Hyperscience at ITC Vegas: Redefining Insurance Automation

Emilio Figueroa: Hello, I am your host, Emilio Figueroa. We’re at ITC Vegas live with Brian Weiss from Hyperscience.

Brian Weiss: Hi Emilio. How you doing? How are you?

Emilio Figueroa: Good. It’s nice to be here today, Brian. Let’s get down to it. Let’s get dirty with this. Tell me about Hyperscience. What are you guys doing in the industry?

Brian Weiss: Yeah, so Hyperscience is an AI company and we are focused on back office automation and really any of the problems that companies have with human information, right? So we bring AI to that problem of sorting out what I’m looking at, what it means, and how can I translate it into something I can automate. Uh, company is about 10 years old, we were founded by machine learning engineers who had taken their understanding of deep learning to actually sell, sold their first company to SoundCloud, but then they turned the concepts of machine learning to problems in the back office.

And the idea is if I can solve for what humans have to do with machines, you see where I’m going with this, right? I can create a lot of value for companies and their first use case around handwriting. Okay. Um, so deep learning and deep learning models. Handwriting is a great example of like, you cannot build a rule to tell me the next piece of handwriting. And so very quickly after that, we become a standard in the back office for really modernizing processes that are stuck with, you know, bad data, human information, that kind of thing.

Emilio Figueroa: Insurance, I mean, we’re very legacy. We don’t have stuck legacy workflow. Oh, we don’t, you don’t really? No. We’re, you know, we’re in hyperspace right now. Every legacy system we have, I mean, it’s, we need it more than any other industry out there. We are risk averse. We try to minimize exposure. Um, but by doing so, we also inherently change the way we move within technology. ‘Cause we are risk averse, right? So the whole point is simplifying these engagements, bringing technology into it that can remove those outdated workflows or human led workflows or RPA led workflows.

Brian Weiss: Yeah.

Emilio Figueroa: Tell me about, you know, let, let’s talk about the AI and what you’re doing to mediate, you know, RPAs and offshore processing and all these other things.

Brian Weiss: Yeah, well, look, the insurance industry is really at a pivot point right now. And it’s pivot not just around technology, but also processes, but it is large. I mean, the technology inflection point with AI is dramatic. It is something that will change and should change this industry. If you think about the way we tried to do any kind of automation, right? I’ve got a human task and I need to try and automate as much as possible. The legacy technologies, the first generation really kind of failed because they only got to about 60, 70%. And the reason they did that is ’cause they were based on very rigid rules.

So OCR for example, it’ll count the 14 characters, and if it got 14, it wins, but they could be all wrong, right? So, right. You could get all 14 and get it completely wrong, and then you gotta hire a bunch of people to go check that the machine got it right? And so there’s this combination of what we can hope machines can do, and then what we have to spend on people when they get it wrong. Now, the BPO industry is roughly $550 billion of “your mess for less” of where machines fail. So the testament to where we as technologists have failed the industry is the fact that there’s a $500 billion market of people to do the work that we can.

Brian Weiss: So what the premise here is that you can actually do what people do now. So with the models that we build internally, we build on customer data, we create sovereign models that are really like digital workers. Um, and we can do 99% accuracy with 98% automation. So it’s a really interesting conversation, Emilio, when we talk to technology people are like, “Well, where does you end? And where do people…” And we look at the whole process now, like what does it take to get all of that work done with 98%, like at human level scale with as much machine as possible. And so we’re creating digital workers that do that work now. And the beautiful thing about AI is they get smarter.

Emilio Figueroa: Correct. Right? My, that’s why the old technology’s topped out, is they don’t actually learn. So you put a learning technology that every time it sees a new document or a new form or a new something it hasn’t seen before, or some handwriting—mine’s terrible by the way. Right? It gets a little smarter. So this concept that you can do that work now is novel and it’s kind of groundbreaking and it’s breaking a lot of silos that live in businesses. I mean, you have that experience.

Emilio Figueroa: Yeah. I mean, we’re in the infancy of AI, right? And then, but it’s exponentially getting bigger, better, faster, removing the human element when it comes to errors on simplified tasks that are, you know, outsourced to these BPOs, these RPAs, all these different things. The whole goal is to remove that error. It’s great when you can have efficiency, but if you maintain the error, then it’s pointless and useless.

Brian Weiss: Well, having a machine do it, does it, and…

Emilio Figueroa: Yeah.

Brian Weiss: And it’s a really good point you make about errors and accuracy, because one of the things that we at Hyperscience are known for, have spent a lot of time on is ensuring that accuracy is controlled, right? So a lot of these systems like, “Yeah, I think I got it right. You figure it out.” The way Hyperscience works is it’s learning on your data, but then if it’s at all confused, right? If is an A or a U, I’m not quite sure on the handwriting, instead of dumping it out, it actually raises its hand and say, “I need a little help right here.”

And so what we end up doing is taking all that people you spend on BPOs, I take a fraction of it and I put it right next to the model, and I help it every little just-in-time, human-in-the-loop. And then the model gets a little smarter. And so what that ends up doing is creating this accuracy harness whereby I can tell you it’s gonna be 99%, right? Mm-Hmm. And every, for every penny you spend helping it, it’s gonna get a little bit smarter every time. So we get, we see efficiency gains that are eye-watering with some companies when they put Hyperscience in. Um, 70, 80, 90%, you know, efficiency in the ROI and the cost savings that drops outta that and some of these very complex processing is kind of astonishing. I mean, it’s refreshing when our customers do it.

Emilio Figueroa: Oh, absolutely. And it’s, you know, when they get that well feeling and, you know, that’s what makes you coming back, you know? Yeah. When someone’s excited, you’ve simplified a process, you’ve engaged them in the richer human sense, which is even better. Um, looking at the current processes for the insurance industry, the workflows that we have, you know, what do you, you know, how do you guys see… what do your clients say when they see that we have a reduction in workforce roughly 40% in the next three to five years?

Brian Weiss: Yeah.

Emilio Figueroa: You know, people are scared about not having enough data, enough… not having enough human, uh, you know, enough humans to be able to do processes like these. They wanna remove the errors, but they have that, but they also don’t wanna, they have the fear that you might be replacing them. Yeah. Tell me a little about that.

Brian Weiss: Well, you know, it’s interesting when we approach our customers, it becomes, they’re so used to spending that, and it was a reality that because they’re only 67% accurate, they’re gonna spend a lot of money on people, right? And so, most of our customers actually just spend that forward on growth. So instead of thinking about reducing head count, I’m actually gonna give you another 10, 15, $20 million to spend next year that you didn’t count on, right? So there’s this combination of repurposing manual, uh, rote tasks. I mean, data keying, right? I mean, it’s that’s fairly repetitive and rote. I mean, it’s the kind of thing that you would wanna automate. And most of our customers are seeing that as investment potential in the business as opposed to job loss, for example.

Right? And the other great thing about Hyperscience is because we are allowing customers to build their own AI effectively around their data. So the people who might have been doing knowledge work and data keying, they become experts in training models. So we have a harnessing system that allows business users to train the model.

Emilio Figueroa: And it’s beautiful that you’re creating an additional skillset for the workers.

Brian Weiss: That’s right. So now, instead of… and you don’t have to, they don’t have to be a data scientist. So we’ve simplified all the labeling and the QA and all of that to a very simple interface. And so those folks are being repurposed into model experts that drive the accuracy, you know, that they used to do in keying work, but now you’re getting into 99%, you’re getting a true human level. But what happens, Emilio, is they start taking on the impossible use cases. So what’s really exciting for us is when somebody says, “Man, we’ve never been able to do that. Or, oh, this is a total mess. Like, we can never… like doctor’s handwriting, prescriptions…”

Emilio Figueroa: Gimme a client example of what, you know, the biggest thing you’ve seen a wow factor from one your clients.

Brian Weiss: Well, I just called, I mean, it was Dr. handwriting prescriptions and things like that. They’re so varied, and there are some, there are customers who were genuinely flabbergasted that we can do that to the accuracy level that we can. Um, it’s like, “What do you mean? Have you seen Dr. Handwriting before?” Like, no, we get it, right.

Emilio Figueroa: Chicken scratches. Yeah.

Brian Weiss: It’s horrible. Exactly. We, when we get it right.

Emilio Figueroa: Tell me how Hyperscience is gonna be future proof.

Brian Weiss: Yeah. Well, so I mentioned earlier what we do is we allow you as a customer to start with our models computer vision, right? But on top of that, you’re training with your own data. So you are effectively building your own AI with your data that becomes proprietary to you. And so you’re building competitive advantage and it’s highly controlled. It’s not only accurate, but also transparent. You know why it got that answer, you know, if it got answer and you can start to stack it. So you’re competitive advantage just builds up. The opposite way is, “Oh, I’m gonna use a third party model. I’ll pick one of the hyperscalers and use one of theirs.” Well, what if it gets it wrong? Or what if they change the model? Like, I can’t really call them up and ask them to make it better. You got this. So you are creating like competitive differentiation over time by owning the sovereign models you build in Hyperscience.

Emilio Figueroa: Now, like any model, everything’s just, everything’s based on the data. Making sure that your data’s cleansed, making sure you have some, some type of standardized data. In this industry, we don’t, no one has clean data. ‘Cause we are legacy, we have historical years of data and data lakes that just sits there. How do you guys simplify that?

Brian Weiss: Well, there’s two ways, and you’re absolutely right. The dirty secret of anybody who tells you they’re gonna do magical things with data is you have to do all the dirty work first, right? Data cleansing is something that always trips up any of these innovation ideas that come up, whether it’s AI or whatever it might be. That’s, you have to start with really clean data.

We do that two ways. First of all, there’s an accuracy harness in the model itself, or if it’s not confident, it’ll ask you to help, right? So in that regard, if it sees variability, you’re gonna help it in line as opposed to having to wonder why it got it wrong later. So it’s kind of like a self-healing cleansing process inside the data process. The second one is that Hyperscience lives on an orchestration pipeline. So in addition to “what does this page say and what do you want it, what do you need?”… so what kind of a decision are you trying to make… so we do all sorts of secondary validations against it. So you know what, I need to make sure that my next few steps where I’m gonna attach this to a particular customer who’s got this much in their annuity, et cetera, et cetera, we can go look up all of that data and do the cross-referencing for metadata that ensures what you’re looking at is correct. And that’s a big difference with Hyperscience.

Emilio Figueroa: That’s huge.

Brian Weiss: Yeah. We are a full orchestration platform. I mean, we’re very, very good at getting data in a clean way, but what you do after that is very often done in Hyperscience as well. And I think we talked about it earlier, I mean this… you mentioned the RPA market like what RPA did and does, like we can do almost all of the important stuff inside the same environment where you’re understanding the information without breaking…

Emilio Figueroa: Without breaking.

Brian Weiss: But I mean, that market is so… I mean, it is, it was a, by definition, a bandaid, and it had its moment where, you know, you can’t tie these systems together. So let’s get bots to do things. But anytime there’s anything that’s not, you know, a rule or very, very rote and repeatable, they break. You can’t handle any kind of an unstructured…

Emilio Figueroa: Any UI changes, anything, any real change to break it, right? Yeah. Um, yeah. It’s crazy how we’re still working with technology like RPAs when we have new technologies coming out into the world, like Hyperscience. Um, engaging the industry with it is super important for companies and individuals to use as a tool. Yeah. And so they don’t see it as a replacement, which is not a, it’s not gonna be a replacement. It’s a tool to simplify, engage and manage extremely complex and heavy dataset. Whether they come from the actuarial side or underwriting side or claim site, tons of data everywhere. Yeah. We need to be able to filter it as fast as possible. Um, you know, gimme a real life world example as to how Hyperscience is doing that.

Brian Weiss: Yeah. Um, uh, I’ll, uh, AIG Corporate is a customer of ours. Okay. Um, they, you know, do all annuities, insurance. Uh, they started out with sort of basic simple forms, right? That they were looking at and they were trying to modernize their process for understanding that. And the first couple steps in that they saved roughly 70%. Um, you know, they got a 70% efficiency gain on top of that. But as they then progress, they look for all of the harder pieces, some of the more complex pieces that have handwriting in them. And now what they’re doing is they’re going on and looking all of the downstream, what happens after this? What’s the next step that happens?

And usually it’s somebody has to go into a system, look up some information… they’re actually taking the underwriting work that comes after that fact and automating in Hyperscience. And the impact to that business is, I mean, it was, they got 70% efficiencies right out of the gate, but they’re gonna see five times that in the very, very expensive part of the business. Like, what do underwriters do with that work? All of that will get collapsed. The benefit to the company, yes, it’s cost savings, but the speed to which they can process is collapsed significantly. Um, and that benefit to is, um, it’s hard to quantify the business, but you can say, if I can do it in five hours instead of five weeks, what’s the benefit to my customer experience?

But they’re seeing both of those impacts of Hyperscience. And they started really slowly, they came in with a model they started building, and then they realized how efficient it was. And then they started, they sort of progressively, and they had to go change the culture internally because “we don’t need a BPO anymore. I can just take a little bit of help over here and get this.” So their internal transformation, probably a two year journey, that their both their cost savings and the speed to which they can process claims is off the charts.

Emilio Figueroa: And that’s monstrous. I mean, you have to look at like the other costs that are saving by even by being able to process a claim faster, you’re gonna lower your legal liability from it. Yeah. You’re gonna close out the claim. You can look at your, your EULA and change the way you’re looking at things because you have lower legal expenditures due to the closure of the claim. Yeah, yeah. You know, and it’s not only about efficiency, but it’s about bottom line when you look at the overall aspect of it. Um, looking at workflows, you’re simplifying these workflows. I mean, insurance is laden with workflows. We love our workflows.

Brian Weiss: Yeah, yeah.

Emilio Figueroa: Um, you’re minimizing those workflows. Gimme an example of, you know, what you guys have done to do that.

Brian Weiss: Yeah. So, um, at Canada Life and their subsidiary Irish Life, um, very similar process, whereas what they have done is taken the information that’s coming through Hyperscience and they are using it to begin to populate an automation bot, right? To be able to help a claims adjuster make a decision about it. So yes, they’ve gotten significant automation rates coming through, but at the same time, they’re also populating a RAG architecture. So they’re taking that data and using it to feed an LLM, if you will, that knows their business.

Brian Weiss: And so at that point, they can ask questions about not just the particular claim, but the history of the claim. They can ask questions about, um, or they’re also looking for fraud at the same time. So there’s this concept of being able, not just to run it, pass it through, but also collect and accumulate information over time, which then allows you to build potentially AI on top of your own data to ask questions. You hear a lot about Agentic AI and “how do I make an agent that knows my business?” Um, that’s a company that’s doing exactly that type of work. They’re kind of at the bleeding edge of it. I think there’s an intersection between what back office automation does in creating clean data and the desperate need for that clean data by folks who wanna do broad AI initiatives inside a business so that Canada Life from Irish side are customer of ours that we innovate with on that front.

Emilio Figueroa: That’s pretty amazing. Um, you know, and looking at, looking at what you guys are doing, I mean, you can grab historical data from the carrier and not only simplify the current workflows that they’re gonna have, automate future workflows that they need to build, um, quoting, binding claims, et cetera. But going back historically and analyzing the data that they have so they can see where the problems have been financially, and you can track loss differentiations, you can look at revenue changes. There’s so many things you can do, right? Whatcha guys doing with that?

Brian Weiss: Well, we’re doing a lot of that right now. You sort of cracked open the box there on what the next frontier is. Um, because everything that you mentioned, it starts out as a side effect. So I, we go to our customers, “You’re already processing this data, it is flowing through the pipes and it is extremely accurate.” And everything you’ve described, being able to do historical analysis, fraud analysis, you better be darn sure the data is correct. Correct. You can’t misplace a decimal place because your OCR was wrong. Like, you can’t get it wrong. It has to be Right. But our system guarantees it’s correct.

So at that point, you have this trove of data, and we have customers are doing what you’re talking about. And the light bulb goes on that this is really valuable business data. And if I accumulate enough of it, I can tell you a lot about trends. I can tell you a lot about what you should and shouldn’t do. I can do all of that. I can train an agent to read that data…

Emilio Figueroa: For underwriting, for claims, for everything.

Brian Weiss: Exactly. So this, you hear people talk about, “Okay, we’re gonna create an AI for underwriting for my company.” And you hear all about the AI part of it, but you don’t hear about how I’m gonna get the clean data that starts the process and guarantee that it’s going in. So you end up with garbage in, garbage out.

Emilio Figueroa: Correct.

Brian Weiss: And then those projects stall. Our customers are coming to us and saying, “Well, hang on, we’ve got the clean data and Hyperscience is guaranteeing it’s clean.” All of our folks from machine learning and AI and all of that, they’re coming to us and asking for it. So there’s this synergy happening between what we do with back office automation, um, and what I think the future of that, you know, building agents for underwriting, for all of the processes that we see that can automate and help as decisioning assistant, right, if you will, is the first couple of steps. And then I know you, I know you work on this and you’re smiling at me.

Emilio Figueroa: Yeah. I’m thinking of the different things that this can be used for. Absolutely. Yeah. Yeah, yeah. Um, Brian, unfortunately, we’re out of time. You can sit here for another hour chat, wait for a beer. But unfortunately our time has come to a close. Uh, once again, Brian, thank you for joining us. It’s great to see you again, everyone. Brian Weiss from Hyperscience at ITC Live.