Bootstrapping with Services at the Cusp of AI and FinTech: Vahe Andonians, Founder of Cognaize (Part 3)

Sramana Mitra: Now we’re in 2017?

Vahe Andonians: 2018.

Sramana Mitra: What problem were you going to solve next?

Vahe Andonians: It all started with Moody’s. They approached me with a problem. The problem was extracting financial PDF. They had identified this as a pain point of the market. They had just created a unit called Moody’s Accelerator which was trying new ideas and products. They approached me along with others and asked us if we can automate it.

I looked at the problem. I thought that this is a solvable issue. It’s not a natural language issue. It’s much more graphical. It’s tables, headers, and these kinds of things. We developed AI models for that. Operations had a few people, but the company was large.

I wanted to do this with El, who is now the CEO. El was busy at that time. He’s also a serial entrepreneur. I had to wait a little bit for him. The two of us founded it. We started with extracting financials out of PDF.

Sramana Mitra: What was the relationships with Moody’s?

Vahe Andonians: Only client relationship. It was fully bootstrapped. The idea was not to go over 30 people and just do AI and science. Deep learning, in a very rudimentary way, mimics the brain. That may be an exaggeration. In contrast to the brain, it learns much slower. You need a huge amount of training data.

The training data was generated by Moody’s through a separate team in India. We’re developing the models. I figured that was not a good setup. You had a linear cost in India. If you need double data, you pay double. The utility of the training data though has a marginal increase. It was clear to me that this is not economical. I figured that that was not a very good solution.

I said, “I have an idea. Build a platform where we can develop AI models, get them into production which annotates the document for you, and then a human verifies.” We switched from a model-centric view to a data-centric view. It’s more of data rather than the feedback loop. We created such a platform. If you want to develop deep learning models to extract information from documents, this is the right platform.

Sramana Mitra: What is the go-to-market for this?

Vahe Andonians: They’re still our client for not only that unit but other units as well. We also have other clients. In the beginning, it was mostly word-of-mouth. Later, we were actively targeting large corporations.

Sramana Mitra: Are you selling the same use case to multiple large corporations?

Vahe Andonians: What we do is SaaS. We have this platform. They run it on their premise or on their cloud subscription. The idea is that it is use case agnostic, especially for these larger companies. For any document where you want to extract information, it works. We have a set of foundation models that outperforms cloud AI because it is targeted on the industry. It outperforms generalistic models.

We have just recently started a second strategy where we go to the midmarket with specific use cases in an almost self-service way. We push it to them so they can try it out. Then the sales team contacts them and tries to close the account.

This segment is part 3 in the series : Bootstrapping with Services at the Cusp of AI and FinTech: Vahe Andonians, Founder of Cognaize
1 2 3 4