for cybersecurity and compliance. Because the full licenses SKUs had a full set of features, the bot kept defaulting to the more expensive licensed product, however the client needed less than half of the feature set while needing the security and compliance elements.
To overcome this issue, I had to describe, in detail, how the feature set would be used and to articulate that the other features are not necessary. This forced the bot to make an exception and send the sales quote request to an outside “real” sales representative, to solve the problem. Which worked. We are all making great progress, but throughout the history of technology, the road has been fraught with curves, potholes, and danger.
Technology progress can be annoying or entertaining…especially entertaining to us “Techies.” Thanks for listening to my diatribe.