Webinar replay: How carriers are leveraging large language models (LLMs) and automation to drive better decisions
Watch Now
  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 Peter Camp

Watch Christopher M. Wells, Ph. D., Indico VP of Research and Development, and Peter Camp, CTO & Founder, CampTek Software 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 10 with Peter Camp

 

Christopher Wells: Hi, and welcome to another episode of Unstructured Unlocked. I am your host, Chris Wells, VP of research and Development at Indico Data. And my guest today is Peter Camp of CampTek Software. CTO and founder. Peter, how are you today?

Peter Camp: Good. How are you doing, Chris?

CW: Good, I’m good. I’m looking forward to this conversation. to get started, why don’t you tell us a little bit about who you are and then what you all do over at Camp Tech?

PC: Sure. That would be great. Yeah. So I’m Peter Camp. I’m the founder, CTO of CampTek Software. I’ve been in automation for the better part of 20 years doing RPA development. previously worked at a company where built a r p a product from the ground up similar to UI paths. So I know all the trip tips and tricks and, you know, the strengths and weaknesses of the tech. my company, Campex software has been around for almost five years. we approached this a little differently than some of the other leading RPA vendors out there. service providers, we’ve tackled this from a managed service perspective from the beginning, cuz I think that’s the best way to handle this. and, and support. So we really build all our solutions with a focus on support from beginning to end. So when we scope something, we build it, and we have a methodology in and around that. so, you know, we’ve had quite a bit of success in, you know, healthcare, supply chain financial services even retail started to pick up a little bit lately. So, you know, we’re really we’re closely aligned with UiPath and obviously, we’re partners of Indico as well. So

CW: Yeah. Let that cat outta the bag right away. Yeah. Strong partnership. Watch this space, as they say, for good things coming. great, thank you. thank you for giving us that rundown. Peter, it sounds like, if I’m hearing all of that correctly, you’re really deep and wide when it comes to RPA and all things automation. Is that fair?

PC: Yeah. Yeah. So I mean, one of the things we tell our clients is we kind of have, we have a best-of-breed approach providing solutions. So, you know, in the case of like INDICO versus, you know, some of the other document understanding tools, every, you know, Indico does its job really well with certain types of unstructured data, where some, where you may not need it for something’s more highly structured or there’s kind of a good example there. it’s the same thing with the RPA approach or even solutions or activities you call within our RPA bots. We always try to figure out the best and cheapest and highest ROI solution set.

CW: Yeah, absolutely. And that’s a theme of this podcast right tool for the job. And you, you guys certainly have the expertise to know how to choose that. So it’s gonna be great to dig in. Why don’t you start by, I’m really interested in the introduction. You talked about seeing, you know, offering RPA as a managed service, to begin with. What is it that convinced you that over your, you know, long career of doing this?

PC: Yeah, and that’s a great question. So like I said, I’ve been in this space for about 20-something years, but originally I worked for a legacy r p A company. And when I first got hired there, they, they’re doing a lot of trainings, you know, training people how to build, you know, scripts or what we’re calling bots these days. Yep. as time went on, less and less of those people on the client side were wanting to take the training or were even available or didn’t have the right background. so customers came to us to help them support these bots that are already in production and, you know, so it kind of became more of a services approach versus just a straight technology software approach. And, you know, as, as RPAs evolved, you know, with UiPath and others, really taking this to scale at the enterprise you know, saw that in like 2017 18, that if this thing’s really gonna, you know, scale from a wide level, you have to have support, you need to have managed services around it because companies can’t hire quickly enough to keep up with tech, not to mention there’s a million different pieces or tech that you can use. I mean, indico, you can use Google, you can use Microsoft now chat, G P T coming up. Yeah. I mean, it’s on and on. I mean, there’s no way anyone can keep ahead of that curve. So you need to have managed services companies where their job is to keep ahead of the curve, understand what’s good, you know, kind of figure out which, which direction to go based on, you know, the need.

CW: That’s great. There are a lot of areas in there that you covered. So, you know, I can think of sort of just day-to-day operations oversight, qa. What’s the spectrum of services that you offer and that your clients find valuable?

PC: Yeah, I mean, well on the whole delivery front, we can do everything from, you know, okay, coping identification, you know, onto development, testing, QA of everything, support is really where all our bots are at this point. The majority of our bots are okay. Try to get it out of the development to support. so whether it be an IP solution or a straight back office unattended robot or attended robot, you know, those things go to support have, that’s basically step one.

CW: Great. Okay. All right. So you talked about the proliferation of tools and shiny new toys these days. Yeah. Like chatGPT. Tell me one of the things that have become critical for really getting the most out of your automation solution is finding a good IDP solution or cobbling one together from a number of different tools. so talk to me a little bit about how you and camp tech software think about unstructured data. How is it different? How is it the same? What are unique challenges?

PC: Yeah, so we think of it in three different ways. Like most do structured, semi-structured, and then finally unstructured. so structured is what I would call, you know, legacy ocr r techniques, positional, you know, data doesn’t move fixed form, you know, frankly, there’s free OCR r tools out there that can do the job at this point. even those have gotten better, you know, 10 years ago they, they had a hard time just reading text. Yeah. Now it’s, it’s highly sophisticated even on over the shelf. so the semi-structured is, is really where a majority of the use cases traditionally have come from, in my opinion. Like Abby has done a lot of work in semi-structured Yeah. Un passes a lot of work in semi-structured. And it is really become very fuzzy. And, it becomes hard for clients to understand the difference between semi-structured and completely unstructured. And, I’ll even go a step further to say it’s not just unstructured. You know, this could, this piece of data could be anywhere on the screen, but is it, is it unstructured need based on the volume? So like, if you have semi-structured and you only have, you know, I don’t know, let’s say you’re processing a hundred a month, you can, you can get that limping, you can limp those through, you know with, with back backend support. But if you’re talking like millions of documents a year, or millions of pages a year you know, 7, 10, 12, you need something that can handle the level of volume and, and handle anything that comes through. So you, you could be majority semi-structured, you know, 60% semi-structured, but it’s the 40% that will kill you. That’s not, yeah. You know, and, and if you’re, if you’re pulling in a lot of different sources you know, this use case we’re working on with you guys, I mean, they have thousands of physician offices pouring in these orders from all different states. And, you know, they don’t know. They may have worked with a doctor one time. They may have worked with a doctor a hundred times. They may work with, you know, it, so the repeatability isn’t really there, but the, need for the discreet pieces of data are.

CW: Yeah. You raise an interesting point, which is even if the, even if one document is sort of, you know, semi-structured and the next one is semi-structured, it’s slightly different. If you’ve got millions of documents, millions of pages, like even getting your arms around the problem initially to define it in a semi-structured way, you know that that’s months or years of work, right?

PC: Yeah. Yeah. So I mean, you know, you look at it strategically for us being advantaged service, so obviously wanna make money on support, right? Yeah. we wanna get people to support us, but we also need to make money too. We don’t wanna spend an infinite amount of resources supporting an IDP initiative because of the retraining or training of the models. So I think Indico, to me, has the best solution, best go to market at volume of high volumes where you don’t, you know, you basically build it, correct it, and it’s, it’s continuing to learn on its own, but the benchmark where it starts is a lot higher from a reliability standpoint than any tool that we’ve seen.

CW: It’s an amazing compliment. I happen to agree with you <laugh>. and I’ll, I’ll encourage the folks out there who are curious to either talk to you or me about what’s possible coming back to that point of what is possible what do you see unlocking access to unstructured data potentially doing for organizations and for, you know, the knowledge workers within those organizations?

PC: Yeah, I mean, obviously, you know, the value is so high if you really think about taking all these digitized assets, assets and, and making them understandable and not having that middle swivel chair person processing them. Hmm. Both from a labor standpoint, and I think from an accuracy standpoint as well as a speed to business and speed to decisions. You know, if we think about this, I, I’m actually in the middle of a blog right now thinking about what it would look like, you know, in 10 years, truly having unstructured data solutions that could really handle a lot of these review situations. I mean, we work with customers that are having to review faxes, you know, hundreds and thousands of pages of faxes a day, and they have a team of people doing this, reviewing that they get this information. A lot of the data is exactly the same, but it needs to be put into a system exactly the same way.

You know we’re working that with you guys doing physician orders. There’s a team of people that are having to look at each one of these orders every single day, look through and make sure this information is not only reflected in the order, then also in the system of record. And they have to go through two different steps to do that. And that is a very, very common use case. But I think regardless of what industry you’re in, financial services, supply chain, healthcare, wherever there’s that middle man or middle person that’s having to process this stuff, and, and it can’t be just done automatically without some type of logic behind it. so it’s not just simply grabbing the data and putting it someplace, but actually making some type of decision to reflect that, it is accurate.

CW: Yeah. I love that. I love that framing of speed to decision. And I also, cuz we’re, we’re seeing the same thing right now in the insurance industry broadly, just whether it’s claims or underwriting, you know, quote submissions,

PC: Right?

CW: The person that gets, you know, the company that gets that data out the fastest and to a point where you can look at it and say, yep. Quote it, Nope, deny the claim. they’re the ones that get paid first. And so I like thinking about that 10-year future, it’s kind of an arms race, right? The folks that get there first and the fast followers have an advantage, there’s some real potential for companies to be left behind in that.

PC: Oh, yeah, absolutely. I mean, I’d say I think not only with I D P, but you know, even r p A and, and implementing, I’d say bread and butter, rpa, you know, bots to just alleviate some of the work. Yeah. let’s not talk about complete digital transformation just yet. But even that, just getting that off the ground, I think would make a huge impact on the majority of companies. And, you know, we’ve seen, you know, per, from 2018 at 23 you know, a real evolution in this market. you know, high, high co o e usage in 1819, companies building their own, you know, practices and Yeah. You know, and then 2020 kind of really scrambled that all up. It’s budgets got constrained, you know, and the bots that were built in 18 and 19, whether it be a solution provider coming in or a c OE or turnover, a lot of these bots are like on the shelf right now

CW: Mm-hmm.

PC: Huge opportunity to go back and, you know, fix ’em or get people realigned. I mean, we have several new customers coming online just with that particular problem. I think there’s a lot more out there. with that being said, I think, you know, as we move forward because I d p in 2018 and 19 is far different than it is right now. I mean, yeah, as you said, it’s an arms race. Everyone’s really, you know, there’s a lot of value, I think a lot of revenue coming into this particular space. and companies that may have evaluated solutions, you know, in 18, 19, now they come back in 23 and it’s like, gosh, well I didn’t realize this was out there. So that’s why we’re advocates of what you guys are doing and what you guys are trying to do. cause we feel like it is the best of breed.

CW: Awesome. Yeah. I would say that 18, 19 era, it was five to 10 to one hype to actual, you know Yeah. Meat and sausage when it came to those technologies. And so a lot of people got burned yeah. You know, by vaporware. And now, like I said, I do think we’re the best, but there’s a, there’s a lot of really good stuff out there. And one of the key things is finding someone who’s done this before and the right technology to do it with. And so that, you know, that makes me super pumped about this partnership.

PC: Yeah. Yeah. And I think if you look at the strength that UiPath brought to the market, the reliability. Yeah. you know, and it’s really not, a lot of times it’s not, it’s not, I’d say 99.5% of the time, reliability isn’t unreliability is not a result of the, of the technology. It’s more around the implementation, identification of good use cases, you know, proper support, maintenance change management.

Change management, yes. And all that. It’s not, it’s, the technology is definitely there and there’s a lot of different ways if it’s not there to improve it, you know, in a custom way too. So, you know, we, we feel really strongly that the go to market with the technologies that are out there right now can make a lot of this realistic versus, you know, before 10 years ago, it really wasn’t realistic at to the scale that was, you know, needed.

CW: No, yeah. It was science fiction. The only companies that had any technology like this were using it for themselves. And they’re probably building it into products like Google search and Facebook. Right. It was all locked up. I wanna lo I want to keep looking at the future a little bit. so 10 years from now, I don’t wanna steal the thunder from your blog post, but it’s an interesting topic. so we’ve gone through this arms race, you know, some people have fallen along the way. now everyone’s really sort of fast to that decision point. so then what’s the bottleneck from there? if decision-ready data is not the bottleneck.

PC: Yeah, that’s a good question. I think, well, the bottleneck is actually transforming that data into something meaningful and you make decisions off of versus, cause I mean, you guys, that’s probably gonna, the next evolution of your product is to take the data you, you’ve already done intelligently and reliably, but then actually provide decision trees or models or, you know, so if you say you’re going through, I don’t know, I think there’s a huge opportunity if you’re cataloging this data, pulling this data, actually doing something with the data versus just, you know, spitting it out in a J S O and having a bot go de plug it into co into an application, but actually mining the data and, and creating, you know, understandable models that, you know, things could fall out of that. So like, you know, if you see, like in the insurance case, and they’re probably already doing this, most sophisticated insurance companies probably do this already, but like, you know, taking all, like say there was a car accident, like insurance and, you know, taking all these discrete parts of that data and then figuring out, okay, this is actually someone who’s under the age of 25 that drives the Jeep, they’re unmarried, you know?

Yep. Whatever. maybe we don’t, I don’t know, whatever the decision tree falling out of that is, you know, and like yeah. Figuring out what to do with that. And they, and I know a lot of companies already have these models, but I think the thing is they’ve not had the discrete data you can pull from these documents. I mean, frankly, I also, one of the things I wrote in my blog too is imagined if you could take all of a company’s, you know, their entire lifecycle of every printed document, scanned document, and you collected all that data and made it into something. I mean, you would have these supercharged companies. I mean, they’ve got all this like, you know, digital debt sitting there. Yeah. So, you know, I think that’s really where, where it’s gonna go. because I think a, as people are looking for more advanced solutions or more decision making pre-human, you know, they’re just gonna be looking for data however they can get it.

CW: Yeah. I couldn’t agree more and don’t tell anyone, but I think unstructured analytics is the future of working with unstructured data. Like, you know, to your point, all that digital debt sitting around, it’s kinda like crude oil sitting in the ground, right? Like, and to date, we’ve built drills and we’ve built some pipes to get it out of the ground and move it around. And there’s a lot more of that to do <laugh>, by the way. yeah. We’re not gonna go, we’re not gonna be bored anytime soon. Yeah. But the future is really refinement. Right. And get it into, you know, from Rod Bites in a PDF to a, a real you know, usable asset on the other side.

PC: Yeah. Yeah. I mean, first, you just gotta get the piping in there first, you know, and then, and then we can start figuring out best ways to use these, and the, and the analytics tools are gonna get better. I mean, they were already on a lot better, you know, where you don’t, you don’t need a full data scientist to create a machine learning model. I mean, I could do it, and not to say that I’m not sophisticated, but I’m not a data scientist or anything like that. So,

CW: Yeah, that’s a great point. Coming back to that idea of just the digital debt that companies are, some companies are drowning in right now. How do you think about that debt? Like what is the cost? What’s the risk of just maintaining unmanaged, unstructured content as, as we’re doing it now, sort of status quo?

PC: Well, you know, I’ll pull it into the analogy. I frequently use about IT and legacy systems. so they say an average ID department spends about 70% of its budget maintaining legacy systems. So paying all the support and all that. And then, and then there’s another implication for the organizations themselves having to use these legacy systems. So the problem just kind of balloons. So if you’re like working on a legacy system, then you have your new system, there’s a lot of, you know, going into the system to make sure everything’s cool and going into the other system. Yep. And I kind of look at, I kind of look at, at this as the same situation where if you don’t get a strategy for I D P or document understanding now it’s only gonna get worse because your company’s gonna grow and there’s gonna be more inbound.

And then you have that issue, and then, and then you’re, and then it falls down after that too, because it’s not being indexed properly. So, you know, if a doctor comes in and says, where’s this order? I, this was in here. I, or the Dr. May not even know there was an order and have to go place another order of the same exact order, you know? Yeah. And that kinda thing. So like, it’s, it’s really a problem that now that there’s a solution, there’s really no reason why people shouldn’t, or companies shouldn’t really adopt it.

CW: Yeah, a hundred percent. So we talked a little bit, so we’ve looked into the far future. We’ve looked a little bit into the recent past. Let’s dig a little bit more into the past. So talked about the tools getting better. You’ve talked about people being more educated about what’s possible. You’ve talked about the right approach, being managed service to handle all of the things that you’re not thinking of. When you think about, let’s just automate this part of this process, how have you seen the use cases that are being approached change in terms of, you know, whether it’s vertical or complexity or whatever it is as we’ve hit this sort of inflection point that we’re at?

PC: Yeah. Well, it really depends on where customers are on their automation journey. summer’s still back in the, oh, great, I can take the spreadsheet of data and put it into a system. You know, that’s, there’s still a lot of those out there. Yeah. and those are a lot of our, like, quote unquote legacy bots, I guess you could say. but it is getting a little bit, it’s getting more sophisticated, particularly the use of APIs even within our bots themselves to, you know, hook into like Salesforce or ServiceNow or, you know, I, I know even on the EMR standpoint, there’s fire, the new fire standard in healthcare is something that highly evaluating, and we have a partner that we’re working with on that. so it’s, it’s, it’s also too, we’ve seen uptick on attended automations, you know, kind of the human loop aspect, which, you know, IDP definitely falls right into as well.

Yeah. so we’re seeing quite a bit there. you know, we’re also seeing Yeah, a real, a real need to, okay, we have this problem you know, we’re having an issue onboarding off onboarding employees at scale, you know? Yeah. seeing that quite a bit where, you know, that was something that was certainly mentioned as a high value use case, but not really taken seriously enough because there was, you know, not a, the labor market wasn’t as tight as is right now. So like, we don’t have someone in that, from a compliance standpoint in, and any company is a problem. you know, as you try to maintain your credential or your licenses and things like that and your certifications, that has to be done quickly. so we’re seeing a bit more of that. you know, I also think companies are doing this at scale. Like they’re approaching this with, with mi with a mindset of scale. we have several customers that have said, we wanna automate 16 different departments within our organization that’s a 5 billion, six mil, billion dollar revenue company. Huge. And, you know, we’ve had other ones in the Fortune five, fortune 1000 realm doing the same thing. We need to get, we want to double our revenue, we want to get going here, but we’re not gonna be able to hire enough people to do this. So we need to come up with automation first strategy.

CW: Yeah. 

PC: We’re 18, 19, when we started this company, it was like, well, I think I have that one thing we wanna do, or I have three, you know, one to two different bots we wanna do. And, you know, I, I think UiPath and others, and Indico have done a really good job of marketing what’s possible. and letting people know, even Microsoft’s entry into, you know, the RPA space has been very helpful. Yes. For gathering awareness. So, several years ago, we talked to people in IT who had no idea what RPA was. Now they kind of know, oh, like power automate or whatever. So it’s it all, it’s all helpful.

CW: Yeah. Absolutely. let’s let’s see. There’s something in there that you said that I wanted to drill into. I’m gonna have to edit this part out.

What was it? It’s a great little nugget. Oh, yeah. You were talking about the difference between attended and unattended bots, and you were talking about how I D p sort of drives you to the attended side of things. my experience has been folks that come to indico mature in automation they often don’t understand the need to have a human in the loop. Not only the need, but the value of doing it. You know, the assumption is I put document in, everything comes out perfect. I never see it. but of course it almost never works that way for a lot of reasons. So how do you talk, how do you talk people off of that ledge of just, I thought it was gonna be a hundred percent straight through.

PC: Yeah. I mean, and I mean, with your verification station too, there’s that human loop element to this, but Yeah. Yeah. I think, I think what we, you know, we try to set the expectation to say, you know, if it’s indico or whatever we’re trying to process Yeah. Thought quote unquote is gonna be able to do, I don’t know, 80% of this right. Without, without even worrying about it. But there’s gonna be that 20% where, you know, you need to review what’s going on, and you, and it may be something that it’s simple click, oh, yes, this is a, this is a ID number, boom done bot panel, but it may, may require more intervention, like there’s missing information or the, the information that the bot or the indico was trying to get just isn’t completely there. Or maybe something that they not need to pick up the phone and call someone. Yeah. Or, you know, so like those kind of things, you have to be realistic about what is actually possible because it, there’s no turnkey solutions. I mean, even our bots that are unattended that run, they all create reports and there’s exceptions on all of them. Yeah.

CW: Okay.

PC: Just say once in a while. I’m not gonna say they’re all all there, but if there’s exceptions, you know, that someone will go look at the report after the bots run process them. So that’s a human loop scenario. absolutely. At at scale. If there’s a lot of ’em, then you build automation into help process those. But if it’s, you know, it’s just a handful, there’s no need to create a bot that does all that legwork.

CW: No, that’s right. And I think one of the things, one of the things that’s important to point out is that pure RPA bots don’t get smarter over time, but your document processing solutions, assuming they’re ML based, can, and so it’s really valuable to keep, as your data flows change, as your analysts change the way they do the job, it’s important to keep flowing that supervised data back into the model to keep it it’s almost like you know, ongoing on the job training, right. The bot is, you know, it’s mimicking that human knowledge worker, and you need to keep it up to speed on how things are going and what’s coming through.

PC: Yeah. And the bot, I mean, it does get more intelligent in a way that, you know, you get, it’s a feedback loop coming from the SMEs to say, Hey, Canec, we need this adjusted with the bot. We need to add these things. So it’s, it’s more of a manual intervention versus an automated intelligence. And frankly, and I know that UiPath and others are thinking of ways that they can, you know, have their bots peel or get more intelligent. But I’m a little weary of that because it’s, you know, you can’t just say, please go do this, and then you have it go make a bunch of, you know, business rules or whatever without some type of intervention. but I know, I know that it is coming. so we’ll just have to keep an eye on that one.

CW: Yeah. Have you tried the experiment? Like, Hey, chatGPT, write me a UI path bot that does X and Y <laugh>?

PC: No, but I know it’s possible

CW: <laugh>. Okay. Stay tuned for that. That’s exciting. 

PC: No, that’s stuff’s really exciting to me. but I still think any, any kind of auto-generated code always scares me to some degree. And granted, you know, everything is fairly componentized at this point and object oriented, but you still need to know kind of what it’s doing. you know, and, and we were talking about automating at scale with enterprise level systems, you have to be very careful. Yeah. You know, it’s no different than creating a power shelf script that can wipe out your entire active directory with like one, one back slash you know, like,

CW: So That’s right.

PC:  You have to, you have to be careful.

CW: Oh, yeah. No, it’s a great point. It’s, I mean, it’s the same problem as like a room full of devs and none of them are listening to what the architect is saying, right. Like, and chat GP t ain’t no architect at this point. Maybe someday.

PC: And it could be, I think there’s a lot of potential there. you know, particularly even from a code review standpoint, you know, it could certainly do quite a bit there. you know, it could also help, you know, developers that, you know, h how are they developed? Maybe they’re development cycles that kind of start at the higher end and then build the depth there. you know, but even from a reusability standpoint, if you’re like, I wanna use this library, boom. Yeah, please do that, then it just throws it in. So, you know, I think it is exciting. but like a lot of other technology takes a little while for it to take hold. But I mean, the fact Microsoft’s really making a pretty decent size bet on it, and they want to Yeah. Into all their applications, it’s only gonna get more sophisticated.

CW: No, yeah, you’re absolutely right. so, you know, yeah, go ahead.

PC: That’s also, that’s also, that’s also the reason why manage services are so important, you know, because Oh yeah. You know, you can, a company like ours can do that research figure, figure out what works, what doesn’t. You know, we, we do a ton of r and d behind the scenes you know, evaluating these technologies, working with them, trying to figure out, okay, these use cases we had before, how much better is this technology than what we were using? Or, you know, go out and try this new RPA platform, see how it compares. You know, so it’s like that just saves a lot of legwork and confusion for our customers.

CW: Interesting. All right. I wanna circle back to that, but while we’re talking about, you know, the right tools for the right, you know, project within an organization, how does Camp Tech go about helping your customers find the best candidates for a process or workflow or whatever it’s that could be automated?

PC: Yeah, so we have a few different approaches. we have the traditional interviewing SMEs, trying to figure out where the pain points of the business are, focusing on those areas. you know, doing scoping sessions, things like that. Having business analysis, figure out where pain points are in departments or across the business and, and focusing on those. That to us is, is one way of doing this. another way that we’ve really started to stand up within the last year or so is using task mining. So we’ll task mining agents on desktops. So we have a few organizations that are like, we do wanna automate, but we’re not really sure what we wanna automate and we wanna close data from our employees, but our employees are so busy they can’t sit with you for, you know, hours on end. And, you know, then it becomes very subjective and employees have a way of, you know, describing the process different than it probably is described.

And it’s, it’s, it’s all biased. in some ways task mining isn’t task mining will sit there, observe a group of people doing similar job function for, you know, a week or two weeks or however long it takes, and then it’ll provide, okay, this is where, these are the common tasks that people are doing. This is where we not only think there’s opportunities for automation, but this is how we’d actually automate this, you know, and coming up with those. Interesting. So like in the case of the joint use case you guys are working on, we actually did a task mining study with that group and found how much time they were spending on, on that task. Fascinating. And putting real numbers. Cause they couldn’t, they couldn’t figure, they couldn’t tell us exactly. So we, we have really good identifiable numbers in and around that.

 so, you know, I think that’s, that’s definitely a way and, and that technology’s only getting better and better. you know, whether you do UiPath or whether you use Celonis or any of those tools, but basically you’re using that as the engine to kind of figure out what people are doing from a top level. Then I think the other thing is to, once you kind of get in there, use cases will just sort of pop out once they kind of understand, oh wow, I’m, this bot’s doing this for me. I’m sure I have the same exact thing. Or I have a group of people doing a similar type thing at the other part of the organization that I believe that automation could use. So you, you tackle it from both the hands on the ground aspect you know, the automated with task money and then using, you know, a backlog delivery tool, a backlog creation tool where people can submit ideas, you know, they’re evaluated like automation hubs and you can, you can put ROIs and look at the historical values and then attached the PDDs to the different opportunities and track ’em.

And you know, that’s how you scale this. You know, and you can also use process discovery too which is a whole nother animal in itself. But, you know, finding the low level tasks, I think task mining is a good place to start. And then you can look at more of the higher level business process management tools of processes discovery.

CW: Yeah. When, when you do get to that higher level, how much do you all invest in sort of process redesign before you sit down and put hands to keyboard?

PC: Yeah, so <laugh>, one thing a lot of these tools do discover is, I mean, all three approaches all end up having some level of process redesign or, okay. Or, or sometimes it’ll be like, oh yeah, we thought this process was more mature than it is. We need to figure this process out before we’re gonna automate it. So it eliminates the, you know, automating a bad process, which is rule number one, never do. so, you know, there is that process improvement aspect. The other thing we have seen too, quite a bit now in the post covid age is using task capture as a way of documenting a process, because a lot of businesses don’t have their processes fully documented, you know,

CW: Interesting.

PC: So, you know, with turnover and things like that, that’s little like task capture, which has nothing to do with automation other than it’s a UiPath product to capture their day-to-day. And then the, it creates a full for fully formed process document for them to say, okay, this person doing this with s SAP once a month and here’s the process and boom, boom, boom, there it is documented, save it. So if that person’s not there that month or you know, they leave the company, at least there’s some documented, you know, process. yeah. So, you know, we’re seeing that quite a bit too.

CW: Interesting. Hedge against the key person risk bill right in.

PC: Yeah. Yeah. Right.

CW: I like that. It’s exciting. Yeah. So let’s go back, you were talking about one of the value ads for Camp Tech is that you’re constantly kicking the tires on the new and shiny when it comes out. just as an organization, you know, what’s the sort of whether it’s time or dollars or whatever it, what, what’s the proportion of your sort of budget that goes towards that r and d type of stuff?

PC: Yeah, I mean, as a newer company quite a bit we’re at about 25 to 30% I believe. we have budgeted this year. last year is a little bit more. but yeah, we, we basically take 25 to 30% of our budget and put it towards r and d. And r and d can include the technology, but also can include operations, building our operations out, making sure that we’re continuing to serve our customers in a proactive way. so, you know, one of the, one of the things starting the company is to make sure their operations are sound and they continue to improve cuz we can always improve our operations. So like, you know, we put a, we put a significant amount of our revenue that we’ve gotten from our clients to r and d.

CW: Right. On that sounds healthy. Sounds like a way to stay ahead of the curve. so let’s, let’s talk a little bit now about the way you engage with new clients. And I want you to, I want you to sort of spill the tea a little bit on, when you go into an engagement, what do you see warning signs of? Like, this isn’t gonna go well, these people don’t have their stuff together. Like, what, what is it that that sort of tips you off to that?

PC: One of which are you talking pre-sales or actual, when we’ve signed it, signed the deal?

CW: Oh, I’d, I’d love to hear about either. You can take it in whichever order you want.

PC: Yeah. So pre-sales, a lot of times it’s, they don’t really have a budget or Sure. Isn’t necessarily high of mind yet. you know, I think also too, one thing we do in pre-sales before we get statements of work or we sign customer agreements is to do some scoping of what they think they wanna Sure. Write then and there. It’s usually a starter, non-starter. So that what will, what will happen frequently is one, they may not be repaired for the scope call, which does happen, that doesn’t happen as often as, oh, in the middle of it. Like, oh gosh, this isn’t something we should automate, or we still haven’t gotten this defined yet. Or, you know, this isn’t the right person that should be showing this to you guys. you know, those kind of things kind of like help identify. We’re not gonna, we are not gonna sign a cust, we’re not gonna sign a contract with a customer that we don’t believe it’s a good automated automation candidate.

It’s not worth it. I mean, from a resource standpoint, from a lot of different ways, it’s not, it’s not even worth it. And we’ve had to actually turn some people away or some customers away where we’re like, you’re not ready. You want to do this, but, you know, do X, Y, and Z before we can move forward. that being said you know, when you get to the post sales implementation aspect, it’s really, you know, you talked a little bit about identifying use cases and Yeah. You know, building the backlog. It’s, I think the hardest things that we’ve faced with new customers with one, which sounds really, and you guys probably run the same thing, is getting access. Getting access. Oh yeah. Getting all that stuff approved that can take, that takes longer than building the bot. Like we can build a bot in four to six weeks, but sometimes access can take three months, you know?

Yeah. And, and that, that’s just, that’s always been that way since I’ve been doing this. sometimes access assets access can take like three hours and it really, you know, so I think that’s, and that to me says a lot of different things. the company’s not ready for external party to come in and manage the service, or they’re not structurally ready to supply these access controls to us. Yeah. we, we do a lot you know, particularly with some of our larger customers, well, most of ’em at this point, we do security reviews with them before we sign a contract. So sit with it, explain where we’re coming from, where the data’s gonna live, you know, in our cloud. maybe in their cloud, on their pre, not on their prem, depending on what their security model looks like. you know, as things have gotten more cloud receptive, a lot of companies are like, we don’t really wanna manage this.

We work with other managed services like you guys, as long as you’re so too, and, you know, you have the references and all that other stuff, we’re good to go and that they’re faster moving. but, you know, it’s things like, it is access is the problem and then it, then it’s SM e access. So we run, we’ve run into issues where fully formed co you know, simple bots. I mean, I’m not talking really tough ones where it’s like, we’re working with a customer and the SME is not available for weeks. You know, and that’s the one person, the whole organization that knows that one process that they wanna automate. So that slows things down. and then it’s, we always find the first few are probably the, the hardest to kind of get the company’s work and their head around it, thinking about ways to automate or what, what kind of questions we’re gonna be asking after the first few, it really starts to take off and Yeah. Cause they see what our process is. They understand what they need to do, what they can expect. and, and it really starts to get some level of scale. And, you know, we’re, we’re seeing that with all of our customers. They grow over a 30 to 40% clip year over year.

CW: Wow.

PC: But the first year, first six months to year, depending on the, the scale can be a little challenging.

CW: Yeah. There’s a little dip then you take off. Right. You have to wander in the desert for a while. It’s interesting you mentioned the SM e access. You know, we we have SMEs often in our platform supervising the models, right. Just labeling the data. And they’re often the worst model supervisors because they have all of this business knowledge in their heads and they’ll see three different things, and then they highlight this as the key piece of information, and then they’re shocked when the model doesn’t pick up on it. It’s like, well, you know, you gotta give it a B and C before you get to D. yeah. Have you had similar things where the SMEs sort of like, it’s almost like they don’t even rem remember why they do things anymore. They’ve been doing it so long.

PC: Yeah, yeah. Or, or they don’t or they’re really not the SM e

CW: <laugh> <laugh>. Well, yeah, that’s a whole other ball of wax

PC: No, they, we’ve had, we’ve definitely had that issue with SMEs that are like, I don’t know if they get brain freeze or something when they talk to us. but yeah, no, I mean, God love ’em, you know, but that’s why, that’s why I think task mining could be really helpful because Yeah. Sort to get the idea of what it is before we even talk to them. so if we do talk to the SME or we scope with sme, it’s more of a validation of what we’re seeing with task mining.

CW: Yeah. And then they can help with the things that, you know, look like outliers or strange circumstances. Yeah. Yep.

PC: Yes. The SMEs are technical, so if you’re giving SM e that’s not technical, you know, they may be a world-class accountant or finance person, but they, they don’t have the logic, the business or the development logic to be like, okay, the dependency stuff. So once again, managed services are helpful in that respect too, because we can just take what the SME gives us from a document standpoint, teach the models, figure that out, and help them over, you know, when that verification needs to occur.

CW: Yeah, absolutely. We’re about to round out on our time together, so I want to, I want to focus on we talked about the far future. Let’s talk about the near term future. Again, if you could have, you know, wiggle your nose, rub the bottle, whatever it is, if you could have one wish for what technology could get built into these IDP platforms or what feature could be built into these IDP platforms that you work with, what’s the one thing that was, that’s missing right now that would unlocked the most potential?

PC: Ooh, that’s a good question. I, I think a reliable template. I mean, I know Abby has done a decent job with like invoices and things, but real intelligence built into the intelligence.

CW: Yeah.

PC: You know, so if you’re trying to do invoices, you’re trying to do whatever it is, like, you know, in healthcare, like pull out medical record numbers, have these pre-built models as a starting point. and, and it, it really could help a time in the market a lot quicker for development team. And, you know, the issue that we’ve seen with these pre-built models with other companies is they’re not that reliable. So you end up having to do a lot more work outside of the model, but just really building. So that’s kind of what I was talking about earlier. cataloging this data over time and understanding, okay, this is what a telephone, I mean, not to say you guys don’t know what a telephone number is, but Yeah.

CW: <laugh>

PC: Or, or specialized id. So if you’re working with insurance company, they have their own ID system, you know, understanding what those are, what the most common ID structures could be for insurance or, you know, financial services or whatever it is. that kind of thing. I think that would be hugely helpful. you know, I think the feedback loop stuff with the customer is good. It can break down too a little bit at times. yeah. Yeah. But you, you need, you need that like, handholding with, you know, whether it be you guys with your support or our support or whatever it might be.

CW: Good. Good. I like that answer. I, I think that again, back to the concept I mentioned before where we’ve got pipes and we’ve got plumbing and you can get things, you know, you run it through this pipe and this comes out the other side, right? Really capturing a representation of a document along the way and being able to say, ah, I’ve seen, I’ve seen this before and it tripped me up. And and that, that, you know, that’s another one of the things that excites me about models like G P T three and a half, is they do seem like they sort of learn on the fly as they see examples. So I think there’s a bright future there. Yeah. alright, bring us home, Peter. what, what, what should the folks listening out there, whether they’re, you know, sort of boots on the ground building bots, they’re leading a c o e, maybe it’s different answers for everybody involved, but what should they take away from this as we wrap up?

PC: Well, I think one of which is the technology both from RPAs perspective and an intelligent document processing standpoint is here. it’s, it’s available. It, it’s reliable, it’s, you know, industry tested you know, from all different perspectives. I mean, RP itself has been around for 30 plus years in one way or another. but now we have these highly sophisticated enterprise platforms that can deliver, you know, INDICO is using techniques that have been around for quite a while as well, but really taking it to the next level using intelligence and, you know, artificial intelligence and machine learning to tackle it from a, a completely different perspective. non-positional, which is huge. There’s no relation between, it’s not key value pairs any longer which is, you know, Amazon uses that with text React. But so, you know, really tackling the problem, having, having the, the technology can really support the use case, but also understanding what the use case is and, and what technology could really be the best to use in that use case.

And not having to go through, you know, a whole years of discovery with different IDP tools, right. And just kind of zipping into the head of the class and saying, you know what, this is what it’s gonna be. I can get the roi. you know, we’re not gonna have to maintain this as much. It’s gonna work, you know, whatever it might be. You know, and I think that’s really what I I’m focused on, especially this year. Yeah. I think it’s gonna be a big year for intelligent document processing, you know, a more sophisticated R P A bots.

CW: Yeah. I love it. So, to, to sum that up into a few bullet points, you have a need to deal with your unstructured data. The tools are good enough, and make sure you work with someone like the good folks at Camp Tech who have done this before and know what the best practices are and can provide the services you need to get there. Is that fair?

PC: Yes.

CW: Great. Well, thank you, Peter, for joining me on the podcast. it’s been a, it’s been a pleasure. And just to send us home, this has been unstructured, unlocked and I’ve been joined by Peter Camp, who is the CTO and founder of CampTek Software. Thanks again, Peter.

PC: Thanks Chris.

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

Subscribe to our LinkedIn newsletter.

Unstructured Unlocked podcast

March 27, 2024 | E43

Unstructured Unlocked episode 43 with Sunil Rao, Chief Executive Officer at Tribble

podcast episode artwork
March 13, 2024 | E42

Unstructured Unlocked episode 42 with Arthur Borden, VP of Digital Business Systems & Architecture for Everest and Alex Taylor, Global Head of Emerging Technology for QBE Ventures

podcast episode artwork
February 28, 2024 | E41

Unstructured Unlocked episode 41 with Charles Morris, Chief Data Scientist for Financial Services at Microsoft

podcast episode artwork

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.