1 00:00:00,134 --> 00:00:04,784 How much effort do you need in your organization to discover one obscure 2 00:00:04,784 --> 00:00:07,544 workflow that is quietly burning millions? 3 00:00:08,264 --> 00:00:11,414 Today we're exploring how task mining teamed with autonomous AI 4 00:00:11,414 --> 00:00:15,914 agents might deliver savings big enough to move the CFO's dial. 5 00:00:16,424 --> 00:00:19,441 My name is Alexandre Nevski and this is Innovation Tales. 6 00:00:19,667 --> 00:00:22,247 Navigating change one story at a time. 7 00:00:22,277 --> 00:00:26,477 We share insights from leaders tackling the challenges of today's digital world. 8 00:00:26,747 --> 00:00:30,257 Welcome to Innovation Tales, the podcast exploring the human 9 00:00:30,257 --> 00:00:32,117 side of digital transformation. 10 00:00:32,572 --> 00:00:36,617 That invisible workflow siphoning millions doesn't vanish with another slide deck. 11 00:00:37,437 --> 00:00:40,257 It takes data you can see and action you can trust. 12 00:00:41,517 --> 00:00:43,957 Tuhin Chakraborty lives at that intersection. 13 00:00:44,107 --> 00:00:48,367 As CEO and co-founder of Mimica AI, he turns messy click streams 14 00:00:48,487 --> 00:00:50,407 into crystal clear process maps. 15 00:00:50,977 --> 00:00:55,147 With engineering stops at LinkedIn and Pandora, and a patent for a supervised 16 00:00:55,147 --> 00:01:00,067 learning model, Tuhin knows how to convert AI hype into CFO-level ROI. 17 00:01:01,132 --> 00:01:03,472 Today he shares battle-tested tactics. 18 00:01:03,862 --> 00:01:05,812 Where to start test mining pilots? 19 00:01:06,082 --> 00:01:07,702 How to prove value in weeks? 20 00:01:07,942 --> 00:01:11,182 And why pairing agents with human oversight will unlock 21 00:01:11,182 --> 00:01:12,922 sustainable scale in the future? 22 00:01:13,522 --> 00:01:16,732 If you're juggling stakeholder skepticism and budget pressure, 23 00:01:17,062 --> 00:01:18,472 this playbook is for you. 24 00:01:19,492 --> 00:01:22,752 Without further ado, here's my conversation with Tuhin Chakraborty. 25 00:01:23,204 --> 00:01:24,334 Tuhin welcome to the show. 26 00:01:24,518 --> 00:01:25,328 Thank you for having me. 27 00:01:25,722 --> 00:01:26,772 I'm glad you could join. 28 00:01:27,642 --> 00:01:31,662 I'd like to start us off, as my listeners know by now, with a good story. 29 00:01:31,752 --> 00:01:34,752 You and your company specialize in task mining. 30 00:01:35,262 --> 00:01:38,227 Would you mind describing what that is, but from the perspective 31 00:01:38,227 --> 00:01:39,097 of one of your clients? 32 00:01:39,432 --> 00:01:40,447 Yeah, absolutely. 33 00:01:40,447 --> 00:01:46,917 So task mining is a new technology that can make inferences about the work that 34 00:01:46,917 --> 00:01:49,317 people do on their computers using AI. 35 00:01:50,172 --> 00:01:54,462 So essentially what we do is record the clicks and keystrokes into the 36 00:01:54,462 --> 00:02:00,462 computer, and we use advanced machine learning and AI to process this data 37 00:02:00,792 --> 00:02:05,172 to determine what are the tasks and processes that people are working on. 38 00:02:05,752 --> 00:02:07,522 Where are they spending their time? 39 00:02:08,002 --> 00:02:11,692 Where are the best opportunities for automation and process improvement? 40 00:02:12,202 --> 00:02:15,982 And then we can generate these process maps that are used for 41 00:02:16,042 --> 00:02:18,142 process improvement and automation. 42 00:02:18,213 --> 00:02:23,763 So for example, we do a lot of work in the insurance industry and one 43 00:02:23,763 --> 00:02:26,763 of the big insurance companies in the US is one of our customers. 44 00:02:27,453 --> 00:02:33,543 And essentially what they did is use Mimica to identify all of the different 45 00:02:33,543 --> 00:02:36,123 activities in the claims process. 46 00:02:36,333 --> 00:02:42,603 So this company spends $240 million a year on claims and yet they have 47 00:02:42,603 --> 00:02:44,733 no idea how this money is spent. 48 00:02:44,853 --> 00:02:47,853 What are the different activities involved in the claims process? 49 00:02:48,213 --> 00:02:50,343 How much time is spent in each one? 50 00:02:51,183 --> 00:02:54,213 Where are the opportunities for process improvement and automation? 51 00:02:54,573 --> 00:02:59,182 And what they did is use Mimica to record 1000 claims agents. 52 00:03:00,153 --> 00:03:03,903 those recordings, they were able to identify all of the different activities 53 00:03:03,903 --> 00:03:10,023 involved in the claims process and create productivity benchmarks for each activity. 54 00:03:10,443 --> 00:03:15,063 So now for the first time, they're able to measure how is that $240 55 00:03:15,063 --> 00:03:19,143 million breaking down across all the activities involved in the work. 56 00:03:19,893 --> 00:03:24,033 And one of the key insights that Mimica was able to identify for this customer 57 00:03:24,423 --> 00:03:29,778 is that claims involving a tow truck, so these are auto claims, so you know, when 58 00:03:29,778 --> 00:03:35,328 you have a car accident, you file a claim, claims involving a tow truck take 30% 59 00:03:35,328 --> 00:03:37,848 longer to process than claims that don't. 60 00:03:38,088 --> 00:03:38,778 That's unexpected. 61 00:03:39,423 --> 00:03:39,883 Why is that? 62 00:03:40,065 --> 00:03:43,845 The reason it was happening was because the case management system that they 63 00:03:43,845 --> 00:03:49,435 used to process the claims, the workflow around claims that involved a tow 64 00:03:49,555 --> 00:03:51,715 was significantly more inefficient. 65 00:03:51,776 --> 00:03:55,706 The clicks and keystrokes required to execute that workflow and process a 66 00:03:55,706 --> 00:04:00,876 claim of that type took much longer than claims that did not involve a tow. 67 00:04:01,946 --> 00:04:05,366 And this was a new insight that was created for the company. 68 00:04:05,366 --> 00:04:09,326 And in this way, they thought of Mimica as a new category of business 69 00:04:09,326 --> 00:04:14,156 intelligence, and they got the product manager who owned the case management 70 00:04:14,156 --> 00:04:16,286 system to rebuild that workflow. 71 00:04:16,616 --> 00:04:20,456 And they were able to get the delta down by 80%, which resulted 72 00:04:20,456 --> 00:04:22,586 in about $5 million in savings. 73 00:04:22,981 --> 00:04:28,361 That's an amazing story and the way you're telling it, I think, to me 74 00:04:28,361 --> 00:04:33,341 sounds the way that it would been experienced by the managers, the 75 00:04:33,341 --> 00:04:39,161 leaders, the executives of the company, that before they had no visibility 76 00:04:39,581 --> 00:04:45,881 and now they have this very precise, accurate breakdown of those 240 million. 77 00:04:46,766 --> 00:04:52,236 But what about the experience of the claim specialist 78 00:04:52,236 --> 00:04:53,286 that you were referring to? 79 00:04:53,886 --> 00:04:57,696 Doesn't it feel for them, like a bit of an invasion of their privacy? 80 00:04:57,978 --> 00:05:00,558 Now the people that are involved in the process, they are 81 00:05:00,558 --> 00:05:02,808 active participants in this. 82 00:05:02,958 --> 00:05:07,288 So it's not a background recorder that sort of, surreptitiously 83 00:05:07,288 --> 00:05:10,588 captures information, while being hidden from the user. 84 00:05:11,128 --> 00:05:15,238 We actually put them at the center of the recording experience, and the 85 00:05:15,238 --> 00:05:21,148 reason is that every employee and really every person has something that they 86 00:05:21,148 --> 00:05:23,368 would rather be doing with their time. 87 00:05:23,518 --> 00:05:30,028 So that team has a remit, and that remit has certain activities that are probably 88 00:05:30,028 --> 00:05:36,078 more core to what that team is responsible for doing than these repetitive and 89 00:05:36,078 --> 00:05:40,708 mundane activities that they're forced to do on a daily or weekly basis. 90 00:05:41,278 --> 00:05:45,358 And the people that we record are often eager to participate in this recording 91 00:05:45,358 --> 00:05:49,738 process because Mimica enables this work to be taken off their plate. 92 00:05:49,738 --> 00:05:54,248 It enables the work to be automated, and this gives, that individual or that 93 00:05:54,248 --> 00:05:58,358 team more time to work on the things that they would like to be working on. 94 00:05:59,228 --> 00:06:03,258 Now in order to create, an experience that protects their privacy. 95 00:06:03,423 --> 00:06:06,243 There are a number of controls that we put in place. 96 00:06:06,273 --> 00:06:11,433 One, the recorder always sits front and center on the person's desktop. 97 00:06:11,433 --> 00:06:12,993 So it's not hidden in the background. 98 00:06:13,153 --> 00:06:16,473 Two, it has a start, a stop, and a pause button on it. 99 00:06:16,833 --> 00:06:21,363 So the user is in complete control over what is being recorded and when. 100 00:06:21,903 --> 00:06:23,343 They can turn it off at any time. 101 00:06:23,343 --> 00:06:26,223 If they're ever doing something unrelated to the work, they can simply pause the 102 00:06:26,223 --> 00:06:30,003 recorder and then resume it when they get back to their day-to-day activities. 103 00:06:30,693 --> 00:06:34,933 And then finally, and this is maybe the most critical point, Mimica 104 00:06:34,953 --> 00:06:38,753 anonymizes, all the information that flows through the system. 105 00:06:38,813 --> 00:06:41,903 So there's absolutely no confidential or personally identifiable 106 00:06:41,903 --> 00:06:43,073 information that's captured. 107 00:06:43,523 --> 00:06:46,943 Things like password fields are never recorded and in addition, 108 00:06:46,973 --> 00:06:50,333 names, emails, phone numbers, addresses, that sort of thing. 109 00:06:50,513 --> 00:06:52,463 All of that is stripped out of the recording. 110 00:06:52,463 --> 00:06:54,953 So when you look at the result in Mimica, you don't see any 111 00:06:55,043 --> 00:06:56,183 confidential information. 112 00:06:56,286 --> 00:06:57,996 Yeah, that makes total sense. 113 00:06:58,836 --> 00:07:05,456 And I wonder, you know, I've been in this business for a while, so before, we 114 00:07:05,456 --> 00:07:10,286 had these automated ways of discovering a business process, we would have 115 00:07:10,316 --> 00:07:14,786 business analysts going around and asking people questions and drawing diagrams. 116 00:07:14,786 --> 00:07:20,006 And I think this is a bit different because it provides a level of, let's 117 00:07:20,006 --> 00:07:26,806 say, objective reality that was a bit hard to achieve because it was very 118 00:07:26,806 --> 00:07:31,936 often the case that every actor would see what was going on a bit from their 119 00:07:31,936 --> 00:07:37,229 own angle and not necessarily agree on where the time was wasted or not. 120 00:07:37,229 --> 00:07:44,159 So can you share any anecdote or any story that shows how at the personal 121 00:07:44,159 --> 00:07:48,134 level that transparency can be valuable? 122 00:07:48,684 --> 00:07:51,504 There are many examples of this, but one that immediately comes to 123 00:07:51,504 --> 00:07:56,616 mind is some work that we did at a large telecommunications company. 124 00:07:57,366 --> 00:08:02,316 And it was the first time that we delivered a Mimica result to a customer. 125 00:08:02,796 --> 00:08:06,126 And the person receiving it actually got a little bit 126 00:08:06,126 --> 00:08:08,916 emotional when they saw the output. 127 00:08:09,796 --> 00:08:13,046 Mimica identified an automation opportunity. 128 00:08:13,115 --> 00:08:15,455 That was a process that this person was working on. 129 00:08:15,875 --> 00:08:21,105 And essentially what the task was is that they had to take a spreadsheet of 130 00:08:21,105 --> 00:08:25,395 all the candidates that the recruiters were meant to reach out to, and they 131 00:08:25,395 --> 00:08:30,075 would break that spreadsheet down into 10 or 12 different, smaller 132 00:08:30,075 --> 00:08:34,845 spreadsheets and essentially use that to distribute to the various recruiters. 133 00:08:35,685 --> 00:08:39,675 And we learned on that call that this individual had been doing 134 00:08:39,675 --> 00:08:45,960 that process every day for one hour a day for the last 15 years. 135 00:08:46,513 --> 00:08:46,903 Wow, 136 00:08:47,403 --> 00:08:52,783 For him Mimica was a gift because it enabled the automation of a 137 00:08:52,783 --> 00:08:56,713 process that was very repetitive, very mundane, that he had spent a 138 00:08:56,713 --> 00:08:59,503 significant amount of time, working on. 139 00:08:59,503 --> 00:09:04,273 And he was emotional about the fact that for the first time in 15 years, 140 00:09:04,273 --> 00:09:05,833 this work would be taken off his plate. 141 00:09:05,982 --> 00:09:06,492 Crazy. 142 00:09:06,782 --> 00:09:12,402 So coming back to what I was saying about how it used to work with business analysts 143 00:09:12,402 --> 00:09:17,962 coming to model a business process before the advent of automated technologies 144 00:09:17,962 --> 00:09:22,842 like, like yours, there was a number of conversations that needed to happen 145 00:09:22,962 --> 00:09:28,332 for people to figure out what the current process, the AS-IS process was, and 146 00:09:28,332 --> 00:09:30,702 then discuss what the TO-BE should be. 147 00:09:31,332 --> 00:09:39,402 So can you maybe explain how that works with the data that Mimica compiles? 148 00:09:39,882 --> 00:09:44,382 I'm curious about how the model, first of all, make sense of all 149 00:09:44,382 --> 00:09:45,972 the signals that it's getting. 150 00:09:45,972 --> 00:09:49,722 Because people are using their computers to do many things, and 151 00:09:49,722 --> 00:09:53,022 some of those things may not be related to the business process. 152 00:09:53,442 --> 00:09:58,422 But then how does it get to a point where the process actors 153 00:09:58,422 --> 00:10:01,482 actually start recognizing, oh, that is what we're doing today. 154 00:10:01,542 --> 00:10:06,647 And then from there, how do they start thinking about what should we change? 155 00:10:07,197 --> 00:10:11,957 The traditional way of understanding processes is by interviewing 156 00:10:11,957 --> 00:10:14,747 people and asking them questions about the way that they work. 157 00:10:15,737 --> 00:10:20,297 So oftentimes you will be looking over somebody's shoulder. 158 00:10:20,327 --> 00:10:23,507 At one of our customers, they call them desk rides. 159 00:10:23,904 --> 00:10:28,924 So you're basically, you know, shoulder surfing essentially, and asking someone 160 00:10:28,924 --> 00:10:30,394 questions about the way that they work. 161 00:10:30,394 --> 00:10:34,654 You could be taking screen recordings and going home in the evening 162 00:10:34,654 --> 00:10:38,044 and reviewing them, and manually creating these process maps. 163 00:10:39,304 --> 00:10:46,324 And as you go through the transformation journey, as you're making improvements 164 00:10:46,324 --> 00:10:50,134 to the process, you're doing that based on this anecdotal data. 165 00:10:51,154 --> 00:10:55,834 And the problem is that anecdotal data is rarely accurate. 166 00:10:56,254 --> 00:10:59,284 So people often misremember what they do. 167 00:10:59,734 --> 00:11:03,244 They mischaracterize the inefficiencies in the process. 168 00:11:03,244 --> 00:11:06,274 They over or underestimate the automatability. 169 00:11:06,964 --> 00:11:10,864 So in a word, they are biased about their own work. 170 00:11:11,001 --> 00:11:16,761 What Mimica does is, provide a way to get the get to the source of 171 00:11:16,761 --> 00:11:18,921 truth, which is the data itself. 172 00:11:19,191 --> 00:11:22,971 So rather than relying on the person's anecdotes, you can put a 173 00:11:22,971 --> 00:11:26,891 recorder on somebody's computer, they control it, of course. 174 00:11:26,891 --> 00:11:28,691 The data's anonymized, of course. 175 00:11:29,141 --> 00:11:33,911 But from that you can get really accurate information about what 176 00:11:33,911 --> 00:11:35,081 are the tasks and processes? 177 00:11:35,081 --> 00:11:37,861 How much time do they take, what do the process maps look like? 178 00:11:39,226 --> 00:11:44,566 And from there, your transformation journey is rooted in the data itself. 179 00:11:45,136 --> 00:11:49,276 And within the tool, you can actually model different scenarios so you 180 00:11:49,276 --> 00:11:54,646 can see, okay, if we were to deploy let's say vanilla RPA, robotic 181 00:11:54,646 --> 00:11:58,486 process automation, to automate that process, what would the impact be? 182 00:11:58,996 --> 00:12:02,406 Or if we were to bring in an advanced technology like OCR or 183 00:12:02,406 --> 00:12:06,286 IDP, or Agentic AI, how would that change the automation potential? 184 00:12:06,286 --> 00:12:07,906 What would that new process look like? 185 00:12:08,266 --> 00:12:11,846 Are there fewer steps that would need to be done? 186 00:12:11,846 --> 00:12:14,276 Steps that can be eliminated, steps that can be optimized. 187 00:12:14,816 --> 00:12:16,736 So what does the TO-BE process look like? 188 00:12:16,736 --> 00:12:20,681 All of that can be modeled within Mimica using the underlying data itself. 189 00:12:20,801 --> 00:12:24,701 And this is what we've seen time and time again be critical to companies 190 00:12:24,701 --> 00:12:27,071 for the success of the transformation. 191 00:12:27,461 --> 00:12:28,001 Okay. 192 00:12:28,301 --> 00:12:31,961 Well, you've mentioned a number of solutions, let's say to get to a TO-BE 193 00:12:32,071 --> 00:12:37,341 process, and I don't really want to spend too much time on the traditional stuff 194 00:12:37,341 --> 00:12:41,661 that's been around for a while, like RPA, but you've mentioned agentic automation. 195 00:12:42,031 --> 00:12:49,411 So how do you see this symbiosis between what you do in task mining and 196 00:12:49,441 --> 00:12:52,681 the possibilities that are becoming available with agentic automation? 197 00:12:53,481 --> 00:12:58,381 Yeah, so task mining and agentic automation are inextricably linked. 198 00:12:59,506 --> 00:13:04,347 In fact, when we started Mimica, we built the company on the vision 199 00:13:04,347 --> 00:13:06,237 of a new type of automation. 200 00:13:07,197 --> 00:13:12,297 One that did not need to be explicitly programmed, click by click, but 201 00:13:12,297 --> 00:13:15,087 rather could learn by observation. 202 00:13:16,017 --> 00:13:20,547 So in the same way that humans learn, it can watch what someone does, it can learn 203 00:13:20,547 --> 00:13:22,947 from it, and it can reproduce the task. 204 00:13:23,307 --> 00:13:27,237 And this is how we thought computers would be programmed in the future. 205 00:13:27,483 --> 00:13:31,533 In order to build this technology, the first thing that you need to build is the 206 00:13:31,533 --> 00:13:35,610 ability to interpret the work, recording clicks and keystrokes and deriving 207 00:13:35,610 --> 00:13:37,380 meaning from those clicks and keystrokes. 208 00:13:38,370 --> 00:13:42,630 And what we've built so far, which is an analytics tool, is really just the 209 00:13:42,630 --> 00:13:45,270 first step on that long-term journey. 210 00:13:46,170 --> 00:13:49,260 So we've gotten really good at understanding the clicks and keystrokes 211 00:13:49,260 --> 00:13:52,890 that people do, and we've created an analytics layer on that, and we've 212 00:13:52,890 --> 00:13:56,280 put a ribbon around that product and we sell it in the market today. 213 00:13:56,280 --> 00:14:00,750 But really, all the technology that we built is in service of the automation 214 00:14:00,750 --> 00:14:05,220 vision, where we want to take that learning of the clicks and keystrokes 215 00:14:05,250 --> 00:14:07,440 and play them back on the computer. 216 00:14:07,591 --> 00:14:10,391 And that will be the final step on our product roadmap. 217 00:14:10,742 --> 00:14:11,042 I see. 218 00:14:11,042 --> 00:14:11,792 Very interesting. 219 00:14:12,077 --> 00:14:16,532 I might have pushed you into using some of the technical terms, but 220 00:14:16,742 --> 00:14:17,972 we didn't really define them. 221 00:14:17,972 --> 00:14:22,482 So what is agentic automation from your perspective, and maybe just re 222 00:14:22,872 --> 00:14:27,162 retell that description, but from the point of view of a business 223 00:14:27,162 --> 00:14:29,382 leader, like why should they care? 224 00:14:29,382 --> 00:14:34,512 Why would, how would that look like in the future once you have that capability? 225 00:14:35,397 --> 00:14:36,687 So that's a great question. 226 00:14:36,840 --> 00:14:42,310 The definition of agentic automation is still forming in many ways. 227 00:14:43,420 --> 00:14:49,811 But from Mimica's perspective, agentic automation is one that does not need 228 00:14:49,931 --> 00:14:53,651 any specialized skillset to build. 229 00:14:54,341 --> 00:14:57,431 So typically when you want to automate something, you 230 00:14:57,431 --> 00:14:59,891 need an automation developer. 231 00:14:59,981 --> 00:15:02,261 This is a, a, specialized skillset. 232 00:15:02,261 --> 00:15:05,561 Someone who has trained in writing computer code. 233 00:15:06,356 --> 00:15:10,796 And they may have learned to build robotic process automation, which 234 00:15:10,796 --> 00:15:14,576 is one type of automation, or they may have learned to build Python 235 00:15:14,576 --> 00:15:17,156 automation or Excel macro automation. 236 00:15:17,156 --> 00:15:21,326 These are all different types of automation that are used to automate 237 00:15:21,386 --> 00:15:28,046 different processes, but in all of these cases, you need someone specialized who 238 00:15:28,046 --> 00:15:34,286 knows how to write code and has trained in building that specific type of automation. 239 00:15:34,429 --> 00:15:39,139 When the development of automation is bottlenecked by this specialized 240 00:15:39,139 --> 00:15:45,739 skillset, it really caps the value that can be created from it because 241 00:15:45,739 --> 00:15:48,422 these developers are in short supply. 242 00:15:48,472 --> 00:15:53,122 In our view of the future agentic automation, you will not 243 00:15:53,122 --> 00:15:58,382 require a specialized skillset in order to build the automation. 244 00:15:58,862 --> 00:16:04,112 But in fact, anybody will be able to deploy an automation simply by 245 00:16:04,232 --> 00:16:08,132 demonstrating their process to the agent, and the agent will be able 246 00:16:08,132 --> 00:16:10,712 to learn from it and reproduce it. 247 00:16:10,739 --> 00:16:15,179 So what this means for a leader in the organization, if you look 248 00:16:15,179 --> 00:16:16,859 at the before and the after? 249 00:16:17,489 --> 00:16:23,444 So before you would deploy a business analyst who would interview the person 250 00:16:23,444 --> 00:16:28,214 who's doing the process, and they would manually create a process map. 251 00:16:29,414 --> 00:16:33,614 And that process map would then be given to an RPA developer who 252 00:16:33,614 --> 00:16:37,724 would use it to manually build an automation click by click. 253 00:16:38,504 --> 00:16:43,724 And that entire process from start to finish can take six months or more with 254 00:16:43,724 --> 00:16:45,254 all of these different people involved. 255 00:16:46,334 --> 00:16:51,884 In the Mimica world you won't need a business analyst or an RPA developer. 256 00:16:52,124 --> 00:16:55,844 You will simply have the person that does the process and the Mimica recorder. 257 00:16:56,464 --> 00:17:00,644 The recorder will capture the process, it will generate a workflow, and 258 00:17:00,644 --> 00:17:04,784 then that workflow will automatically be used to power the automation. 259 00:17:06,134 --> 00:17:10,034 And in this way, we will automate the automation process. 260 00:17:10,934 --> 00:17:12,434 That's a very ambitious vision. 261 00:17:12,674 --> 00:17:13,244 I like it. 262 00:17:13,874 --> 00:17:17,294 But if you don't mind playing devil's advocate for a minute 263 00:17:17,294 --> 00:17:22,984 here, isn't there part of the process that is not on the desktop? 264 00:17:23,104 --> 00:17:27,454 Instead, it's in the process actors' minds, right? 265 00:17:27,674 --> 00:17:29,714 The thinking that they're doing, the decision they're doing. 266 00:17:29,964 --> 00:17:37,734 Can you clarify, how in your vision, does that get captured somehow and automated? 267 00:17:37,734 --> 00:17:42,924 Or is that, on the countrary, stuff that remains after the automation is complete? 268 00:17:43,424 --> 00:17:46,184 I'm glad that you brought this up because this is the crux 269 00:17:46,184 --> 00:17:48,104 of the challenge in many ways. 270 00:17:48,104 --> 00:17:51,494 Is how to deal with the decision making that the person does? 271 00:17:51,494 --> 00:17:54,434 And there's so many different types of decision, and you can 272 00:17:54,434 --> 00:17:56,164 categorize them by complexity. 273 00:17:57,049 --> 00:18:01,369 Now in, in the simplest case, we talk about rules-based decisions. 274 00:18:01,429 --> 00:18:07,739 So that is, if the invoice is greater than 50,000, then it needs to be approved. 275 00:18:07,739 --> 00:18:09,679 It needs to go through an extra layer of approval. 276 00:18:10,149 --> 00:18:13,279 So that is a business rule that is going to be different from 277 00:18:13,279 --> 00:18:14,629 organization to organization. 278 00:18:14,629 --> 00:18:17,149 These business rules are unique to every organization. 279 00:18:17,149 --> 00:18:21,139 So the question is, can this rule be learned from the data? 280 00:18:21,139 --> 00:18:22,849 Our belief is yes. 281 00:18:23,269 --> 00:18:28,579 That if you observe this decision being made many times and that sometimes 282 00:18:28,579 --> 00:18:32,869 path A is taken and sometimes path B is taken and you correlate that with 283 00:18:32,869 --> 00:18:37,579 the underlying data in the process, then you can ultimately predict that 284 00:18:37,579 --> 00:18:42,769 it's a very high likelihood, let's say that the amount of the invoice 285 00:18:43,189 --> 00:18:44,989 is contributing to the decision. 286 00:18:44,989 --> 00:18:47,569 And from that, you can ultimately derive a decision rule. 287 00:18:48,859 --> 00:18:54,289 Now, the same logic can be applied to more complex decisions as well. 288 00:18:54,289 --> 00:18:59,599 And we've seen with the latest advances in LLMs that they are capable of reasoning, 289 00:19:00,049 --> 00:19:03,259 taking data from different sources, unstructured sources, interpreting 290 00:19:03,259 --> 00:19:08,479 them, reasoning to ultimately take a decision on the right course of action. 291 00:19:09,079 --> 00:19:14,639 As these models become more sophisticated and their reasoning capabilities become 292 00:19:14,639 --> 00:19:19,229 more sophisticated, the complexity of the decisions that they will be able to take 293 00:19:19,229 --> 00:19:21,449 independently will also grow over time. 294 00:19:21,689 --> 00:19:26,189 Short answer to your question, yes, we believe that we can understand how 295 00:19:26,189 --> 00:19:29,969 the decisions are made, certainly for rules-based decisions today and for 296 00:19:29,969 --> 00:19:31,979 increasingly complex decisions over time. 297 00:19:32,350 --> 00:19:33,730 Okay, great, thank you. 298 00:19:34,370 --> 00:19:37,940 where do you see Mimica's role in this evolution? 299 00:19:37,940 --> 00:19:45,830 Do you see like a specialized provider of the task mining or, I think you maybe have 300 00:19:45,890 --> 00:19:52,685 alluded to that, that you would take the company to provide the full spectrum of, 301 00:19:52,785 --> 00:19:55,195 business process automation capabilities? 302 00:19:55,572 --> 00:19:56,937 Yeah, it, that's a great question. 303 00:19:57,284 --> 00:20:01,784 If you look today in the enterprise, the vast majority of 304 00:20:01,784 --> 00:20:03,284 the work is being done by people. 305 00:20:04,994 --> 00:20:10,349 In the future, you will still have work being done by people, a lot of 306 00:20:10,349 --> 00:20:14,309 the work that's currently being done by people will be given over to agents. 307 00:20:14,459 --> 00:20:19,259 And you'll also, alongside these people, have a large number of agents 308 00:20:19,259 --> 00:20:22,409 that are also working day and night to execute the work in the enterprise. 309 00:20:22,512 --> 00:20:26,647 Now we see task mining as serving two functions. 310 00:20:27,107 --> 00:20:33,762 One, it's a layer on top of all the digital work happening in the company. 311 00:20:34,722 --> 00:20:37,242 And the goal is to make that work more efficient. 312 00:20:38,022 --> 00:20:42,702 Now, part of what we're doing by providing that analytics is helping the organization 313 00:20:42,702 --> 00:20:50,832 determine which aspects of that work to peel off into the agentic layer, right? 314 00:20:50,952 --> 00:20:53,802 So all the work today is being done by humans. 315 00:20:54,702 --> 00:20:59,562 Iteratively, we will give over more and more of that work to the agents. 316 00:21:00,592 --> 00:21:04,432 And Mimica will help identify what aspects of that work should be given over. 317 00:21:05,302 --> 00:21:10,402 And as the work that the humans are doing becomes more complex, the analytics 318 00:21:10,402 --> 00:21:15,292 that we provide will again, become more sophisticated to give insights into how 319 00:21:15,292 --> 00:21:17,902 to make that creative work more efficient. 320 00:21:18,682 --> 00:21:23,077 And at the same time, our ambition is also to own the agentic layer and the 321 00:21:23,077 --> 00:21:24,877 transition between the two layers. 322 00:21:25,237 --> 00:21:31,337 So you can imagine in the future you have, for any given organization, an army of 323 00:21:31,487 --> 00:21:37,187 people that are doing work, an army of agents that are doing work and Mimica is 324 00:21:37,187 --> 00:21:41,297 the kind of infrastructure layer on top of both that's transitioning work from 325 00:21:41,297 --> 00:21:43,657 human to agents more and more over time. 326 00:21:43,957 --> 00:21:48,202 And maybe someday we will have a world where all the the work is being done 327 00:21:48,202 --> 00:21:52,012 by agents and the humans are just all sitting on a beach somewhere. 328 00:21:52,242 --> 00:21:55,342 I think we're still a ways away from that, but, but we hope that, 329 00:21:55,392 --> 00:21:58,402 that Mimica will, will be running that work every step of the way. 330 00:21:59,155 --> 00:21:59,605 Okay. 331 00:21:59,665 --> 00:22:00,025 Okay. 332 00:22:00,075 --> 00:22:05,330 for our listeners, I think a lot of them, this vision of sitting at the 333 00:22:05,330 --> 00:22:09,440 beach while all the work is being done by agents is probably a mixture of, 334 00:22:09,490 --> 00:22:13,770 obviously that would be cool, but also there is probably anxiety around that. 335 00:22:13,770 --> 00:22:20,370 So, let's, before we conclude, try and provide some more practical, short 336 00:22:20,370 --> 00:22:23,160 term advice for business leaders. 337 00:22:23,380 --> 00:22:27,620 What would be the main key takeaway that you would want them to come 338 00:22:27,620 --> 00:22:29,690 away with from this conversation? 339 00:22:30,950 --> 00:22:33,915 Where's the first step, let's say for them? 340 00:22:34,548 --> 00:22:41,680 There's absolutely no doubt that the advent of LLMs, you know, foundational 341 00:22:41,680 --> 00:22:48,470 models, ChatGPT, that technology is already capable of fundamentally changing 342 00:22:48,470 --> 00:22:50,450 the way that we work in the enterprise. 343 00:22:50,590 --> 00:22:55,030 There's absolutely no doubt that the advent of that technology has unlocked 344 00:22:55,150 --> 00:23:00,250 a completely new category of work that can be automated and therefore 345 00:23:00,250 --> 00:23:04,355 a completely new level of efficiency that the organization can achieve. 346 00:23:05,500 --> 00:23:10,690 So the main question as a leader that you should be asking, is 347 00:23:10,690 --> 00:23:15,280 not whether the technology can transform your business, but how? 348 00:23:15,280 --> 00:23:22,810 The first step to take is really to do a holistic assessment of all the manual 349 00:23:22,810 --> 00:23:27,520 work happening in your organization to determine what percentage of that 350 00:23:27,520 --> 00:23:34,075 work and which parts of it can be given over to an LLM or to an agent that 351 00:23:34,075 --> 00:23:36,925 can begin automating that, that work. 352 00:23:37,585 --> 00:23:42,355 So that's where we would recommend using Mimica because, at the click 353 00:23:42,355 --> 00:23:46,285 of a button, you can get a complete understanding of all the manual work 354 00:23:46,285 --> 00:23:52,795 happening in your organization and you can instantly capture, map, and 355 00:23:52,795 --> 00:23:55,575 ultimately optimize those processes. 356 00:23:56,185 --> 00:24:01,045 And what Mimica will give you essentially is a roadmap for how to get from 357 00:24:01,045 --> 00:24:05,485 where you are today to that future vision of your enterprise, where a 358 00:24:05,485 --> 00:24:10,255 lot of that work, or as much of it as possible, has been given over to agents. 359 00:24:10,703 --> 00:24:11,153 Makes sense. 360 00:24:11,153 --> 00:24:14,423 Get clarity, get visibility of all the work that happens first. 361 00:24:14,853 --> 00:24:15,462 Exactly. 362 00:24:15,552 --> 00:24:19,182 And, And from that position, you know the answer for how to transform your 363 00:24:19,182 --> 00:24:21,192 enterprise, it will become self-evident. 364 00:24:22,002 --> 00:24:27,822 Because you will see which aspects of the work can be automated AS-IS, which 365 00:24:27,822 --> 00:24:30,702 ones will continue to be done by humans. 366 00:24:31,122 --> 00:24:35,232 And the next immediate vision, I think, for the enterprise is the 367 00:24:35,232 --> 00:24:40,422 world in which you have humans and agents working alongside each 368 00:24:40,422 --> 00:24:42,402 other to execute this process. 369 00:24:42,852 --> 00:24:46,242 So the humans are gonna continue to do the parts of the process that require 370 00:24:46,242 --> 00:24:48,282 intelligence and complex decision making. 371 00:24:48,792 --> 00:24:52,662 The agents are gonna do the rote, mundane aspects of the work that 372 00:24:52,662 --> 00:24:54,842 can be done by a robot essentially. 373 00:24:55,392 --> 00:25:01,062 And so your goal as the leader is to really understand your process and 374 00:25:01,062 --> 00:25:04,752 understand which components of it can be given to the agent and which will 375 00:25:04,752 --> 00:25:06,372 continue to be done by the human. 376 00:25:06,612 --> 00:25:10,302 And if you can understand the process in enough detail to understand where those 377 00:25:10,302 --> 00:25:17,202 handoffs should be and what that new sort of hybrid workforce looks like, then 378 00:25:17,202 --> 00:25:19,032 you can begin to execute on that vision. 379 00:25:19,192 --> 00:25:19,672 Perfect. 380 00:25:19,672 --> 00:25:20,812 Very good advice. 381 00:25:20,902 --> 00:25:24,082 So as we're about to wrap up, I usually ask a couple of questions at the end. 382 00:25:24,652 --> 00:25:29,062 And the first one is what book, tool or habit has made the most 383 00:25:29,062 --> 00:25:30,682 impact on you in the last 12 months? 384 00:25:31,695 --> 00:25:36,525 It's gonna sound cliche to say it, but the tool that has had the biggest impact 385 00:25:36,525 --> 00:25:39,965 on my life has definitely been ChatGPT. 386 00:25:41,500 --> 00:25:43,960 I use it for everything. 387 00:25:44,173 --> 00:25:49,213 There's probably no decision in the company that I make today without 388 00:25:49,213 --> 00:25:53,093 consulting ChatGPT, its reasoning models. 389 00:25:53,213 --> 00:25:57,713 Whenever I am trying to solve a new problem, whenever I'm trying 390 00:25:57,713 --> 00:26:03,443 to think about a space that I'm unfamiliar with, I always use it as a 391 00:26:03,443 --> 00:26:05,363 starting point for my brainstorming. 392 00:26:05,903 --> 00:26:09,818 And even when it's incorrect, and it's not always correct, but even 393 00:26:09,818 --> 00:26:14,978 when it's incorrect, it always provides or brings something to the 394 00:26:14,978 --> 00:26:17,408 conversation that was not there before. 395 00:26:18,038 --> 00:26:22,808 So I would highly encourage everyone to become proficient in this tool. 396 00:26:22,808 --> 00:26:27,098 There's absolutely no doubt that it will increase your creativity, 397 00:26:27,098 --> 00:26:28,418 improve your productivity. 398 00:26:29,108 --> 00:26:31,388 Maybe I should be a salesperson for Open AI? 399 00:26:31,688 --> 00:26:32,633 Yeah, I think you should. 400 00:26:33,143 --> 00:26:33,503 Yes. 401 00:26:33,513 --> 00:26:34,173 definitely. 402 00:26:34,173 --> 00:26:37,983 An LLM as a thinking partner is a great way of using this tool. 403 00:26:37,983 --> 00:26:38,433 Of course. 404 00:26:38,813 --> 00:26:39,803 And then the last question. 405 00:26:39,803 --> 00:26:43,283 We've been talking about everything that is likely to change and your 406 00:26:43,283 --> 00:26:46,763 vision for that, but is there maybe one thing that you expect to remain 407 00:26:46,763 --> 00:26:48,023 constant in the next 10 years? 408 00:26:48,323 --> 00:26:56,333 There is a lot of discussion right now about what the future of AI looks like. 409 00:26:57,623 --> 00:27:01,283 What I think will remain constant and I feel very confident is in 410 00:27:01,283 --> 00:27:05,033 this is our reliance on people. 411 00:27:06,387 --> 00:27:11,003 So I joked earlier about a world where the humans are all sitting on a beach and 412 00:27:11,003 --> 00:27:14,843 the agents are doing the work, but it's not a vision that I truly believe in. 413 00:27:15,423 --> 00:27:19,823 I believe that people will always have a place, in this world. 414 00:27:19,873 --> 00:27:23,683 And there are some sort of AI optimists, let's say, that believe 415 00:27:23,733 --> 00:27:26,623 that, that humans will be automated and not have anything left to do. 416 00:27:26,623 --> 00:27:29,623 And I disagree with that notion. 417 00:27:29,643 --> 00:27:32,543 And that's something that I think will absolutely remain constant in 418 00:27:32,543 --> 00:27:35,723 the next 10 years, regardless of all the advances that we see in AI. 419 00:27:36,023 --> 00:27:36,983 You and me both. 420 00:27:37,193 --> 00:27:38,213 I also hope the same. 421 00:27:38,783 --> 00:27:41,273 Tuhin, thank you very much for this great conversation. 422 00:27:42,475 --> 00:27:43,195 Thank you so much. 423 00:27:43,195 --> 00:27:44,365 It was a pleasure being here. 424 00:27:44,887 --> 00:27:49,267 What resonated the most in today's chat was Tuhin's insistence that numbers 425 00:27:49,297 --> 00:27:51,487 not narratives should drive automation. 426 00:27:52,057 --> 00:27:53,257 Capture the keystrokes. 427 00:27:53,377 --> 00:27:54,427 Quantify the waste. 428 00:27:54,637 --> 00:27:58,777 Then identify where planned agents could deliver undeniable payback. 429 00:27:59,857 --> 00:28:03,667 So how might you translate your next AI initiative into a metric 430 00:28:03,667 --> 00:28:05,107 the CFO actually champions? 431 00:28:06,117 --> 00:28:09,807 If you're ready to test the waters, Tuhin and the Mimica team offer a proof 432 00:28:09,807 --> 00:28:14,307 of concept assessment and a free CXO guide to data-driven cost cutting. 433 00:28:14,667 --> 00:28:16,857 The links as always are in the description. 434 00:28:17,967 --> 00:28:22,077 In the meantime, we have more exciting topics and guest appearances lined up, so 435 00:28:22,077 --> 00:28:26,787 stay tuned for more tales of innovation that inspire, challenge and transform. 436 00:28:27,477 --> 00:28:29,337 Until next time, peace. 437 00:28:30,594 --> 00:28:32,934 Thanks for tuning in to Innovation Tales. 438 00:28:37,674 --> 00:28:41,364 Get inspired, connect with other practitioners and approach the 439 00:28:41,364 --> 00:28:43,404 digital revolution with confidence. 440 00:28:44,334 --> 00:28:47,724 Visit innovation-tales.com for more episodes. 441 00:28:49,064 --> 00:28:50,084 See you next time.