Christopher M. Wells, Ph. D., Indico VP of Research and Development, talks with Munish Arora, Associate Director of Advanced Analytics, Sun Life, in episode 2 of Unstructured Unlocked. Read on to discover how enterprise data and automation leaders are solving their most complex unstructured data challenges.
Chris Wells: Hi there and welcome to another episode of unstructured unlocked. I’m your host Chris Wells, VP of R&D at Indico Data and I’m really excited to be joined today by Munish Aurora, Associate Director of Analytics and Robotic Process Automation at Sun Life. Munish, welcome to the podcast.
Munish Arora: Ah, thank you Chris. I hope to learn something from you and also contribute here. Thanks so much.
CW: Yeah same. I hope to learn something from you too. If you could, why don’t you tell us a little bit about your career journey; how you ended up in analytics and RPA and then a little bit about what you’re working on these days.
MA: Yeah, so starting with the current organization, as you mentioned, I’m working with Sun Life. So I work on analytics and automation projects. Prior to that, I worked in mainly consulting organizations, where I worked on analytics that include say insights data engineering or predictive modeling and prior to that if I go back I started my career as a developer I got my hands dirty with all different tools and technologies. I still like to do that and now here I am responsible for building analytic insights, mining the data, understanding things, understanding what’s going what’s beneath that and also to process mine and build automation on top of it.
CW: That’s great. How often do you get to get your hands dirty these days?
MA: Usually when people face problems and we need to fix those problems, then I get into those projects and try to help them. But yeah, other things also keep me busy.
CW: Yeah, someone has to manage things I guess. Right, that’s good, tell us a little bit about you. This podcast is all about centers of excellence, centers of enablement for automation and analytics. Tell us a little bit about your experience, and how COEs are sort of set up, how they’re structured. What have you seen out there in the marketplace in your career?
MA: Yeah, so COE is a way to embed and implement RPA deeply and effectively into an organization. That is how I see a COE because that’s a very central part of any organization, and it really helps you to perform many functions. Say for example you want to implement process mining or to go deep into that; you would have people from diverse backgrounds implementing things, bringing in diverse viewpoints, different technologies.
Say you would want to have an infrastructure setup. Now, which tools to use? Which technologies to use? What kind of benefits will your automation bring in? And I guess most importantly: enforcement of standards that you would want all your teams or all your developers to follow. Say what kind of a source code? What kind of coding standards or what kind of design framework should you have? So I guess to enable or to implement those things the COE plays a very very important role. Otherwise, it’s more like different teams are scattered and then not really able to bring in the actual value to your clients.
CW: Yeah, it can end up being one of those things where a lot of different people own it, and that means nobody actually owns it. So, the center of excellence sort of stands as the bastion for those things. You mentioned an interesting one; actually there were two things I wanted to point out. You mentioned sort of just standards and then technology choices. So for those two things which voices are involved in those conversations as you’re making those decisions?
MA: Different sets of people I would say, and I feel we should get diverse views. So first and foremost, definitely those technology experts: architects or designers belong in that area who should at least draft the initial versions. Then we do interact with external consultants or external vendors also. We would want to know what our peer groups are doing. What are the standards available in the market? How are we competitive to those standards? So in that case you would interact with different consulting organizations. Then maybe some organizations which build those platforms also play a great role in that. Those primary technology companies also have a set of standards which they would want their clients to follow or they would prefer at least. I guess we try to take input from all these organizations, our own consulting, and product organizations.
CW: So the center of excellence, especially for automation analytics, is a relatively new thing in the world. Really, I would say it’s started to crystallize in the last five years. How consistent is that advice you’re getting from consultants, vendors, and other professionals that you network with? Is it still sort of scattered or is it starting to centralize around a few things?
MA: No, this is a great point. Who would you trust or who would you go with?
CW: Yeah, I didn’t want to ask it exactly that way, but yeah, how do you make those decisions?
MA: Actually, as I mentioned earlier that we have a standard set of tools within organizations we call them architecture groups. Our own architects sit there, and then there are divisions like governance bodies. So we take all the views, we listen to all. At the end of day we validate, we try to follow a prototype model. They would fix some guidelines on everybody, we would test it, we would build some small POCs, prototypes, and see how it is working, if those standards make sense. Obviously, for older technologies: say java.net where you have that kind of a background, things are already defined.
CW: That’s a good point! I guess I would summarize that by saying: you’d listen to all and then figure out what works for you, and the organization, and the people. Right? I think that’s wise counsel.
Let’s transition a little bit. You can make this specific to you and what your team’s working on, or you can make it more general and talk about it as an industry: the automation industry, the robotics industry, the analytics industry. What are people working on today and how is that different from even just a few years ago?
MA: My view is as an industry, as an organization, companies or organizations have matured at least by five or ten points in the last five years or so. For example, when we started our journey, I saw many organizations about four years or five years back, at that point in time people were focused primarily on task automation. You would want to automate specific tasks. You may want to give some tools to your employees for them to run. From that point, I now see organizations are moving towards process automation. Maybe everybody might have used their low hanging fruits and now they are actually focusing more towards end-to-end automation? In our overall current exposure I also see more AI or problems like unstructured data. These are the things which we could not solve in prior years. We kept at it, and saw it, and now we’re trying to solve those things to automate that complete process from scratch.
MA: Okay, so the low hanging fruit is gone so you got to climb up a little bit higher on the tree. I’ll Circle back to the AI and the unstructured because we want to get to that, but I’m guessing that someone who’s gonna listen to this podcast is reaching that point where they’ve just about got all of the low-hanging fruit and so things are about to change for them and their role leading a COE. What advice would you give to someone who’s making that transition from desktop task automation to full-blown process automation?
MA: My advice would be in that case, I would say if you can skip that low hanging fruit because of the availability of new tools, of process mining, and the technology has advanced. Rather than focus on end-to-end process understanding, see where your process inefficiencies are and then build a solution. In the longer term I see those solutions will give better results. It’ll improve your client experience. One more issue it will avoid: creation of a technical debt. That’s the last thing you would want to do; create an automation and then scrap and build something new. So if somebody can do that it would be great.
CW: That’s interesting, the advice to not be tempted by the low-hanging fruit and to go right for the stuff that really has a high impact is interesting. I haven’t heard that advice. But I think there’s some wisdom to it. The fact that you have processes running on someone’s desktop that they want to automate sort of tells you that there’s something inefficient or broken about the way that you’re working. Right?
MA: The thing is, “low hanging fruits” always help you to convince the largest stakeholders to start automation. You know, I can say that’s also a pivotal role, but in a case where you have an organization which is quite mature; they will understand all those technologies or their COE understands different technologies, tools, toolboxes and things available. Maybe they’ll be able to start at a higher level, so you are increasing the benchmark. So it depends really at which place that current organization is.
CW: Okay, so if you’re still trying to prove out the viability of the COE then pick all the low-hanging fruit. But if you’ve got a mature organization with big processes to automate, go right for those, that’s what you’re saying I think.
MA: Yeah, because nobody would put his or her money, like a million dollars, for you to automate and you don’t know whether that project will be a success or failure.
CW: That’s right! Good. Now I want to give you a chance to just kind of brag a little bit and you can make this as general and abstract as you want. It can be about something going on now or something in the past, but talk to me about an automation or analytics project that you were involved in, or you led, that really had an impact on an organization and really forced you to solve some interesting problems.
MA: Okay I’ll make it quite a practical example. Talking about the insurance sector, coming from an insurance landscape. Insurance is also quite a traditional sector, and you would have a lot of papers involved in it. The best use case anybody would want to have, is to have a straight through processing involved, or have a setup where you can understand what is coming out from those documents. You have doctor prescriptions, you would have medical transcripts, you would have images.
Now, there was one project where we actually tried to automate this area itself wherein we were actually getting a lot of emails. Now, an email would tell you that a client would want to have your spouse added to an insurance policy or they would want some kind of a change. So what you would do is: you would call a service test or you would send them an email. Now a typical process would take a good amount of time because your executive will understand those emails. Somebody would need to read it or a person who is listening on a phone would take notes or process something. So a typical organization will have multiple processes or multiple tools in between before it actually reaches to the final point where you would process that request. In a typical scenario this would lead to a poor client experience, or maybe not as good because it takes time. You would have your request completed after two days, three days or maybe a week or so.
So we try to understand where exactly our process inefficiencies were. Which tools or technologies which were there in this complete life cycle were actually putting in any value? Do we really need to change our platform or do we need to give automation to those executives who are listening to phone calls or people? Can you build something which can read those emails and trigger the next process? In this example we tried to put up a multi-point solution. A solution which is given as unattended bots to call center folks which will help them to navigate all our internal platforms or tools and trigger automation. Then automation will help to file a case, give it to the relevant team to validate the information, and then process it. We also try to implement a text-based solution on the emails which we were getting, and classify those emails where you would want to have a change in your policies or change in your request and try to automate those things as an unattended bot because emails can come in any time of your day and you really don’t need a person just to scan it.
Then there’s going to be some emails which a bot cannot process, so we will give those exception cases to the actual executive or you need the help desk executive to process it. So this way we really were able to shorten our time and our clients were happy. We were turning out the request in many days. Now it just takes a day or two for us to process it and then deliver back the output.
CW: Okay that’s a big project I’m just going to play back to a few things I heard. One you spent time up front trying to understand the whole process end to end. I imagine that produced some artifacts, schematics, and diagrams documentation of what the process is today. Is that right?
MA: Yeah that’s right and one thing I can also specify here is that in this particular process there are multiple teams which get involved. It is not really your automation team. You would have actual business users and your operations. Many organizations follow that consulting people are typically good with process understanding. So it’s the club of all those individuals coming together in at least the first phase where you actually try to understand the process. It is important that business should be involved from day one and not at a later stage.
CW: Absolutely. Okay, so get the process documented, and then I think I heard you say instead of being one massive automation this is several smaller: some attended automations some unattended. You’re surgically putting the automation where it belongs in the process is that right?
MA: That’s right and it was also step by step. It was not really like that we started all this project on day one at the same point. So we identified the process, and we figured out where the inefficiencies lie. Then we took the first project only, or first part of and built that project also in an agile fashion, iteration by iteration, giving something to the business. Once we have delivered that, let the business run that project or run that bot or let them run the attended version first, see how it behaves. That builds their confidence in us. The clients, the actual agents, they are liking the product, they are liking the bot.
Here one challenge which I typically see happen with many organizations or many teams and that is a challenge to do adaptability. You build it and people don’t want to use it. So it safeguards us as well. So let them build, let them use it. Once they have used them for a while, they like it and then they go towards the second part of inefficiency and try to fix that. But we definitely have a big picture in our mind when we start doing that. How will different pieces tie together once the complete project will end.
CW: Great! You mentioned a point which I always love to emphasize, which is you’re asking people to go from their 15 year old Honda Civic to a fully self-driving electric car. Maybe let them get behind the wheel first. Right? Change is hard. Change, it’s hard for all humans but business process change is especially difficult.
So let me ask you this then, you talked about the clients being much happier and you talked about the business being involved from the beginning, super important. The agents and the account executives that are actually working on these cases that come through, after you go through the process and you give them a chance to step into it and get used to it. What happened to their job satisfaction? What were the reviews from the people that are now doing the work with the help of the robots?
MA: I would go back to your point which you mentioned. So, it takes time for people to adapt to change. There is a good effort which is involved first for training those folks. You will have an apprehension, but one thing which really helped us in this was our way of working. That agile way of working. Because when you build prototypes the business would see a few things, or at least something at each stage. They will have something to test every two weeks or every week when we’re going to give them products. So by following this we have seen less apprehensions or surprise factors in terms of business eyes. What exactly have you built? because they have seen it and they have tested it. I would not say the complete business. Let’s say if you have 100 users you would have at least 10 users part of that journey. Those 10 users who have tested your product for the previous three months every single day, every third day, every few weeks, then those 10 individuals will become your brand ambassadors, to convince your [other] 90 users that you need to use it. So I think that strategy of pilot users who are involved while you’re building your automations and then they will drive your adaptability.
CW: That’s great. There’s a lot of conversation out there on LinkedIn and elsewhere about automation, and I think we often forget that people are still involved, and they’re going to be involved. That process you identified right there of taking 10 laboratory test subjects and turning them into people that are gonna talk about the good news to the rest of the organization, this can make the job better and it can make the client experience better, that’s just absolutely critical and it’s a great process to build into your your COE.
MA: I see as such also the users will also get an opportunity to work on most complex tasks right so they also get to avoid those very simple tasks which they were used to doing. Right? So it’s a win-win situation for everybody.
CW: Yeah if you spend your whole day doing mindless stuff, you only have so much mental energy and that would be better spent on the stuff that you need human mental energy for. Good. Well you brought it up a little while ago and I’m going to circle back to it now. Let’s talk a little bit about unstructured data. Not everyone here listening has tried to tackle unstructured data in the past so why don’t we start? I’ll just ask you what unstructured data means to you?
MA: I see it everywhere. All images, video, email, social media footprint, everything. Everything which is probably not in a relational format, those typical RDBMS definitions, everything else is unstructured.
CW: Yeah, everything is unstructured basically if it’s anything interesting. I mean the use case you mentioned was email right? Which is maybe in a lot of ways the most unstructured because people can put whatever they want into an email body. I like the framing basically if it can’t fit into a relational database then it’s probably unstructured. That’s a nice heuristic for people to carry around in their heads.
You talked a little bit about your experience with unstructured in one particular use case. Do you see unstructured as sort of the next big horizon for the automation Center of Excellence or the analytics Center of Excellence or is it still early days?
CW: Absolutely I think the time has come. Businesses really want to have end-to-end automations and I’m very sure when anybody or any organization,you pick up any industry, you’re going to try to automate end-to-end processes. You are going to get unstructured data in between, be it images or voice or some other way. You will face that and if you really want to enhance your client experience you would need to solve that problem. The time has come. I don’t see that as a futuristic thing, yes very much in the present.
CW: All right, you heard it here folks. The time has come for unstructured data automation, Minish Aurora. Good! I think you’re right.
Back to your comments about low-hanging fruit and seeing the whole process and trying to automate the whole process, there are very few processes which drive business value that don’t have some form of unstructured data somewhere along the way as a key component. so thinking back to the early days where you were just doing sort of task automation, robotically clicking through web forms or filling out data, thinking of that which is at this point the very simple stuff, what are the unique challenges that present themselves when you’re trying to automate a process that involves unstructured data?
MA: The biggest challenge which I see at this stage is accuracy. Now if I compare that challenge with a challenge four years back, at that point in time, it was maybe the unavailability of right tools and technologies or organizations who can solve those challenges. So accuracy is a major issue. You really would not want to have your process automated by some technology which will give you say accuracy of 50% or 60% because in that case you would spend a good amount of time in processing the other part and then you may need a good amount of time to validate what has been processed earlier. So accuracy is the main thing.
But yes, as I said earlier, it was a case where you had very few organizations or say boutique players like your organization which specifically worked on this particular problem statement. So I don’t remember many organizations trying to solve unstructured data problems, say six years back or eight years back. Luckily, now we have, I think, a good amount of technologies available and different organizations available. So yeah, probably we want to solve that.
CW: Yeah, interesting point. Six or eight years ago, the organizations solving unstructured, were really the big research labs at Google and Facebook and wherever else. Right now those technologies have been open sourced and there are lots of products built on top of them. So the tools are no longer the excuse I guess.
Circling back to that conversation about accuracy, that’s one conversation that we have a lot. Obviously anyone working with unstructured data does. So I want to frame it this way: the end-to-end process before the document comes in and eventually key details have to end up in a database, that end-to-end process you really need to be a hundred percent accurate right. So how do you decide, like how do you set the benchmark for the tools in between that are taking part of that job. You said 50 to 60% is not enough. Does it have to be 99 how do you decide where to stop in terms of accuracy to get enough value out of something like that?
MA: I would not say 99%. I think I’ll be too insane to ask for it. I put things about 85 percent 85 to 90, 92. I think that’s a good number to have. Then I would say it also depends on the use case. What kind of use case do we have? If I give you an example of a medical transcript you will have analytical transcriptions, medicines written over it versus where you would have a claim amount written on your receipt. So you may manage with the first one bit of a lower accuracy but you would want to have a very high accuracy in the later stage where you process your claims.
Then versus if I change the use case to say social media for some different industry where you are mining data from social media platforms. Then you even would have a challenge first, accuracy is a different matter. You would have a challenge: what exactly you would want to have or what exactly you would even want to store. Then you have different challenges like BAA, and those sensitive things are there. So I think that’s what drives the use case or the technologies we would want to pick up.
CW: Yeah absolutely. On a related note have you seen use cases where the automation, whatever it’s built out of, AI rules whatever; where it gets to 99% accuracy, but it’s still so business critical that there’s there you’re always going to have a pair of human eyes reviewing before it goes to a downstream process?
MA: In my experience, I always see at least some pair of human eyes at least if not say 100% of the cases but either as a random sample where you would randomly validate a few cases. Whether we are calculating your false positives or your false negatives. Whether you’re training so then you would have a trade-off between which will exactly lead to losses or which will exactly lead to poor client experience. Then you will take your call.
CW: Yeah, excellent point. I repeatedly tell anyone that will listen, but a lot of our clients, like accuracy in and of itself, aren’t very meaningful because you don’t have any units, you can’t convert it to dollars or time saved. But that distinction you just made between: are we willing to live with false positives which create one kind of work for us, or false negatives which create a different kind of work or risk, that nuance is really important and I hope everyone listening writes that down because that’s where we really decide like is the process working or not. That’s great. I appreciate that a lot.
Good, so we’ve talked about what unstructured is what some of the challenges are. You made the point earlier that basically everything is unstructured data so given all of the things that you could be working on in the unstructured space, how do you decide which projects are the right ones to take on at a given moment?
MA: Well one of the major things will be client experience. We should pick the channel which will have the maximum impact on client experience. I may want to pick up things related to voice first compared to data. Somebody is listening to you and it’s he or she who is on the call. You may want to solve that problem first. Then there are secondary cases where I would say process to process maturity, where exactly that particular problem fits in whether by solving that particular issue, will my complete process be automated because that will also improve my client experience. Third, I may want to look at my profitability also versus my cost in building that. What will be my payback period? I don’t really prefer to pick up a three-year or five year project. with the payback of again for another three years. You want to have a process which is cheaper to build, faster to build and can give ROI in a shorter time frame. So that use case I will pick.
CW: Yeah all the chief financial officers listening out there just cheered looking for that sub one-year payback for anything. I want to circle back to something we talked about earlier which is: you talked about listening to a lot of voices and deciding which tools and practices when it comes specifically to unstructured. As you’re talking to consulting firms and peers and vendors out there, how do you feel the maturity is in terms of best practice, tooling and the advice that you’re getting in the unstructured space?
MA: That’s really at a starting level I would say, not at a mature level. The advice which we get towards structured automation or process mining, process automation that is quite matured. But, this I may see early to mid-level. And yeah, I don’t blame them because the way things are moving, the way data is changing. I think people may take some time to standardize that.
CW: Yeah absolutely. What are the specific areas where you would say there’s the most friction or the most noise in terms of working with unstructured data right now?
MA: Maybe images. I think that’s the most difficult area at least which I have seen up till now because you know there is a lot of ambiguity in the way you would write or I would write English language and specifically for organizations which operate in different geographies other than say North America, Europe or India Pacific. You would have Regional languages so that’s, I think, a difficult nut. Maybe a lot of organizations are trying to solve that. I expect that is what maybe Google or Microsoft might be processing or training their engines on. So what’s your experience on this one?
CW: Wow! That’s a power move right there, taking over the interviewer’s chair. I love it.
My experience is that, so here’s what I’ve seen in my journey: the last five or ten years working in the enterprise. So RPA comes onto the scene, and RPA is very successful, grows extremely rapidly in the last decade, and sort of dominates the marketplace. So you can get straight through processing on some of that task automation and you can get to really high accuracies on task automation. That has led to the market being saturated by people talking about 99% or 100% accuracy and by the way you can do it with your documents too or your images or your audio. The truth is, here’s where I think the market lacks maturity: those processes with unstructured data, they require intelligence. That’s why we have human beings doing them today. Because when you look at a document and I look at a document, the same data field might occur in five different spots but one of them’s the right one, or maybe two of them are actually the right one, and you and I are going to pick different ones. That noise in the intelligence is an inherent obstacle to straight through processing. So, having the maturity to realize that your humans are central to the process today and they’re going to stay central to the process tomorrow, but they’re going to get a lot of help from the AI and not just the AI but good tools for them to work with that weren’t written as a console application in 2005.
So, maturity in terms of tooling, maturity in terms of understanding that your process has inherent ambiguity in it, and ultimately accuracy is not the number that drives your return on investment. It’s really how efficient and effective your human workers are. Because despite Tesla’s claims that we’re going to all be riding around like The Jetsons, we’re not there yet and we’re not going to be for a while. So that’s where I see a lack of maturity is just understanding what the tools can do and thinking about them in the right way as a partnership with your human knowledge workers. So that’s my two cents.
Now, something we haven’t talked about is, and actually it’s a good segue from task automation to process automation, you automate something and like that email case you talked about it.
It had a bunch of bots, a bunch of new tech that creates organizational overhead; there has to be oversight of those things. So given that when it comes to unstructured data and automating processes that involve unstructured data, what are the unique sort of costs or process controls or governance checks that you have to put in place to manage something like that involving unstructured data?
MA: It depends again on the use case or which kind of data we are trying to mine. Things which are coming to us through traditional routes, as I said medical transcripts or you know people calling and asking for help. There, we have more control on who can see the data, how we can process that. Because you would have a standard pipeline. But if I compare that with somebody reaching out to you on Twitter, somebody is commenting on Facebook, there you would need to have strong controls on who or how you can process or who can see that data. So there are very strong guidelines defined by the organizations. I see that in a majority of organizations, where they have a governing council, they have a security council in place. Where, before you can actually mine or you can process data you would need to go to those councils and put your use case, that is the data which we are getting, how we’re going to process it, how we’re going to anonymize, how I’m going to handle my PIA information. Then there are standards defined for that you need to follow. Then there are different country-specific rules, different geography specific rules which you need to follow. So I think that’s a big work and, at least in my experience, I’ve seen specific councils take that. But yes, I see there is one challenge in their work also because you would have a new set of data, a new set of use cases coming in every few days. They also need to adapt. So that is where, as a COE, we also try to help them when we try to explain our business cases. This is what we are getting and this is how I’m going to store it, these are the controls, then we take it forward.
CW: Yeah, that’s the sort of data and security council framework. I really like that. I hadn’t heard that one before. That’s really interesting. As you’re trying to build out cases to take to these councils: How do I ask this question? One of the tricky parts about unstructured data is you don’t know what’s in it until you actually look at it. So do you find yourself building out a lot of tools to do analytics on unstructured data to help make that business case or do you have to build that new every time you’re faced with a new data set?
MA: I would go back to the model of POCs here. You would take some part of data, obviously, you would have some human who would see that data, who will process. But definitely you would only give that opportunity or permission to those individuals who have the right to watch that. If you have any geography, time geography boundaries or you have any other kind of boundaries you would definitely first subset those kinds of individuals which can at least process it or get that data. Then as a second point, you would give that raw data also to your councils. This is what we have sourced out and then a kind of a pilot on it to take it right to the council because I think that’s the duty of a COE also; to help those Security Council Members understand how we are getting that data, how we’re going to process it, because they don’t know. Someday, if an upcoming use case you may mine something or bring data out of a blockchain. Now you really don’t have any controls over that without technology. I think they’re also going back to the whole point of how COE contributes. So COE also contributes and brings that education or brings that maturity to different stakeholders like this governing council and then they approve or deny your cases.
CW: Yeah I like that. Partnership so that security and the compliance you know they want to know are these things in here which are risky for us and so the COE’s role is to really expose those things and bring best practice to working with data like that so that that question could get answered. I like that partnership a lot. Let’s see, we’re coming up on the bottom of the hour so I want to switch to looking towards the future. You talked a little bit already about the challenges for today. Now is the time to pounce on unstructured data. If we do this podcast again in a couple of years, what challenge do you think we’ll be facing as the next big challenge a few years from now?
MA: That’s a good question what kind of challenges will be facing.
CW: And that counts as an invite to come back to the podcast in a couple years too.
MA: I would love to come back. I see maybe we should be able to solve our basics of unstructured data problems. I could say we will be able to solve problems in silos. Solving maybe a voice problem, solving a text problem. We may lack in solving my end-to-end journey because there is one issue which I see now and which may still be present in the future is what to do with the existing technology landscape in mature organizations or old organizations. I see startups, smaller organizations which are just new to any sector they may be far ahead in the curve. they might solve things related to, they may start working on blockchain or you know maybe newer stuff. Maybe elevating client experience through metaverse. Getting that AI or BI flavor because that will increase your or improve your client experience to a very high level. I see organizations, or mature organizations or older organizations, may need to fix all those old technologies from their bucket before they can latch those things. I think different organizations will be at a different place.
CW: Well that’s interesting. I think what you’re saying is because it’s kind of a fragmented space in terms of technology and vendors and consultants, that in the future we might have accumulated quite a bit of tech debt as we’ve brought in a bunch of different tools and a bunch of different homemade processes to solve problems in silos. So piggybacking on that, are you thinking that now is a good time to try to start building solutions that aren’t siloed in the unstructured space or is it too soon?
MA: I think we should start because as I said earlier, we have those technologies in place, we have organizations like your organization. I see the talent is available. A lot of work has been done by companies like Microsoft, AWS, you have connectors available. Then there is a lot of interest in academics as well where people are trying to solve this. Also as you mentioned, we have already solved the problem of typical structured stuff and if we really need to bring value to the business, we need to solve that. So yeah, in order to avoid that technical debt which will come after a few years it’s better to spend maybe more time and solve at least organizations which are a bit mature, who have already used their low hanging fruits, who start the journey now on unstructured.
CW: Yeah, great! Start the journey towards maturity and unstructured today. I love it. This should be marketing for Indico honestly. All right, so two more questions. These ones are a little bit more controversial: There’s data out there that suggests that something like as little as one in five to as bad as one in ten AI or automation initiatives are successful. So maybe 20%, probably more like ten percent. Why do you think those initiatives fail in the enterprise?
MA: Maybe we expect a lot or maybe we make those initiatives very costly to fail. I believe we should fail fast and fail cheap. I don’t think you can ever build a solution or ever build a platform which will be 100% perfect. It’s not humanly possible. So it is better to fail fast, learn from your mistakes and then move to a different solution. So I would not want to make that very blanket statement that one out of ten is unsuccessful so stop doing it. Right? Because in that case you will stop innovating. You will always go with the safer route and I will not do it. I will go on a safe path.
CW: There you go. Yeah it’s funny, I was talking with someone the other day about the James Webb Telescope that was just launched a little while back and something like 350 individual points of failure which they all had to go right for this thing to get to space and we were talking about why can’t enterprise automation follow this path, and as you know their budget was like five billion dollars. That’s why. if you have five billion dollars and the best engineers in the world you ought to be able to do it but we have to make do with a little less in the enterprise I guess. Last question for the new COE leader out there and then same question for the mature COE leader out there that’s listening: best piece of advice you could give to those two personalities?
MA: Always involve the business with you while you try to solve a problem. Don’t try to solve a problem in the silos. I think the more feedback you will get the more folks you will interact with the more folks you will tell your solution, your options, they’re going to like it and they will give you brilliant suggestions. That’s what I feel is common for every center of excellence. Second, I would say we discussed that point right that we should see an end-to-end process including both structured and unstructured. You may be able to build a better solution with very less technical debt.
CW: Yeah, absolutely. Well, it’s been an absolute pleasure. I hope we get to talk again in a few years about the next stage of the future.
MA: Absolutely, it was a pleasure interacting with you Chris and I also learned a lot
CW: Same! Well, thank you for listening to another episode of Unstructured Unlocked. My guest today has been Munish Aurora and it’s been an absolute pleasure. Enjoy all the automation out there. Munish we’ll see you again down the road.
MA: Thank you.
Check out the full Unstructured Unlocked podcast on your favorite platform, including:
Read the full transcript of the podcast here.
To learn more subscribe to our LinkedIn newsletter.