Thought Leaders in Artificial Intelligence: Grant Ingersoll, CTO of Lucidworks (Part 3)

Sramana Mitra: The nitty gritty of this is very interesting to me. To the audience, it would be very interesting to know to what extent did you achieve this personalization. What I told you earlier about my own experience, where we were doing clothing search, you have many cluster-based personalization opportunities.

For example, size is one of them. Then there is also color, different hair colors, eye colors that match better or worse with different types of clothing. We had all of that in our taxonomy and in our rules. This particular area is very rich for personalization and search. Does this all fall within your purview?

Grant Ingersoll: It definitely does. I would say that big change from what you’re describing is a large majority of that is automated and learn from behavior. Of course, there’s a bootstrapping problem in there where you, as a company, providing other stuff needs to, at least, have an opinion. The reality is these days, a lot of that can be derived from a feedback loop whereby you’re looking at user behavior and it’s automatically factored in.

There will always be a need for a rule engine in the sense that company business goals will always have an editorial component where you’re going to say, “I want these jeans to be at the top of the results list because we’re running a sale on them this week.” Regardless of the core feedback loops and ranking algorithms, you still want to be able to come in and say, “Put this at the top.” What you don’t want, though, is your merchandisers to spend all day doing that kind of stuff.

Automation sounds like you’re getting rid of jobs. What we find is that it is actually much more effective. It’s helping people to set up experiments. Instead of thinking about, “I know that this result should be at the top.” They instead say, “Let me take and run an experiment that determines that.” I’m sure many of the listeners here think, “That’s what Google and Amazon do all day long.”

Actually a lot of companies struggle with this, especially on the backend. They might do several hundred A/B test a year as a whole on their site but most of those A/B tests are focused on the user interface. Very few of them are focused on the data. One of the things we really try to do and help our customers with is become much more data-driven and have that experiment mindset that says, “How do I go from a model where I have a right to test whether the model is right or not?”

That’s a big challenge because once you come into that, you say to yourself, “How do I do 10x the number of experiments a year?” Some of the big giants in this space, people don’t realize that they might run 10,000 to 20,000 experiments a year. Most people can’t even fathom that in their current setup because it’s so antiquated.

This segment is part 3 in the series : Thought Leaders in Artificial Intelligence: Grant Ingersoll, CTO of Lucidworks
1 2 3 4 5