The layer down from that is the data layer. Instead of talking about the fact that I have employees here and there, I want to look at specific data elements and really understand. Based on the cardinality of the data and the pattern distributions, I want to build very specific rules on how to construct business processes to reconcile that data. It could be automatically corrected so this is building autonomic behavior with respect to data. This could be building data stewardship processes so that certain things are like a matter of taste.
This is jumping out a little bit. If company A buys company B, I want to look at sales by region. Company A has three regions. Company B has four regions. Which way do you want to look at it? Is it an AI problem or is that something that, potentially, AI needs to punch out and say, “I can do this for you, but you need to tell me what model this should actually conform to.” That’s done at the data level.
This is getting into the specifics of the data and getting specific exception processing and data manipulation on the data that you’re seeing. That’s all done by training the system on what’s actually happened before. It lends itself very well to a supervised training model. The supervision can actually be direct with business analysts. It can work a little bit unsupervised as well based on these other examples that we see where there’s prior work. It will generally be a hybrid, but we tend towards supervised.
Sramana Mitra: If you were starting a new company, what would you start?
James Markarian: Having been around just integration and data, the company that I imagine starting ended up being SnapLogic because we start from scratch when the foundation is so solid. That being said, there’s still a lot that we can do with AI around autonomic behavior of our compute cluster. That continues to be an interesting problem. Exceptions on the compute cluster level aren’t just infrastructure exceptions. They can be customer exceptions as well.
There’s a whole lot of things we can do on the integration side innovating in various ways including with AI. In the whole world of analytics, the big thing that nobody’s bitten off yet is for a long time, we talked a lot about closed-loop analytics. You would expose certain facts about the data and then generally speaking, be incumbent on an analyst or somebody to take that insight and turn it into action. The big area which is too big to get your head around is, “Is there an application of AI to close that loop?”
To actually take the output of analytics and instead of presenting it to a user, actually taking the action. It’s slightly mind-bending to think about what that might actually look like. I think that if we really want to take the state-of-the-art analytics home to the promised land, getting non-obvious insights into data is good but doing something with it is the holy grail. As we progress through this technology, we’re going to see investment and innovation there.
Sramana Mitra: Thank you for your time.