We explore AI Connectivity in this discussion.
Sramana Mitra: Let’s start by having the two of you introduce yourselves as well as Pandio.
Gideon Rubin: I’m the CEO and Co-Founder along with Josh Odmark. I’m a serial entrepreneur. I focus primarily on using big data and data science to gain an advantage. Pandio came from the fact that Josh and I both came from different startups. We were talking about what to do next. We realized that IoT and a lot of the data sources were getting really good.
On the other side, there are a lot of great models out there to get value out of that data, but the in between was very costly and very custom and extremely time-consuming. At the end of the day, you didn’t even know if that infrastructure would support what your in-project was; because when you started, you didn’t always know how many data sources you’d need, what type, or how much data.
We built Pandio to help companies get to AI faster. They wouldn’t have to think about the infrastructure that needs to be built just to implement machine learning models for large enterprises. That was where it came from. If you think about where Josh and I were before this, I was working with a European company helping them launch in the US.
We were providing mapping and navigation data to the major mapping and navigation companies – Waze, Apple Maps, and Google Maps. It was very time-consuming and complex. That showed me the need and the pain really well before we started this.
Josh Odmark: I’m the CTO and Co-Founder at Pandio. My background is in software engineering. I spent the last decade working in the machine learning and AI space trying to build those initiatives from the ground up. This is in the insurance industry with actuaries and analysts. In the process of trying to coordinate all that and have those teams work together and produce something of value, I saw a whole lot of inefficiencies and all the difficulties in doing that. Pandio was created to address those pain points and accelerate the adoption of AI in general.
Sramana Mitra: I understand the general problem that you are pointing to. Let’s get down to specifics on your early customers that you went in to solve the problem for. What kinds of systems did you encounter and what kind of environments did you encounter?
Gideon Rubin: I don’t know if I mentioned this earlier, but I do have a Master’s in Management Science focusing on Applied Statistics. To answer your question, one of our first customers was a large-scale media company. The challenge they had was huge amounts of data from many sources. They have hundreds of thousands of small business customers for whom they manage marketing campaigns.
They manage these campaigns across 30 to 50 channels. It might be paid marketing on Facebook, or Google paid search. They have dozens of opportunities for these small businesses. They have hundreds of thousands of small businesses. Some of these campaigns were as little as a few hundred dollars a month. There’s no way to have humans manage that many channels across that many small businesses and act effectively.
If they just did heuristics or managed in a spreadsheet, it doesn’t scale well. You have some companies that are not getting good results in Tulsa, Oklahoma while someone in Atlanta, Georgia might be getting great results. What they wanted to do was empower their data scientists. They had about six data scientists and 140 people on the overall product team when we started.
We gave them access to all the channels of data. Then we allowed them to map all that data together. They might have a customer number in one department and a different customer number in a different product set. For the businesses they acquire, they merge over seven times over the last 20 years. They have a lot of legacy technology and platforms in there.
Then we allow them to automate the processing, in other words, ingest the data. They give an estimate when a small business comes to the website. They get an estimate that’s based on millions of small businesses doing marketing campaigns across all these channels. That was the first output of the model.