Sramana Mitra: What I’d like to do is pick a few of your real-world clients who are using AI to do business transformation and start with what problem domain are we looking at, how are you applying AI to their problem domain, and what kind of business transformation are you achieving?
Josh Sutton: Let me break this down first before we jump into client-specific opportunities to set the stage with what are some of the macro use cases that we see being impacted by the broad suite of AI technologies. The three that I see most commonly are along the lines of insight generation, conversational engagement, and business acceleration.
Let me walk through what I mean by each of those and then I’ll walk through some real examples. Insight generation is taking all of the data that we have, from both a structured and unstructured point of view, and using machine learning and natural language technologies to extract insights about your customers and business in a way that enables you to fundamentally drive better business decisions and better results.
One real world example with machine learning is an online retailer that we work with. By applying a suite of machine learning algorithms that we’ve built on top of their entire set of data, we’re able to help them dramatically improve their recommendation engine at the very onset. When customers get to the site, based on what was already known about them, we’re able to fine tune the initial recommendations. Uplift was somewhere north of 10% over a period of three weeks. So, that’s a tremendous bottomline impact out of the gate.
Sramana Mitra: I have a bunch of questions on that. This is a subject that I know a lot about. When you took on this project, what did your evaluation of off-the-shelf recommendation engine products yield? Did you decide to use any of the off-the-shelf products or did you write the software yourselves? If that was the decision, what were the gaps in the existing software?
Josh Sutton: In this particular case, we chose to write the software ourselves, because it was actually much more cost effective than using some of the off-the-shelf products. That’s one of the nice things about your straight machine learning use cases; there’s a lot of open source or extremely inexpensive tools available.
Sramana Mitra: Which ones did you choose to use in this context?
Josh Sutton: I don’t know which one off the top of my head. We tend to do a lot with Microsoft Azure and Tensor Flow. While we play around with others, those tend to be our go-to’s more often than not.
Sramana Mitra: It sounds like you’re happy with what Microsoft Azure is offering in terms of baseline machine learning capabilities such that you can write heuristics on top of that for domain-specific purposes.
Josh Sutton: We’ve been working with them to create a suite of API’s that will live on top of their core platform to make it a little bit easier for developers. I don’t want to go too deep into that because I think we’ll be announcing some stuff in a few weeks. We do like the underlying platform and we partnered with them to do some things that we think are going to make it much easier for companies to leverage that in a way that isn’t going to require machine learning developers, but will enable developers that don’t have as much familiarity to start leveraging the platform.
Sramana Mitra: This is interesting because I met with Mark Russinovich, the CTO of Microsoft Azure, about a year ago. He and I discussed the Azure strategy of creating a programmable platform such that it’s not so difficult to write code on top of Microsoft Azure. It sounds like they have made progress and you have been one of the test sites for this kind of evolution.
Josh Sutton: Yes, we work very closely with them. One of our legacy agencies, Razorfish, was at one time owned by Microsoft.