Indico Data releases industry-first large language model benchmark for document understanding tasks
Learn More
  Everest Group IDP
             PEAK Matrix® 2022  
Indico Named as Major Contender and Star Performer in Everest Group's PEAK Matrix® for Intelligent Document Processing (IDP)
Access the Report

Interview with Automation Center of Excellence expert Ozan Eren Bilgen

Watch Christopher M. Wells, Ph. D., Indico VP of Research and Development, and Ozan Eren Bilgen, CEO & Founder, Base64.ai, in episode 10 of Unstructured Unlocked. Tune in to discover how underwriting leaders are solving their most complex unstructured data challenges.

Listen to the full podcast here: Unstructured Unlocked episode 11 with Ozan Eren Bilgen

 

Christopher Wells: Hi there. Welcome to another episode of Unstructured Unlocked. I’m your host, Chris Wells, VP of research and Development at Indico Data, and I’m really pleased to be joined today by Ozan Bilgen CEO and founder of Base 64.ai. Hey, Ozan.

Ozan Bilgen: Hi Chris. How are you? Good to see you.

CW: I’m good, thanks. Good to see you. How are you?

OB: I’m doing pretty good to you.

CW: Excellent. Well, why don’t you start off by telling folks who you are and what you do over there at Base 64?

OB: Well my name is Ozan again. I’m the CEO of Base 64.ai. We automate document processing. this is an AI service that can process all types of document. We mostly specialize in structured as structured documents in insurance, banking, K Y C, and you know, healthcare industries.

CW: That’s great. Yeah, I just to let the cat outta the bag, base 64 is a partner of Indico and we’re very happy about that. And I can tell you of all the companies we worked with that have sort of like the model Menagerie style, and I mean that in the best possible way. base 64 s coverage of what’s out there is, is you know, it’s just not matched. It’s a really good stuff.

OB: Great to hear that from you. You’re also very, very happy to partner with NICO Data.

CW: Awesome. Well, I think that’s mostly the end of the commercial. So let’s, let’s talk a little bit more about what’s going on over there. So, as c e o of you know, an AI company working with the enterprise which is a tough job what are your, what’s your day-to-day look like?

OB: Well, MyDay is coordinating across the teams. So our sales teams go out and try to find new customers in the insurance world. we are mostly focused on liability insurance. that’s you acor forums, loss runs, and those kind of work. we also have customers in vehicle insurance which have very different documents that driver license, vehicle registration and so forth. And then we have customers in healthcare insurance and they, we process summary of benefits and coverage forms, s spc which is the semi standard document for every US insurance company. Okay. For some of those things, we are the only provider and we achieve that by building it for our previous customers. So now we are trying to expand into this market.

CW: Right on. I know enterprise sales can be a slog sometimes. What do you see as common obstacles? you know, someone wants to do something, you’re ready to do it for them by hooking them up to your APIs. What gets in the way?

OB: Yeah, so, you know, a lot of things. So let’s start with, for example, pricing. There’s no set standard pricing or expectations on what this should code. Cause in our calculations, we provide 20 to one ROI when it comes to manual data entry. So it’s actually pretty cheaper relative with that. Even then it takes a while for companies to adjust to that thing. And then we, we see, like, you know, legal teams, they’re, they have their own concerns and questions for that. We address them with certifications. We have HEPA SAC two, type two G P R certification. We, we are receiving our i o 27 0 1 this year. so all those things that, you know, kind of help elevate the question. Now, you may hit the engineering blocker, so the engineering team’s super busy. They can’t do that. With that. We are working around this, this by having 400 ready integrations to third parties. So whether it’s like Salesforce, a Google drive or database, email or scanner whatever you need, like, you know, just basical before I comes out of the box ready for it. And we can basically you know, bypass that in busy engineering schedule, at least like to get started with it.

CW: That’s fantastic. I want to, I want to double click on a couple things. out the box integrations are great. It sounds like this is a lot of plug and play stuff. I think I heard you say 10 to one ROI on data

OB: Entry 20 to one. Yeah. And one of the insurance customer is 21 ROI for acor forms.

CW: That’s a big number. and then I think I heard you say that one of the, the biggest blockers is that you’re changing the process for the folks that do this. Right? And so that’s a big, that takes time.

OB: It’s a big deal. So it’s a new world. luckily, like, you know, all our prospects and customers already, you know, know and admitted that this is the future. It’s not like in an if question, I’m pretty sure by 2040, let’s say, nobody’s gonna do this by hand. So it’s a main question. So we try to obviously try to get them to adopt the new technology faster. And we believe that this is for best for them in terms of competition and, and, you know, in terms of focus and drive. so this is like a basically scheduling question that’d be helpful. Okay.

CW: So it’s the old saying the spirit is willing, but the flesh is weak. Huh?

OB: <laugh>. Nice. Nice. Yeah.

CW: Good. circling back to a little bit more about you. What, what brought you here? What in your background sort of set you up to, you know, to take on this role and build this company?

OB: Yeah, I come from software engineering background. I am actually bachelor’s and master’s in computer sciences. And I spent about 15 years in top tech companies as software engineer, architect, manager. And those companies include Microsoft, Netflix, PayPal, Uber and pir. throughout those, especially at Uber, when I was in, you know, one of the first hundred engineers there I noticed a big, an interesting thing there. So everything was done manually cause of the lack of automation. And when we were onboarding a driver, you were asking, Hey, can you upload your driver license? Your re registration and insurance card? Those images were shipped to Philippines and being transcribed manually. So it was creating a delay about like, you know, one to two days. And this is like, in the best case scenario, if the customer uploads the wrong document, then there’s a lot back and forth to try to reach back the customer with text messages.

Some of ’em come back, some of I’m done. later in my career at Uber, I became the technical leader of the leasing department. Now, that problem over there was even worse because leading contracts, as you know very big. So you, you know, and Nico is great with unstructured documents, so it’s part of the unstructured document problem. And now you also like other types of things like, you know, titles and citations and you know, all those adverse action letters that needs to be processed through something. And that’s something with humans. So we hired more and more and more humans that we had actually 300 over 300 people in Phoenix, Arizona going through those documents every day. And even then, the average processing time for Elise was more than five days. So <laugh> with so many humans working on this we could not achieve this.

And then I realized like, you know, okay, that needs to solve true in ai, but we are talking about like, you know, mid 2000 tens thousand 17, in fact, all the way there. there was no good AI to solve this, that, that addressed what we were looking for. As, you know, like Transformers, they came out in 2019, 2020, and there were just babies back then. So now I ne recently, only recently the AI was able to like, catch up with that, but luckily the schedule also corresponded with my, you know, Uber’s i p and my departure. And I was like, when I was looking for next big project, I said, I’m gonna fix this. And I started with 65.

CW: Fascinating. So you saw the problem, you lived the problem, and now you’re solving the problem.

OB: I think those are the best. when you see the problem yourself, you’re, you get a lot of free motivation points,

CW: <laugh> Yeah. When you’re feeling the pain. Yeah, of course. That’s great. I love that story. let’s drill in a little bit more. there’s a saying that I love, which is that you ship your org chart. so tell me how you all have been successful in building out a software practice in a, to your point, transformers haven’t been around that long. How have you built out a successful rhythm practice of shipping AI into the enterprise? Which is a, which is a really new field.

OB: It is. So I think the humble beginning is like knowing what you’re good at and like getting the work that you’re not so good at to other people and being very open and transparent about like yourself. So what I was good at was software engineering. So I was initially effecti, welcome in this thing. And then when it, when we bring it to a certain point, we were ready to try it out in the public. And then I approached the friends and co feature colleagues who will become my product manager. And then later my business business development director. And then we hired, you know, more sales folks, more engineers into that team. We baked up the, basically the field of ranks. And we hired also like new roles such as customer success, right? As we have customers that we need to do those things. So over the time you build a li you know, still very small, but functional team that addresses many different aspects of selling business software as a SaaS market.

CW: Interesting. What, what was the toughest growing pain as you were building out the org?

OB: I think hiring is pretty bad. it’s like, you know, I’m, I think I was like very used to working in corporate environment. All the CVS will be beautiful, brilliant people. Like, you know, they come to your way and they apply to your, you know, companies who work at, I hear some that startup, you really have to go and like, reach out a lot. And that is that was new to me. So to be honest, that’s that, that took me some learning. But, you know, we found ways to overcome this. We work with recruiting agencies, we work with, you know conferences and so forth. We can find like, you know, good talent who’s looking for those. that was, that was one of the unforeseen challenges of starting agencies.

CW: Yeah. This is, this is a little off topic, but has, has the rise of, you know, sort of AI and the public consciousness with like chat G P T and stuff like that, has that drawn attention to companies like Bay 64, do you think?

OB:

Absolutely. I think everybody’s talking about ai, or at least I feel like that might be an echo chamber. But I see this on the news all the time. I see like it’s on the conferences. I, I see newsletter that are written by AI and they’re publicly saying that is a proud achievement <laugh>. So those are like, you know, really cool things. I think it’s crazy awareness about ai. we are not in, you know, chat pro gt business. it’s a completely different thing that we’re doing, but I think it’s, it makes basically installs that word AI in everybody’s mind, in everybody’s product line. Everybody’s like yearly schedules. So what are we gonna do with bot ai is gonna be the question.

CW: Yeah. I’m imagining a lot of CTOs and CEOs out there receiving like emails from their CEO E o with articles saying, Hey, what are we doing about this? Right. Exactly.

OB: Exactly. Yeah.

CW: Cool. So you mentioned a bunch of verticals where you’ve had a lot of success. and I want to, I wanna drill in on a couple, I’m, I’m really fascinated by digital transformation in healthcare and, you know, sort of, you know, property and casualty and specialty insurance. healthcare I’m interested in because it’s been, you know, you go to the doctor and the technology is 20 years out of date, right? And that’s sort of the story of that vertical, at least from lowly patients view. And then property like the insurance industry broadly sort of really suffered with the low rates environment that we were in forever and got serious about digital transformation automation. So what have you seen as like some of the, the key things that have made you successful in both of those verticals? And then I wanna circle back to what do you still see as struggles in both of those verticals?

OB: So what I see is like, you know, this industry is has never really thought about automating this process. <laugh>, that’s what I noticed. Like know, when I first started this, aside from few document types, there was no good standards for certain things. For example, there’s no good standard for loss, wrong. there’s no good standard document for medical ill sorry, dental insurance or vision insurance, right? so the titles, registrations, they’re all different in every state <laugh> rather than obviously in every country that you work in. So the sta the, the lack of standards was the a problem and also the opportunity that we, we, we found ourselves in because there was no good standards. What we built really helped that we built a zero shot processor that can un look, see the document and understand what’s the contents of it. And by doing so, we were able to address so many different states and variations on those documents for so many different document types.

 on the other side, like, you know, there were some good movements about making, creating some standards. Acor organization, which we are a member of Yep. Created a lot of great forms that standardize the information exchange. Information exchange standards are very important because it creates a way to easily understand and digest this information, not just a for AI, but also for humans. So there is no, like, you know, I meant to say this, but I actually said you this kind of like a conversations around when there’s a standard in communication for, and we support a lot of acor forms out of the box about like, you know, 20 do too many that are very, very popular, can be found immediately in basics fours. So you can just directly connect basics four, start uploading and get the, the, the data without doing any other further training or whatsoever.

 in the healthcare world, we have summary of benefits and coverage document called S P C, and this is for medical healthcare doc, medical, healthcare coverage, insurance coverage for the us. and those are semi standardized by healthcare go in two thousands. However, they’re not fully standard as well. Different companies, different insurance carriers, they add certain new items into it, or it’s a table looking like document. They merge columns and rows that the values are the same. So that confuses ai. So we had like an untangled dusting, however, we, we built the only processor that does it. We built in 2021 actually in three years ago. And a lot of companies are using it to process rspc. one of our customers, they were processing it 12 days this form, now it’s like 30 seconds. So there’s a big difference between 12 days and 30 seconds. Those, those are two different companies. Those are two different centuries, if you like. Yeah. In how do business. So if we take this companies, we take our customers from 2021 to 2000, you know, thirties, forties. and that’s what we did for them too.

CW: Yeah. So I have a lot of questions I want to ask on this thread. first of all, let me make sure I, lemme make sure I’m understanding correctly. So you have a, you have a zero shot approach. So if someone brings a new document, you can help them get value outta that new document, even though you’ve never seen it before.

OB: Yes.

CW: And then as they, as they keep sending more in, the solution continues to learn until you’ve got enough that you can say, we’ve seen enough of these, we’ve got the pattern, now you can send them in. Everything’s covered. Is that, is that roughly how this, this goes?

OB: That’s pretty accurate in, in many ways. So my TIF in engineering is I can read every document, <laugh>, you can read every document. Actually anybody can read any document. Those are made for humans. So if we don’t have any problems just to learn any document, let’s say I receive a new letter from my you know, utility provider, it’s about my power. Well, even if I never seen this, I can understand what this is talking about and what this yeah, what the, what I should be doing. It says pay here in this amount. and typical machine learning you have to go, a lot of training is, you know, it’s finding new, the correct neural network and all networks, multiple of them executing them training them, deploying them even in the hassle <laugh>, it’s, it’s not an easy thing with this zero shot approach, which we call semantic processing.

Our AI looks at the looks, the document like a human does, it goes page by page, understands like what the information is there, is there tables, is there key value pairs? I’m looking at signatures faces and those kind of things and extracts them into this standard document format that we provide for other documents. So this, this is a way like, you know, we do one ai, one API as well, <laugh> for all types of documents. It’s, everything happens under the hood for our customers going forward. If you happen to see a lot of demand on the document time, we build specialized models for it. Those models, they address things that and standard thing cannot necessarily address very nicely. For example, if it happens to see a lot like invoices that means can say, you know, we, we will find a lot of invoice date and what we wanna do is like, you know, you wanna normalize those date, you wanna convert them to year, month date format. They can come in you numbers, they can come in September, like, you know the word. So you don’t want to receive this data extraction. You want like isoform, you want ear, so you know exactly what that data is. So we do models that, that fetch as information in a stan and standardized information that our neural network layer or machine learning layer fetch. That’s great. This is how we build lot different models very quickly.

CW: Yeah, no, that’s, that’s very helpful. Having that mental model coming back to the business impact. So you talked about in the healthcare space, a form that took, or maybe not a form is too strong, but a document that took 12 days down to 30 seconds, that’s yes. Like miraculous in, you know, in the, in the insurance industry more broadly, you’ve got, you know, requests for a quote. You’ve got first notice of loss, you’ve got all these other big bundles of documents that come in. Someone has to read all of it, and then after that someone has to make a decision about it. When you take a process like that from, you know, a work week down to an afternoon, what have you guys seen as what’s the impact on the business? Obviously there’s a bottom line impact, we all know that with automation, but what are the other impacts that you’ve seen?

OB: It takes a lot of convincing that this can happen <laugh>. So a lot of our customers were running with weeks long backlogs to process documents. now they convert the backlog, entire backlog in like, you know, in in, in a, in matter of hours. Yeah. And get done process on like the new ones. They don’t stop. So actually they had to change certain processes because now things are happening too fast. They had to catch up with speed of ai. so that was an interesting point. The second thing is the teams that were manually doing data entry now become the teams that are reviewing the AI results. So, you know, this, this form that I mentioned, the sum summary of benefits form, it’s typically five to too many pages long form depending on the insurance provider. So typing in everything in there takes a considerable amount of let’s just round them up to an hour if you like.

 but if, you know, you have a task that takes an hour and an hour after hour, you know, if you have time forms, it’s gonna take like 10 hours, right? So we make that, everything is now the results needs to be validated. just like, you know, you just have to look ahead. It’s like, okay, it looks good. Yes, go send. Yeah. And so by converting this massive one hour task into like a few minutes validation task, you create new products within that company. So this company may now for example, ask the users to upload the document themselves. Instead of emailing them to them, they can say, upload the document, do your verifications like that you work with a very big service provider for building management. This is how they’re using r i for acor form. Now they don’t even wanna do it. They’re like, yeah, upload the user. They show it to the user, they like, does it look good? And it’s good. So, and if there’s any ever a mistake or whatever, the customer can pinpoint and change it. So it creates new products, new business opportunities, and a new company out of it through ai. Yeah,

CW: That’s fascinating. Like, one of the things I heard you say in there was that, excuse me, because the AI is helping, what’s the right way to frame this? You have the human who used to have to do all of the data entry, now they just verify the data entry. Or in rare cases, if it’s missed, they go back to the original document. Right? But that whole process of getting the data ready to make a decision is compressed massively. Right. And so now you’re getting, obviously the business throughput is increasing, but I think you’re also saying that because you’ve, you’ve really made you’ve transformed the process in such a way that it’s now like decisions are almost instant that allows you to make new products, right? Because you don’t have this like five day lag between when the docs come in and when, when some decisions are made. Talk to me, talk to me more about where you’ve seen that in the insurance industry. Cuz that’s fascinating. I hadn’t heard that before.

OB: Yeah, so one of our customers is receiving acor 25 forms. Yeah. from customers from their clients if you like. And they get them verified. They get you know, checked against the minimums. Another like contractual liabilities that they have. They, they maintain and keep a, maintain those forms. So previously their product was upload your form, they’ll get back to you. And some of us taking onboarding, taking like days and weeks and so forth, now they upload the form and the form is instantaneous, like displayed to you immediately now. Like, this is the viewer form, right? So this is the information we pull out, is that correct? And they can make like adjustments and changes and they do get notified of course if the customer changes something, which, you know, they wanna check against product and so forth. But this is very pretty easy to compare, you know, if there was any change on, on their own website. So this way, like, you know, if this looks good, then they can very easily onboard new tenants, new clients, new businesses through this technology. Yeah. yeah,

CW: You’re really bringing the customer into the, you know, underwriting process, right? Like you’re making them a partner in it, rather than I’m gonna throw my stuff across the email wall and just wait for you to make a decision. And I, you know, it’s a black box.

OB: So by bringing customers, actually our clients they have a lot of freedom in terms of how they wanna use their resources. Yeah. So it is much more easier to ask the client, let me take it back. It is much more harder to ask the client, Hey, can you fill up the entire form that I need that has like, you know, 200 different pieces of information? Yeah. But it’s much easier to say, like, upload your document and then prefill the information and show it to the customer. Like, does that look good? And actually it looks pretty smart, <laugh> and savvy too. If you do that, people are like, wow, this is happening <laugh>. so that’s, that’s what basics for AI does for them.

CW: That’s exciting. One of the other elements of data entry that it comes up on the podcast actually pretty frequently is how erratic it it can be, right? Like the AI doesn’t change how it makes predictions based on not having eaten breakfast in the morning. Right. Whereas employees have bad days. and so you get this effect where like the downstream data that you house afterwards becomes hard to do secondary processing on it, right. To do analytics and ask questions like, you know, I got this submission, I underwrote this risk. cuz you really can’t trust the underlying data in the database, at least at the level you’d like to. It sounds to me like what, you know, the pipes that you’re building, everything goes into a really nice tight format. Everything’s clean. Are you seeing any companies taking advantage of that? You know, that pristine data that they have post process?

OB: So there are actually a couple ones. So I have okay. Really interesting stories. One of the insurance companies that we work with were, they were getting out of the out of date insurance claim insurance policies. So basically the expiration date was, you know, in the past if you use humans that will take you again like days, weeks to get there. Like to realize, oh, this is actually late. you, you do this like in a back and forth 10, so you get a new one and everybody’s busy. Everything is already like, you know, <laugh> late and this causes a lot of friction. And they’re like, their clients are saying, well, why didn’t you tell us this before we heard you telling this now and not understanding. Like, there are a lot of humans behind this process now with the ai, the moment you upload it, we can tell that dispose expired and you know, you can build a product and say like, are you sure?

Like, you know, you wanna upload No, this is expired just in case. But again, you can say like, you know, to the user you know, you can give them the benefit of the lot. Maybe they uploaded the wrong file on their drive forward, right? Yeah. so that, so this way, like, you know, you can’t make the processes go really faster in a way like, you know, it was not able to be pulling more data and correct data. The second thing is even for very popular document types, when we use humans to manually extract data, they don’t extract everything, right? They’re, they have certain amount of time that they can spend on that document. So they will only like, you know, get the highlights, for example, from an invoice. They will get the total <laugh> and you know, the due date and that’s it.

They’re just gonna move on to the next one. So this is a huge problem, especially in insurance and finance role. The reasons are couple things. One is they don’t get the data that they can operate on for business intelligence. Like you said, the bigger one is fraud. Mm-hmm. So Facebook and Google got hit by invoice fraud. Yes. and these are like top tech companies, right? Yeah. I mean, they cannot do this. So they don’t like, think about other companies. Like they, they could be hit too for like over $200 million. What they do, what this guys do is like, it’s actually pretty, it’s pretty simple. They create an invoice and inside set out, yeah. They send an invoice saying like, you know, Hey, you owe me, you know, one mo 1 million in services I made, you know, graphic design. They’re like, okay, <laugh> your money why this is happening? Well, because they don’t have the teams to go and check the purchase order and the contracts and all those other things that is associated with this because they all, they’re super busy to processing manually these documents. They’re already back, they’re already behind schedule. So they don’t have the teams to actually mandate the fraud. No.

CW: Yeah. And even if the humans were 99% correct, that’s still, you expect to get $10,000 back right. From that million dollar invoice.

OB: Exactly. So the idea here is can you put AI here so it can extract everything in that document instantly. So you can use those data points to, you know, for better analytics, for better market research, finding like a cheaper vendor, for example. Or, you know, combat things like fraud and understand your business better than you can ever to Yeah,

CW: No, that’s a, that’s a great point. You’re understanding your own business better. I’ve been talking to a lot of folks in these industries, especially insurance lately, and everyone’s like really laser focused on if we could just get the documents in faster so we can make decisions faster, we’ll get more business. And that’s not wrong, but like, I don’t know, some massive fraction of those cases you’re never gonna quote on anyway. And if you could get the data and then build the AI layer on top to say, you know what, don’t quote on it. That’s not the kind of business we want to do. that’s where you really win. Like, that’s, that’s the end game for these types of processes. I think

OB: Absolutely. The future of business will be very different, right. So we all can agree on that one. the business of a century ago, if you like, yes, they were using like typewriters and they had those big books that, you know, now it’s super different. Like there are databases, e r p systems, now AI is coming into picture. well what, what’s gonna happen like, you know, a decade or two later. So yeah, let’s think about those things and let’s build for those things and let’s not wait two decades to, to make a complete like switch. Let’s start this, this transition right now. And the one of the things that you can do to transition today is document processing. And this is why we are super excited about it what we are building every day.

CW: Yeah, yeah. The the technology’s good enough. There are enough people out there that know how to be successful with it. Like yourselves. You’ve got a whole bunch of in-house experts and the problems are big enough, so there’s really no reason to wait.

OB: Absolutely. No insurance company that we work they deal with less than a million pages a year. Yeah. So just imagine like stacking those pages in front of you, it’s like it fills up the entire room. Yeah. And in real life in real business life, what’s really most important thing is like the efficiency, how much I’m spending versus how much I’m making this technology makes the spendings go down while it, it increases the revenues as you can actually scale the business now.

CW: Yeah, absolutely. Early on in Indico’s life a potential investor asked the question like, aren’t PDFs going away? Like, isn’t everything gonna be forms? And I think, you know, it’s kind of a naive question but maybe not like at that point in history, you know, the technology was new and a lot of things were being digitized on the web. and it’s interesting, I think what’s happened is PDFs will never go away now because the technology’s so easy to work with that like the interface just is the document. You don’t, you don’t need any other interface.

OB: Exactly. This is a very, very philosophical way to put it out there. So documents are, the interface documents are like XLS and Jasons for humans. Yeah, exactly. Can repeat through those things and we cannot see through code and, and, you know, metrics you don’t wanna see that. We need, we need documents to, if you, if you wanna put humans in this process, and we should always then you know, it is the name is document. So PDF d f is never going away for a foreseeable future. And it’s can, it can be a different format. It can be X, Y, Z but it will be some sort of a format that I can look at and understand until we have like maybe chips in ar so <laugh>

CW: Yeah. Even then I don’t want to have to do my taxes in xml. Like let’s, let’s not go there.

OB: No.

CW: Yeah. I’ll settle for a document. circling back to the point about the document being in the, the interface, you, you talked about it as well in a different form, which is that, you know, the, the case you mentioned, which is, you know, checking for out of date policies or expired policies, the ground truth is the document. It’s not like what’s in the e r P system or whatever it is, right? The document becomes truth and that simplifies the world in massive ways.

OB: I mean, it’s not about like, you know, what AI can do or not. It is the world we build around it. Like legally that’s the binding thing. Your database doesn’t tell too much to the judge the document that you signed, that it is the contract. So there are, there are certain aspects of it that that will keep the documents that on for, for a longer time than, than some investor.

CW: Yeah. I’ve, I’ve worked on AI implementations with documents from the sixties and seventies, you know, and old typewriter and really stretches the old OCR r capabilities right there <laugh>. So you know, looking, looking again towards that future, you talk about how you’ve, you accelerated some of your healthcare customers from today to like decades in the future. there’s gonna be an AI arms race in these you know, in these, in these verticals indico and, and obviously Bay 64 are already seeing it in the insurance verticals as that arm race goes, those goes on, they’re gonna be laggards, they’re gonna be left behind. Like they’re gonna lose market share cuz they can’t quote business fast enough. Right? They can’t, the customer satisfaction’s not there cuz everyone else is turning things around faster. So at some point you reach a steady state where everyone has the same size, AI guns, right? So like putting the crystal ball on the table, what comes after that?

OB: Great question. So, you know, once we do, once you’re done with that processing, now you talk about like what you’re gonna do with that data. Yeah. Right? Yeah. So the technology will never stop innovating either brilliant mind create new things every day, and pretty sure like, you know, this is how it’s going to happen in the, in the next round. So my understanding, my personal philosophy on that is like, if you can imagine it, this is going to happen. So my imagination, <laugh>, the feature looks like a lot of like virtual robots working in the office doing a lot of tasks like, you know, that you expect for a human. So just like your said, like the documents will disappear, everything kind of digital. Well now in the future you can just say stuff like that and expect them to happen. Like, hey, you know, looking at this invoice, can you get like, you a better deal from me for this one?

Yeah. And you know, then this AI will obviously understand this document, but also understands like how to use different websites like Amazon, Alibaba, wherever you source your stuff and then get like, you know, your better deals. So in order to do that, you need the first step is here the processing the documents and all details. So it gives AI to details that it’s need, it needs to find, oh, you are looking for this kind of a cable, this kind of interesting, you know, claim this kind of an insurance that you wanna find better. So I don’t find ever an end when it comes to how to make things more efficient Yeah. And faster and you know, more comprehensively. So today our customers they get it cutting edge because they’re using a cutting edge technology. Yeah. Too many years later. Cutting edge might be entire different thing, entire different meaning to it. You know, ob obviously AI is one thing that everybody’s talking about. Quantum computers are coming, that’s another, you know, all big thing that’s going to change about computation and like how, how how machine’s gonna work. Yeah. With all those things. it is really, really getting difficult to predict <laugh> where we gonna be in 10 years or 20 years from now.

CW: No, that’s right. As a, as a recovering physicist, I’m not, I’m not so sure the quantum computers are coming so fast, but I, I’d be excited if they did for sure. you mentioned a really interesting point, which I, I hadn’t thought of, but you know, like, so my, I have a pixel phone, right? And certain things I can just say, Hey, call this restaurant and get me a table, right? Mm-hmm. <affirmative>. So you’re, you’re describing a future where the insurance, like quoting process is so fast that essentially I’m just sort of dreaming here. The broker becomes like kayak for specialty insurance, right? Like, here’s all the documents, go out and get all the quotes. It happens in a matter of minutes, right? And not days, and therefore it doesn’t have to be this intense personal process, right? You, the broker process is dramatically simplified. I would think.

OB: We actually see this right now with health insurance, so interesting companies, users through this summary of benefits and coverage form that every insurance company has to provide now. Like, you know, the brokerage became like really a numbers game. So this policy is this much and gives you 20%, you know, coinsurance. The other one says $50 per visit, which one do you want? Like, it can show this in a very standard format. And that’s like, you know, the user, which one do you wanna pay for? Like, do you wanna pay for, you know, $500 insurance that gives you that or $2,000 insurance, like, gives you other benefits. So this is how the future of insurance will be and should be too. So the, the more transparency will bring better competition and better services too. So it’s not about like, you know, race to the bottom necessarily.

It will, you know, it will help the companies to it will motivate them to innovate about like new kinds of services. If you think like, you know, everything about the insurance has been done, then, you know. Yeah. That’s, that’s a very different question. I agree with that. There’s so much to do and especially with this new virtual task force, virtual ai, the white collar AI is coming into the picture. Yeah. now, like you’ll have virtual unlimited workforce to build all the things that you imagined once, but didn’t, never had the budget to do it.

CW: Yeah. Yeah. White collar ai. I am stealing that boon. That’s great. I love it. <laugh> good. all right, so we, we’ve looked into the far future. we’ve talked about the recent past. What, what do you see as the next, like what’s the next hurdle that you want base 64 to get over technologically?

OB: So right now we are pretty good in structured and semi-structured documents. We can read them off the box. We have already models for a thousand of them. my understanding is like, you know, we need to build another thousand document types for the remaining like, you know, the second tier important documents. And when we built that, we will be the one stop shop for air document processing. So this scaling beyond without obviously breaking the current things are, are the, the challenge, the, and from the technology perspective, you or the roadmap, we know how to get there. And this is like, you know, basically how we can execute on that roadmap question. Yeah. The other thing is I want to bring this technology, not just for the digital documents, but also paper documents. Hmm. So there’s a lot of paper still floating around as we spoke.

I receive, I still received letters, <laugh>, right? Yeah. That I need to process and understand. This is why we partnered with scanner companies and we integrated our AI into those scanners. Now those scanners, they become smart scanners. Now they know what they scan a scanner from A to B. There isn’t really that much of a difference. They all can do pretty much every exact same operations. They, you know, scanners the pdf, they can email it to us. But now scanners with basic Sephora ai, which by the way it is plugging into pretty much every scanner the moment you put in the file into that physical file into the scanner, the paper, it tells you, oh, you’re scanning an invoice. And it can also say like, you know, Hey Chris, you asked me to, you know, put your invoices into QuickBooks. I’m doing this right now for you.

So not just like scanning, but also processing automatically. So this goes to every industrial, it’s for insurance claims, right? Yeah. For section letters. How do you process them today? So you can do this by putting these files into here and like, you know, getting them into right channels and distributed or maybe like immediate answer as in like, you know, processed for those kind of documents. So this is the future of a 64 ai like end-to-end connected integrated system where you don’t need to worry about you know, doing this mundane tasks yourself. And then for everything else, <inaudible> data for unstructured very complex pol documents such as policies. we are, we are looking at you to, to build the coolest solutions. Yeah.

CW: Cool. We’re looking right back at you. We’re ready. so I mean, you’re really talking about document processing at the edge, right? That that’s what you’re talking about. yeah. How are you gonna get, I, I imagine your financial service customers are terrified of this. How do you get them comfortable with the security on that?

OB: So, couple things. let me tell you something that doesn’t work. <laugh> first and I, I go like, what worked telling, like, you know, hey, we have a great team. You have a big company, we are this and that, or I worked at Microsoft. Those are not great answers. <laugh>, this is not what security is <laugh>, this is the none of ’em really guarantee that there’s a secure solution. There are however other guarantees that we can actually offer. So first of all, certifications, we are stuck to hipaa, GDPR certified, and we get re-certified every year by independent authorities who are doing this for, you know, like CPA firms and so forth. they’re, they’re credible agencies. They come and look at literally everything we do from our financial statements to how we check in code, new code to how we hire and fire people, what are the like, you know, steps we need to do.

So they look at really every single detail. In fact, they even look at like, you know, how we assess our vendors and how we reassess them, <laugh> every like, you know, six months to a year. So there’s, there are really thorough investigations that we open up our books for them to come in and check us, and they, in turn, they give us this gold seal of approval saying like, you’re doing good. and let’s suppose like, you know, you don’t still trust anybody in this world. We have the option to deploy our AI on-premises. So in this on-premises approach, you can give us a computer where we can install our AI in there, and every processing will be done on-prem, on your system, on your service, without any data going out.

CW: So your smart scanner is on the same network as the base 64 tech that it’s talking to. That’s what you’re saying.

OB: Everything is like, you know, connected to each other and your office or your data center, however you wanna do it. Nice. So you feel this and not is out. So obviously then it’s, you know, the most ultimate, the most secure possible approach. you may also a second. Why, so why is not everybody doing that? Well then they have to obviously, you know, provide the hardware. They have to have teams to understand how to manage, you know, IT and so forth in the cloud solution, which is equally secure. there is no need for like dealing with, you know, capacity planning. Like, I need hardware, you know, I need to buy those expansive gps. No, we all are doing this for themselves. Yeah. a third thing, a very critical aspect is data storage. So it’s typically like, you know, not that our AI is processing this.

The, the problem is like, if what happens to the document I sent to that? That’s the, that’s the question. So we address this question by not storing any data if they want to. So this means like, you know, every process, everything in AI and so forth, the service, everything happens in memory in rem. So when the computer request everything is gone, it’s not stored anywhere. And you know, because we don’t have any files, we cannot be breached <laugh>. Yeah. So for customers that want ultimate, the security we offered is on-prem and also in the cloud. so we don’t have to store anything about the document.

CW: Right on. Those are great answers. so I, I, we got here, I asked the question like, what’s the next technology hurdle for, for these industries, right? Healthcare, financial services, and then specifically insurance. What’s the hurdle that the buyer needs to get over? Like, what’s your advice for the folks out there, for example, in the insurance industry, that they have a mandate to automate, they have a mandate to get AI involved, and they’re just getting in their own way. What’s, what’s your best piece of advice for those folks?

OB: I’ll say find one process, hopefully like, you know, the important one. And then talk to us basically for AI or liquidate for that matter. Let’s talk to us and let us show you like, you know, what we can do. And start small means like, you know, don’t start with like 50 different processes. We can discover those processes. We can understand like, you know, what’s in there and we can, this is also very helpful to prioritization and finding the best iri, but we don’t recommend like, starting everything at once. We start, and then you started one process. When you see the results, I think actually the, the rest comes by itself. So we have our customers telling us, Hey, can you do this document type, can you do this process too? Instead of us trying to upsell our ai because they really see this is working and they really see the benefits of it. They, they compare with the doc, the, the processes they did not automate. So it is then like a no-brainer to expand the AI’s capabilities. Give them better and more jobs, more difficult jobs if you like. this is what I would [inaudible] like, you know, there’s come tried out. we are, we have sales teams and customer success teams that are waiting for a call and, you know, yeah, I will be happy to understand.

CW: Yeah, I, I told you I was gonna steal it. I didn’t think I’d do it this fast. But the white collar ai, right? Like when you hire that junior employee, the first thing they’re doing is just, you know, you’re an analyst at one of the big investment banks, you’re just looking at decks all day and getting details out of, you know, whatever proposals you’re seeing. Start with that. Treat the white collar ai like it’s a junior member of the team, and then give it more responsibilities as you trust it, right. It’s a great multi model.

OB: Yeah. I mean, our AI is right now like an intern to new grad level. Yes. So for any task that, you know, you will put in that level of person you can use r AI for, and obviously over the time our AI will like a human will do, get better and better and, and be able to handle more sophisticated tasks. Yeah. And at that point, the difference will be like, you know, for a human to get there will maybe like takes like 10 years if they’re really good at their job, if they really devote themselves in, well this AI will get there much faster. And once it gets there, every instance of this AI will actually get there. You know, you train a doctor, every person goes to third euro, you know, hustling. with AI doctors for example, <laugh>, there will be only one training. So that’s, that’s will be like very, very efficient. They don’t need to retrain every person they hire once will be good enough for the entire division and company.

CW: Yeah, that’s a great point. And continuing the analogy, I think chat, G P T is sort of like a director level employee, white collar ai, but it likes to hallucinate and it also will occasionally lie to you very confidently.

OB: <laugh> <laugh>. Well, I mean this is a very different thing that they’re doing. There are,

CW: Yeah.

OB: So there are chat bots. So they generate words you know, you give them a topic, then they just like, you know, keep talking. So it is likes with different kind of personnel, different kind of person. They can build paint and they can paint stuff. They can write blog posts. However, in document processing especially, you need a very different kind of person. I need a kind of ai, this AI should be analytical thinking, should be like, you know, very attentive to detail it. You know, numbers digits really matter here. So you may want, you won’t want, you want with someone with some AI with more like a, you know, met brains.

CW: That’s right. Yeah. You don’t want your accountant to be creative most of the time.

OB: Yeah. Mean I don’t need you to paint me something <laugh>. I wanna data correctly, accurately. That’s what I’m looking for. So, you know, from the technology perspective, this is also very different things. Those are, they are more like generative AI is called. Right? absolutely. We utilize generative AI as well. So we do, for example, signature data extraction. So we can find the signatures on a document regardless on which page. And in fact, we can compare it against other signatures. Let’s say your check versus your driver license. These are available today, database 64 air. with that, we want, when we extract the check, we actually clean it up. So we remove like, you know, any lines or so forth under the hood. So we have a good picture of the signature. and find like, you know, the, the, the actual signature. Not like, you know, the, the documents templates and you know, all those lines that might come from the document or the letters, you know, that or which they signed. so those are generative AI that we use, like same technology that powers GP teacher because we’ve used in a very, very different methods or ways.

CW: That’s awesome. Well, I think this just about wraps us. We’ve covered a lot of ground today. This has been another episode of Unstructured Unlocked. My guest today has been Ozone bgan from Bay 64 ai. He’s the c e O and the founder. And he’s got a really compelling vision of the future of Aon, the enterprise. Thanks Ozon.

OB: Thank you. Thanks for hosting me.

CW: Absolutely. And go. If you’re listening, go check out their website, see what they can do for you. See you next time. Bye <laugh>. Bye.

Check out the full Unstructured Unlocked podcast on your favorite platform, including:

Subscribe to our LinkedIn newsletter.

Get started with Indico

Schedule
1-1 Demo

Resources

Blog

Gain insights from experts in automation, data, machine learning, and digital transformation.

Unstructured Unlocked

Enterprise leaders discuss how to unlock value from unstructured data.

YouTube Channel

Check out our YouTube channel to see clips from our podcast and more.