AI Benefactor Interview Series: Kris Subramanian, Option3

Co-founder on building an AI and automation company

Add bookmark
Seth Adler
Seth Adler
07/15/2018

One of the founders of Option3, Kris Subramanian, shares the thinking behind the creation of an AI-driven organization and how they built it up to what it is today...

Seth: Kris, thanks so much for joining us. Option3, obviously, is one of the AI and Intelligent Automation Network's benefactors, so you have been with us since the beginning, so we appreciate that. For folks that are new to Option3, just give us a sense of, you know, you said you're one of the co-founders. Why did we set up Option3? What was the original mission, and how, maybe, has that changed?

Kris: Yeah. Sure. So when all six of us had all started this journey a few years ago, we all have worked in the industry many years. In fact, a couple of decades. And we realized we wanted to do something in the automation space, and obviously, we came from different backgrounds. Some of us came from computer industry, some of us from banking, some of us form pure engineering. So automation was something that was very close to our hearts. So with that being said, we want to have a full product, which is going to write automated test cases without having to code for it. That was the vision, and then we will work on automation. We just, it's still continuously one of our flagship products, where we help customers automate their IT landscape by a few drag and drop activities, and build their intel automated fuel. That's how it started.

Seth: Okay. Yeah.

Kris: And also, in addition to that, we were very fascinated with machine learning and data. That was the second key objective that we had to make any of these platforms intelligent, so we built an augmented platform, which could do data utilization, data streaming, and act on analytics. Then our company was born, as a natural progression of it, because we saw a very strong application of our product into the business process, and not just the IT process side of things. So that's when we came into existence.

Seth: There we go. One thing led to another. I know that you at Option3 have a philosophy, which is automate, analyze, accelerate. Can you take us through that, to make sure that we understand it from your perspective?

Kris: Yes. What we think is, even if you have to...the philosophy of our work is even if you have to do it once, can you automate it? That's how we push ourselves and people say that you know they're doing something 4 or 5 times automated but we say even if you want to do it once, we can do it quickly with automation. That's the automate part. When we're doing something even once or twice, we try to automate it so that's how we go about it.

Seth: Right.

Kris: The analyze part is predominantly, you know, how we play a lot of intelligence into our commission that will be analyzed...aspect of the product that we built in. And finally accelerate part, we wanted automation to be continuous so  it would only be a platform to engine build. It's a process, automation, but once you are on the journey you can quickly accelerate your automation.

Seth: Okay. And accelerate kind of with automated solutions?

Kris: Absolutely, yes. We do solutions not just for automation but even for analysis.

Seth: Understood. Okay. Let's make sure that I & we understand your view of the various parts of the globe as you know, full scale enterprises start to truly you know, accelerate their journeys in intelligent automation. I wonder how regions around the world compare in your point of view and so ill ask you, so we've the AI Live APAC event and that'll be followed by an event for India, that'll be followed by an event for the Americas. If you could take each of those 3 segments and share your point of view of where folks are. So what does APAC look like for you as far as enterprises? You know, on the journey of intelligent automation.

Kris: So APAC again is a very tireless job but it provides security when it comes to automation so I'll kind of further segment... if we look at Southeast Asia's geography, they allow automation so it's been a very easy journey for us too because they know that they want things to be automated. Automation takes time and machine learning is a topic that very likely will need a system so that's been an area where we've been able to progress a lot in terms of computer automation. Now other parts of...if you look at India and the geography. Obviously the powerhouse of the business philosophy. Now when you look at India's region obviously you have the big players that are fully implementing quite a few schedules by office proxy whether that be insurance, banking, finance and accounting and et cetera. So I mean we have again different people in different stages of their journeys but what we kind of have customers do is to not look at this as two different pieces. So instead of doing this as two phases, what we are helping customers do in this geography is to take a hybrid approach to our commission.

It's the same process, depending on proofs, that can be done on a role based approach. For example, take processing your top 10 layers you know based on rules and process them within couple of weeks. But the remaining 90% of layers go through the cognitive process in the same process. So you're not doing it twice, you're doing it in the one shot where it takes 2 layers. One process to try to automate it using the rules and if it doesn't work the same document will slow cognitive roles. So that's something we've been able to achieve for them.

Seth: Great. For those folks listening that were earlier on in their journeys, you've potentially got an advantage over early adopters who did it one piece at a time as you've just shared. So that's good to hear that that opportunity is in front of us. That was part of your answer for APAC, it sounds like that's more of a holistic kind of answer for all of the folks that you're speaking with. But how does a MIA compare with APAC as far as you know the intelligent automation journey for enterprise?

Kris: Sure. What we find largest for our customers is based out of MIA and we have been on a journey with them for the last 24 months or so. Seth: Mm-hmm (affirmative) Kris: And we have seen them, there's some testing waters now to their most advanced customer using automation and its the same theme that we have been working with. And we have seen them progress from a little company to over 80 countries and then now when we look at them they say hey, now we can do this cognitive rather than rule based. But we have seen the beauty of change control that cognitive system and that's what we are seeing more and more requests coming our way when people realize that the cognitive automation is the way to go. Again I think its a lot more matured especially when we look at the MIA market and they've been in this journey for the last 2 3 years so there's a lot of maturity in that market.

Seth: When you mentioned that client - you may not be able to mention the client by name and we certainly are not asking for proprietary information but what can you share with us in terms of those cognitive solutions that they are employing? When that say to you "Hey, how can we employ a cognitive solution?" What are those solutions? What are they doing? How much of that can you unpack for us?

Kris: The first is to process unstructured data. So the there is a lot of data coming in because there are no departments which are doing finance and accounting, you know insurance and marketing, telecom operators. A lot of documentation involved. We have seen this evolve tremendously when it comes to cognitive automation, whether it is processing semi structured or unstructured data. There we apply machine learning as the predominant methodology and one of the challenges people have been struggling to do properly with machine learning is to slip on a functioning machine learning platform rate. So obviously there are you know large cloud based platforms available now but the challenge with that is how you take 30-40 documents that come in an email and run it through your cognitive platform and get your data back. Particularly the response time and the commercial aspect. We have been able to, because we are cognitive product...so part of it is completely seamless approach where 30 documents come in and you extract the documents or the items required and then run a cognitive engine on top of it and any exceptions flag on the screen that moves onto a phase 2 process all within 15 mins.

So backups really made our idea seamless from an operational standpoint, all they see is that there are 100 invoices that came in the last 1 hour. They know that 20 of them got processed through machine learning, and the remaining 80 of them are processed through and maybe 15% had an exception and they need to go and fix it. That's one aspect.

Seth: As far as the machine learning that you're putting on top of that OCR, what kind of insights are coming out of that machine learning layer over the OCR? What can you share?

Kris: If you look at the OCR and it has a very specific role to play and that is to extract detail. 

Seth: Right.

Kris: But when you look at the real acclaimed document or you know a bank statement or an audit statement, you are dealing with very complex data where you're talking about tables, sub tables, and so understanding that requires a lot of intelligence when you try to build something. But what we have come up is a feature that is dynamic templating- the machine learning is able to understand without training the bot. So that has helped our main customers, I don't know maybe 30% of the documents will be flagged as unreadable not because of the quality of the document but because of the subject of the document.

Seth: Mm-hmm (affirmative).

Kris: So those have come down from 30-40% to 10-12% so that's where we've seen machine learning kind of able to take down the document max.

Seth: That's huge. That's just huge, you know, because apply whatever you're doing and the money affected to those percentages and you're talking about a lot of money in many enterprise cases. So thank you for diving in on that part of what we were answering with the MIA and hoe basically what you're seeing is the MIA region is more mature than the APAC region and that's certainly reads true. How about the Americas? How do they measure up to the MIA, APAC or both?

Kris: I mean, we have been working with predominantly enterprises so most of them have had a pay stub in the last many months and I mean they are patients who hae not seen a doctor many times. So again I think you're talking about whatever it is when it comes to insurance processing and some of the most powerful AI engines have told that you need the documents to be a lot more structured for them to read. so insurance as a domain is...you take out of an area which is legal right? So that's where we made some real interesting use. And I kind of you know, you take the legal scenario where you have a contract getting signed. Now the final question- How do you verify that the final version is actually what you signed up for? This is a classic case where you are applying a classic language property because once you open the document you are seeing lines of text with no context. So we need to see and highlight the critical things that might have happened after the document was finalized and before you sign. And surprisingly there are quite a few cases where they have and it's quite good because you get that turnaround information in a few minutes and you...it's not just about machines, its about identifying critical data. In passage, we identify after 2-3 years and you get into a litigation so.

Seth: Yeah. Alright so natural language processing obviously we're seeing that now and you're seeing that more...when do we get to what you guys refer to as pure cognitive? What do you mean by that?

Kris: For us, there are 2 aspects of cognitive. One is ability to process data, so machine learning and natural language processing. And I don't know, keep learning to some extent but it's an aspect that needs to be a little more dived into. Once we are learning. Now the 2nd aspect when it comes to RPA is decision making part of it. That's again cognitive nature for us. I'll give a very simple example for a large group of manufacturers who run a fantastic group of ER businesses across multiple countries. someone figures out if those warehouses actually receive those goods. In fact one of my invoices always looks that way because we can't figure out which department...so this is a classic case where we have applied cognitive ability to scan through millions of records and say these are the potential massive coming your way. So instead of going through 100,000 requests or 1000s requests and coming at the physician the- this is through cognitive recognition at a space where you know you're enabling and looking at data and reformatting an understanding, you're making a cognitive decision. Same applies to anything we execute

Seth: Yeah. What I'm hearing, and correct me if I'm wrong, is that no matter how much cognitive ability is happening from a machine perspective, it is outputting a dashboard, so to speak, for a human to make that final decision?

Kris: Absolutely.

Seth: Alright. As we go here, when you look to 2019, what do you see as the future or what happens next? In terms of intelligent automation and artificial intelligence as it pertains to corporate enterprise.

Kris: So I think you know, I mean a lot of progress has been made when it comes to cognitive capability in enterprise and their business runs on a lot of AI based reporting and analytics. Now the attempt is to apply that to data processing our business operation which is very manual. What we see is that there is going to be a very tight integration. Need for data to get into a cognitive engine whether it is a regular business operation or something specifically done using an analytics data platform. Now where we see this mud is all these channels of information will get consolidated and it is truly going to be...a lot of these things if we were talking 12 months ago about AI and say we are now looking at cognitive where machine learning based decision making is being enabled by us. It's kind of cutting across boundaries, it's not just data processing. It's also moving towards decision making.

Seth: Yeah.

Kris: So rather than losing...in the context of it I will talk about one particular usage we're doing right now. Insurance companies, they have huge wells of storing data- when somebody comes up with a historical claim which is not in your real time storing system, you talk to an operator, he runs a query, he gets this information. It takes 4-5 days until you get to the bottom of that query. Because of our cognitive platform, which is again a data storage platform actually, it is not difficult. So those archives can be part of a solution and a query can be processed by a bot and you can get answers within a couple of hours or within the day. So those are usages that are going to save a lot of time because some are responding to the ability to use on our time becomes very high. So this is a classic case of an archive retrieved and processed. Now people are open to look at it from an RPA standpoint where you store the data in a big cognitive platform and it'll extract that because you know it can run through millions of files in a couple of seconds. The storage is unlimited because of the technology that we use that will not require data stored in expensive mechanics.

Seth: Interesting. Just dive in on that so we're sure to understand what you mean. As far as storage...

Kris: So basically when you look at cognitive automation, you're looking at running your request against a million records and try to match it against a million files. If you use expensive storage, the storage itself is going to cost you a lot and running a query is going to cost a lot of money. Whereas our mechanics is based on a cognitive platform which is very inexpensive, it'll cost you as much as a hard disk but at the level of billions and things like that.

Which means your quantum query is the same as your random query or normal query. So it becomes very easy to think of RPM or historical data processing.

Seth: Right. Well, Kris, I and we really appreciate your time here. Is there anything you haven't shared what we absolutely must know?

Kris: So from where we are looking at from an RPM intelligence process automation journey one big goal we have set is to make cognitive automation simple for normal users. Because you know when we go to events, there's a bit of reluctance among people, there's a bit of a fear based on more the complex. Now we want to kind of break this and challenge people to say that though you are going to automate that process we will do the same using cognitive technology in the same time period. And we have done that specifically for a few customers and so now we want to make this available to the larger user community where you know using machine learning algorithms is not that complex because you don't pay them, you use them. How do you make this available to a set of developers who can easily use them, build them, and innovate a cognitive bot is going to be what we are going to focus on in the next 6-12 months. And you will see a lot of things we will announce in the coming days to make this happen.

Seth: Fantastic.Kris, thank you so much for your time and I'm looking forward to checking in with you as we go.  

Kris: Sure. Thank you!

Stay up to date with AiiA.Net to hear more insight from Option3.

Sponsored By:

RECOMMENDED