Building a Shared System of Trust
How did we get here? After an all-consuming spell as founder of Upthere, I decided to take some time to rest, but also dive back into tech that had long been of interest: Machine Learning, and particularly the analysis of unstructured data. To combine the two, I took my bicycle and a small backpack, and biked what was probably the most inefficient route from Tokyo to Kyoto. While pedaling around, I listened to podcasts, read books and wrote some ideas down on my tiny laptop. (I also had some good food along the way but that’s another blog post!)
Soon, the idea of fact processing was born: the goal was to extract facts from unstructured documents (PDFs, etc.) and then store them in a structured way so they can be queried. To explore this space I assembled a small team and we started building a prototype to prove out our approach and did lots of demos.
A mutual friend introduced me to Hemant, who introduced a very interesting challenge he wanted to tackle in the insurance space. Soon it became clear that the technology that we'd been working on was perfect for the problem Hemant was trying to solve, and we joined forces.
What would it take to build a shared system of trust, providing a single source of truth?
Data is the lifeblood for the insurance industry, yet gathering, curating and publishing that data is a hard problem. We’re all familiar with folders full of Excel spreadsheets, PDF files, images and so on. Before the data is useful, it needs to be transformed. We set out to build a platform to do just that!
A few different requirements jumped out as being critical to building that platform.
Trust in your data:
In order to trust your data it needs to be Complete and Correct. We’re taking a wide range of documents, including PCA’s, Engineering Reports, Blueprints, Loss Reports and so on to make sure that we get a full understanding of a property. We then curate and use proprietary Machine Learning algorithms to make sure the Data is correct, including an extensive QA process. Most of the data is gathered either directly or close to the source of information (“affinity”), which increases both the accuracy and timeliness. Additionally we provide evidence links so you can see where we got the data from - right back to the originating document.
Control over your data:
It’s important that you are in control of your data. The customer controls not only who has access, but also who can see what. Data can be reviewed and approved before it’s published on the site.
Ease of Use:
From onboarding to our User Interface - we prioritize ease of use to make the experience enjoyable and efficient. Our onboarding team and technology will help collect your documents, and after you have given them to us - we do the rest. The user interface is easy to use, beautiful and fast. We show the most relevant analytics and insights in your dashboard so you won’t need to look for them.
Some of the data on the platform is highly sensitive. Data is stored and transmitted in encrypted form only, and we use the most modern Authentication and Authorization methods, including 2FA. In addition we monitor for unauthorized access to detect any anomalies.
We’re already storing key information on properties worth 100’s of billions of dollars, and new ones are added every week. Working closely with our customers has been thoroughly enjoyable and will help us to continue to deliver the most relevant functionality.
If you haven’t already, I hope you’ll join us on our journey!
Written by Roger Bodamer
Chief Technology Officer and COO of Archipelago: Previously, he was the Founder/CEO of Upthere, EVP of Product at MongoDB, and VP at Apple Computer.