HOW INNOVATION (REALLY) HAPPENS: INTERVIEW WITH RON GLOZMAN OF CHISEL AI – EPISODE 073
In this episode of The Digital Broker, Ryan Deeds interviews Ron Glozman, CEO and founder of Chisel AI, about the company’s journey from inception to present-day operations. By listening to this episode, you will learn:
- How a textbook-crunching app became a policy checking tool, and what this transformation says about how innovation really happens
- How artificial intelligence, specifically natural language processing, can help insurance agencies expedite processes such as policy checking, saving the agency time, energy, and money
Talk about the things you don’t see coming. When Ryan went to Dig|In earlier this year, he had no premonition that he’d meet Ron Glozman, the CEO and founder of Chisel AI, and that the two of them would hit it off so well.
But that little bit of serendipity is nothing in comparison to Ron’s story. Barely five years ago, Ron didn’t have a clue that he’d be working in Insurtech, let alone be pioneering artificial intelligence within it.
It all started when Ron was a computer science student at the University of Waterloo (“The MIT of Canada”) in 2013. Insurance wasn’t on his mind at all in those days; actually, a lot of other things weren’t, because he had to study so much for his exams. Too much, it seemed. Ron noticed that his textbooks were huge, whereas the corresponding exams were dainty little things that tested students on just a few fundamental ideas. Poring over those gigantic textbooks in pursuit of just a few ideas seemed like a heck of a lot of work. So Ron had an idea of his own: could he get a computer to read the books for him, identify the most important content, and crunch it into summaries that he could then review in preparation for the exams?
Ever the industrious computer scientist, Ron developed a program to do that. He called it Knote, for Knowledge Technology, and it employed a mechanism called term frequency-inverse document frequency (TFIDF). In layman’s terms, this is the comparison between how often you would expect a term or concept to appear in a body of work and how often it actually appears; the relationship would highlight which ideas, terms, or concepts merited the most attention. Knote was able to condense all of Ron’s textbooks into reliable, one-page summaries for each chapter, dramatically reducing the time he needed to study and liberating him to do other things.
His classmates noticed and implored him to make the app public, so Ron put Knote on the Google app store. Knote would later become the basis of Chisel AI and service insurance agencies that broker hundreds of millions of dollars—so it is funny to think that Ron came close to holding the app back entirely because he was a little stingy about the $5 deposit Google required to list an app at that time. “I was double-guessing whether it was worth it. I’d wanted Knote to be an app for myself; I never meant to make any money on it, and $5 didn’t seem to be worth the trouble. At the end of the day, I bit the bullet and paid the $5. Looking back, it’s crazy to think how far that investment has come.”
But the launch was initially unremarkable. For the first two weeks, Knote was downloaded a couple of hundred times, mostly by Ron’s friends and classmates. Then, one morning, Ron woke up to find that, literally overnight, the app had gone viral, having just been featured on Product Hunt, a popular app index. Thousands of people were downloading it all over the world. Mentors and investors accosted Ron, looking to grow the app and introduce it to other markets. There was just one problem: they couldn’t find one. Ron had developed Knote to help students study for finals, but academia wasn’t a market where an app like this could make a serious profit. The legal field seemed more promising—you would expect lawyers to be interested in an app that summarizes lots of information quickly. As it happens, many lawyers are attached to their billable hours and do not get jazzed about the prospect of reducing them. Knote had hit a ceiling: it had grown, but it could not find a niche to thrive in.
One day, Ron was demonstrating Knote at a conference as part of a talk he was giving about artificial intelligence. He hadn’t thought much of the talk: it was another routine promotion of Knote. But it was to be a watershed moment. After the conference, a woman who had been in attendance sent him an email. She was a Senior Vice President at a major insurance agency, and she asked Ron whether Knote, or something else like it, could help her agency resolve some of its own issues.
Her team, she explained, needed help with policy checking—the laborious process by which old and new policies are studied and compared in order to mitigate E&O exposure. The agency was already employing a team of people to meticulously review documents one by one—basically, to pore over large amounts of text like Ron had to do back in college. Because this was an agency with a very large book of business, the volume of documents to check was massive. The policy checking team was exhausted, and many documents routinely had to go unchecked—a huge E&O exposure. Could Knote be reconfigured to do policy checking on its own, saving the agency time and energy, keeping E&O to a minimum, and liberating the policy checking team to do other things?
Yes. Over the next three and a half years, Ron and his team would reconfigure Knote, rename it Chisel AI, and fine-tune the technology to deliver those benefits to insurance agencies. At the heart of Chisel AI is a mechanism called natural language processing (NLP), the ability of a computer to read. This is not the same as scanning, aka, optical character recognition (OCR). OCR is useful for many things, but it cannot understand a document, interpret the content, and make connections within it. A scanned insurance agreement could list the parties involved therein, but it couldn’t tell you what those parties do. For that, you need natural language processing, which is what Chisel AI specializes in.
Policy checking is serious work. The smallest oversight can lead to significant E&O exposure. Wouldn’t you rather entrust that kind of thing to a human being who understands the policy better than anyone? Well, no. It is commonplace, at the larger agencies, for many policies to go unchecked simply because a team of humans alone cannot handle the volume. The policies that do get checked are often vulnerable to human error. Human beings, for all of their amazing attributes, are imperfect readers. They get tired, they zone out, they miss or forget things, especially as the amount to read grows. Under the proper circumstances, a computer is far less prone to making those mistakes, and it can go through a lot more material. Chisel AI doesn’t remove the need for human oversight entirely—Ron, after all, still had to read those one-page summaries on his own after Knote, Chisel AI’s precursor, crunched them for him. Both tools, however, reduce the amount of time and energy it takes to do the same job.
For the time being, Chisel AI works primarily with high-volume insurance agencies, simply because those tend to be the ones that take policy checking most seriously. But if Chisel AI can continue to reduce the expenses associated with policy checking, not only can the company expect to grow its clientele, but policy checking, as a whole, will become more popular and affordable. “Most of the small agencies we meet admit to doing very little if any, policy checking,” Ron explains. “It’s hard for them to justify the expenses. That’s what we hear most often: ‘I don’t have the budget for this,’ or ‘I don’t have the resources for this.’ Our goal is to make our solution so easy and affordable that you can access it online as a SaaS, at your own pace. We hope to be there by the end of 2020.”
Chisel AI has found its niche, and it’s Insurtech. It’s amusing to look back and see how long it took the company to get here. Amusing—but not surprising. Go back to the beginning of Ron’s journey. Was he supposed to know from the start that he needed to develop a product to simplify policy checking for insurance agencies? Of course not. All he had was a problem: a lot to read and limited time to do it. Everything that happened afterward was in reaction to that initial identification. Ron did not set out to be innovative, nor would he have been so successful if he’d tried. He had a problem first, and he solved it. That is how innovation happens.
“It is up to the organization to identify where the business problems are,” Ron explains. “I’ve seen some companies do innovation for the sake of innovation: they try to come up with something cool, and then they try to find a use for it. In my opinion, that does not work. You need to start with a problem and then find a solution for it, rather than the other way around.”
Ron Glozman’s journey is an embodiment of what we try to teach here at The Digital Broker: find your problems, your soul-sucking tasks, and get to work on resolving them. Don’t know what your problems are? Ask your team what it hates doing. Then, commit to develop or discover solutions. They won’t come immediately, or without effort. Think of the woman who emailed Ron after the conference. Presumably, she’d been aware of her policy checking problem for quite some time, and on the lookout for a solution. To find it, she had to keep her eyes open, go to conferences, and ask questions. Please do the same.
Though we recommend going to conferences (by the way, will we see you at Agency Connect this year?), you don’t have to step away from your desk to meet insurance professionals who are curious, knowledgeable, and eager to discuss operational excellence. You can request to join our Digital Broker LinkedIn group and start posting. When was the last time you took a good, long look at your agency’s problems? Do you have a list of what they are, in order of urgency or importance? How far along the road to a solution are you? Would it help to ping our community for guidance or advice? Join and find out.