General Assembly hosts Michael Housman, Chief Data Scientist of RapportBoost.AI, as he speaks about deploying AI and why it’s not a ‘set it and forget it’ endeavor.
“What you need to do is not just deploy an AI model, say my job here is done, go forth and prosper. We need to continually refine those models. Once we’ve deployed a model – once we’ve used it in production and then observed outcomes – the key thing to do is this. Think of a 2×2 table, where an AI predicted either a yes or a no, for the person either opened the email or didn’t, for example. And then, let’s look at the actual outcome, which was the person actually opened it, or they didn’t open it. So in that 2×2 matrix you have (1) we said they wouldn’t open it and they didn’t; (2) we said they would open it and they did. Those are what we got right. But now there are two boxes where we were wrong: (3) we didn’t think they’d open it and they did; (4) we did think they’d open it and they didn’t. Those are called false positives and false negatives. And what we need to do is dig in. We need to figure out which of those is most concerning – and it really depends on the subject matter. We need to dig in and understand, well, why were we off?”
At RapportBoost.AI, we help companies take measured steps toward integrating AI into their customer support strategy. We analyze customer support centers and brand’s live chat conversation data to generate recommendations about how to chat with customers. After just three months of implementation, our customers see significant lifts in key performance indicators. Contact us to turn your live chat channel into sales.Follow us: