A fantastic discussion on the future of search, virtual concierges, and so on.
Sramana Mitra: Let’s start by introducing our audience to yourself as well as to the company.
Grant Ingersoll: I’m the CTO and Co-Founder of Lucidworks. I have a background in search, machine learning, and natural language processing. I have been a long-time contributor to the Apache Solar and Apache Lucene search engine projects, which are both open sourced. Then I also wrote a book called Taming Text. We’re focused on solving key problems for people on the customer experience side.
On the retail side, we focus really on how we can help users or consumers, both pre and post sales, to find and engage with companies and their content. To put that in concrete terms, on the e-commerce side when you go look for a pair of shoes or a hammer, we help you rank that. We help companies get better results to their users while helping users match up their desires.
The other main use case we serve is behind the firewall in the enterprise. It starts off mainly with what you’ve often experienced in your own career, which is, “Where’s all the stuff that we have inside the company? Where’s that document that I wrote?” That’s evolved quite significantly to asking the question, do we enable employees to make better decisions whether those are decisions that need to be made internally or decisions that are designed to help customers get what they need. At our core, we build products on top of this fundamental notion of search and machine learning that are designed to solve those needs.
Sramana Mitra: You sell to enterprises.
Grant Ingersoll: Right.
Sramana Mitra: E-commerce is one of your key use cases.
Grant Ingersoll: That’s exactly right.
Sramana Mitra: Let’s double-click down on that a bit and tell us what’s going on in e-commerce product search.
Grant Ingersoll: The single biggest trend is the ever-increasing demand for better conversions. It’s an on going demand for search and browse activities to be much more personalized, to be much more relevant, and for more flexiblilty in business goals and demands. Those desires translate to how we take advantage of machine learning depending on how you want to categorize it. Despite what you see at the really large companies, a lot of the rest of the retailers in the space are really struggling with modernizing and taking advantage of all of this user data that they have and churn that back into better results.
Sramana Mitra: What is the machine learning angle in this?
Grant Ingersoll: If you look at search from five to 10 years ago, it was primarily keyword-driven. Users type in keywords on engines like Lucene or Solar. It basically says, “What documents do I have that contain those keywords?” It sorts them according to some notion of information theory or some notion of how important those terms are. That’s all well and good, but if often misses out on one of the key things that machine learning brings to the table, which is all the prior user behavior of users who have done that exact search and bought the exact things in that result.
First and foremost, we start with how do we apply machine learning to that ranking or sorting problem. In that domain, this is often called learning to rank whereby you’re taking a look at prior behavior and applying weighting functions to the scores that you get out, and boosting up things that you think are most likely to convert. Then it also goes beyond that to your co-sell and upsell type of recommendation problems. It goes to your understanding of even what question a user is asking for.