← Embracing Digital Transformation
Loading summary
Transcript70 lines
- [00:00]
A
Well, so there is. As economies have evolved over the last several hundred years or even several thousand years, I think we have gone through a similar structure of increasing productivity because of reskilling because of different areas that humans want to focus on. And eventually each of the revolutions that we have gone through from an industrial standpoint has led to not only technology breakthroughs, but also humans becoming more productive and actually becoming happier by embracing the changes brought about by these technology revolutions.
- [00:37]
B
Welcome to Embracing Digital Transformation, where we investigate effective change, leveraging people, process and technology. This is Darren Pulsford, chief solution architect, author, and most importantly, your host on this episode, the Future of Work AI's role in IT management, with special guest Sharish Nimgankar, founder and CEO of Heedless AI. Suresh. Welcome to the show.
- [01:05]
A
Darren. Great to be here. Thanks so much for having me.
- [01:08]
B
Hey, when we talked, I was like, hey, this is really interesting. We found some practical uses for generative AI and AI in general in the workforce. And then I thought my brain went right to, oh no, are we replacing workers? And the answer is, of course we are. Jobs are changing, all that. But before we dive into that, because that'll interest people a lot. On my show, I only have superheroes and every superhero has a background story. So, Suresh, what's your background story?
- [01:39]
A
Well, Darren, I grew up loving the process of experimentation. At a very young age, I started experimenting with different kinds of science projects. Used to experiment a lot with music as well. And then I went to college to study engineering, came to the us, Stanford. I was in Silicon Valley for a couple of years and that was my first foray into the practical applications of technology experimentation. I worked with companies like intel, and then I was a part of four different venture capital and private equity backed companies. Also worked in investment banking for a couple of years. And what I realized over the course of a variety of my experiences is that what matters is a good understanding of what is the pain point or a set of pain points in the market and what might be ways in which you can run rapid experiments to really get an understanding of the pain points, to come up with practical solutions that solve those pain points and in that whole process create some compelling and differentiated value for the customer.
- [02:42]
B
So that whole concept of fail fast, learn quickly, that's kind of what you've adopted then?
- [02:49]
A
Absolutely. And over a period of time, there have been ways in which we've had a chance to get better at it. Clearly you continue to learn because every experiment is different. And some unique challenges that are posed in Every scenario. Having said that, the whole love for experimentation and the desire to come up with compelling and differentiated value is what drove me to my current venture and would love to talk a little bit more about that as we go through the next hour.
- [03:24]
B
Yeah. So before we dive into what you guys do now, let's talk about the problem space a little bit. The problem space is it. Right. And there's just so much complexity involved in supporting it and devices and, and all that stuff out there. And you guys found that even with all that complexity, you found some common patterns. Is that kind of the best approach to.
- [03:51]
A
That's. That's absolutely the case. So think about the problem statement in a very simple, simple way. We're surrounded by devices. Over the last several years, there's just been an explosion in the number of devices. There's so many different varieties of devices, right. So you have not only your laptop, cell phone, tablets, but on the manufacturing shop floor, you have a variety of robots as well. Even a car is essentially a device. Now think about a Tesla, or from a retail standpoint, your point of sale, terminals. And then you have the IoTs, the Internet of things too. And in the home environment, you're surrounded by smart dishwashers and washing machines. And so pretty much everything is a device now.
- [04:35]
B
Yeah. You know, I know that for a fact because on my WI fi, my home WI fi, I have 75 devices on my WI Fi. That's crazy. I don't have 75 laptops in my house. But everything's smart now. Everything's connected, right?
- [04:50]
A
It's absolutely, absolutely. So now you have more devices, you have more varieties of devices and then more distributed geographies for these devices. Right. So, you know, people working from home from so many different geographies. So that sort of increases the complexity of what it is required to actually effectively manage those devices, decrease the downtime and make sure that the users are actually more productive. So while the device number and variety is exploded, the device management model is still very, very old school. What that means is that there is sort of a. The objective is to sort of look at the symptoms of the issue, right? For instance, if you're sick, you might have fever. Fever is actually a symptom. It is not really the root cause of the problem.
- [05:37]
B
Right, right. Yeah, yeah, yeah, yeah.
- [05:40]
A
And so in most of the device management issues, they sort of monitor the symptoms of the issues. For instance, my outlook is crashing or my device screen is sort of blurred or is blinking. That is not really the root cause of the problem. So what is required is not only detecting the symptom, but really moving to the stage of diagnosing what the root cause of the problem is and then coming up with the effective solution that fixes the root cause of the problem. So in most cases, what happens in the real life environment is that because you don't really understand the root cause, the response to device management issues is slow, delayed, it is reactive because it happens.
- [06:29]
B
Let's talk through one of those scenarios. Because it happens all the time. If you're in a big corporation or even a small one. My machine's not working. You call a support desk and they have a script. I know the script. Right. Did you reboot? Yes, I rebooted. Did you power cycle? I powered cycle.
- [06:49]
A
Right.
- [06:49]
B
It's still not coming up. All right, now what kind of machine do you have? Hold left, shift, you know, cross your fingers and power on or, you know, whatever. They have all this criteria that we go through to try and get the system up and running, but not solving the root cause of the problem.
- [07:10]
A
That's absolutely the case. So the script they run through is actually called a heuristic solution. Right. A heuristic solution is an approximate solution that can sometimes address the problems based on symptoms, but because it doesn't really understand what is the root cause of the problem, you have to go through sort of a series of steps to even get to a possible solution, number one. Number two, you don't know whether you've actually solved the problem or not. So we have this, right?
- [07:38]
B
And someone else could be having the same problem. And.
- [07:45]
A
Foreign.
- [07:51]
B
We saw this with, what was it? Crowdstrike. Right. The whole crowdstrike issue that happened brought down a whole airline because they didn't root cause it. It wasn't root caused in the airline fast enough. And I, I see it all the time in, in large organizations, they throw a banner up on their help desk page that says, we're having major outages right now. That's all you get. Just wait. Right? So that, that, that's the old way of doing things. Right. That's kind of the. Okay, so there's not a lot of information sharing between help desk agents. They're. They're looking at symptoms and, and giving you Tylenol for that fever, even though you may not know what's wrong. Right. They're just saying, all right, reboot. Download this one packet because I know, I heard it worked over here because I'm sitting next to another help desk person, whatever the case may be. So that's how help desk really is today is that, that is, that is.
- [08:54]
A
That is absolutely the case. Right. So now let's look at an analogy where you go to a doctor. It's a doctor does a series of blood tests on you. They typically might monitor 20 to 25 different parameters. They're looking at you, glucose levels, your white blood count, your cholesterol, and so on and so forth. Each of these parameters is something called a normal range. And let's say for you, parameters AC and E are outside of the normal range. Then the doctor looks at what you would define as an abnormality. So someone has to A, tell you, okay, Here are the 25 parameters you're tracking. B, they have to define what is called a normal range. And what we have realized is that while there might be a generic normal range, every person has their own individualized normal range as well. Right? That is called a personalization, essentially. And so it's not just enough to look at sort of a general normal range, but we need to look at a personalized normal range. Right? That is the detection and hence figuring out whether it is an abnormality or not. Then the next step is the doctor looks at the parameters which are abnormal, let's say A, C and E. And. And then the doctor does the diagnosis to say, okay, here is the possible root cause and then let me prescribe a solution or a medication that will hopefully bring back the abnormal parameters within the normal range. Right. And then the holy grail is, oh, can I prevent this from happening in the future? Right. So you're essentially looking at the detection, diagnosis, remediation and prevention. And then the question is, can you actually automate this entire process for the different kinds of issues that you might see in a variety of end user computing devices?
- [10:34]
B
Well, that's a good, that's a very good question because in the minute, let's stay on the medical field thing because this is really interesting. In the medical field I have specialists, I have generalists and then I have specialists. So when a problem is really bad, then I send it off to a specialist. Like for me, I have a cardiologist because I had open heart surgery eight months ago. So my protocol and all my numbers are now different than anybody else's. In, you know, normal circumstances, I, I have to have blood work more often. I, I have certain levels I have to keep. I've got tons of pills that I take every day. That's normal for me now, that's my new normal, like you were saying. But I now have a specialist helping me out because I, I have A unique situation, but I've noticed that IT organizations have moved away from specialists and gone to generalist. So there's a mismatch. A mismatch there.
- [11:32]
A
Right, so you're absolutely right. So an effective solution requires a generalist and a specialist as well. Right. And so for instance, when you, when you sort of, when the first level of diagnosis needs to happen, you would typically go to your primary care doctor. Right. And then they would actually look at all these blood tests and they will say, okay, here is a possible range of scenarios from a potential solution, from a potential issue standpoint. Right. And then they would maybe guide you to a specialist. So essentially an effective solution for automation requires a good combination of the general skills and also specific sub issue domain specific skills or the expertise. And essentially that is what we are trying to do in terms of the kinds of solutions that we are building, which is really effectively automate the entire process of detection, diagnosis, remediation and prevention of a variety of issues in different computing devices. And so the general trends are with respect to understanding patterns across the different issues and across different devices and the specific domain related issues are about do we understand these kinds of applications, or do we understand these kinds of operating systems, or do we understand these kinds of specialized devices as well? So there needs to be an interplay of both in order to effectively solve these problems.
- [12:56]
B
So do you see that tools like yours and AI enabled will, will become more and more of the generalists. So they're doing the triage, they're doing things like that and then pushing it down into working with an IT specialist. Is that kind of how you see the marrying of AI and human interaction?
- [13:20]
A
That's absolutely the case. So we look at different categories of sort of issues and the extent to which we can add value. Right. There is category one, where these AI platforms are really good at automating the entire process end to end in terms of detection, diagnosis, remediation and prevention of these kind of issues. And there's category two, where they can maybe do the detection and a little bit of the diagnosis and they can possibly recommend solutions, but they still fall short in terms of closing the loop. And that's where you have the humans come in. And there's a third category which is sort of a novel category where they might be less informed in terms of even doing the detection side. That's where you need more of the human expertise to come in. And as AI evolves over a period of time, our hope is that the percentage of 1 and 2 will increase.
- [14:17]
B
Right.
- [14:17]
A
And so you're sort of building this base level capabilities and also adding in more of the domain expertise.
- [14:25]
B
Well that, that makes sense because we know AIs are really good at reinforced learning, right? So they can learn as you, as they interact with, with humans and, and detect patterns that they've seen before. They can project potential patterns. But if, if the pattern's never been seen before, that falls back on the humans because we have intuition, which we know AI does not have intuition yet. They, you know, they, they can't guess the, you know, they can't, they don't have that intuition to go outside of the bounds of, of their programming, at least not yet.
- [15:07]
A
That's correct. That's correct. So you know what, what AI is really good at at this point is sort of taking care of the low hanging fruit, which is sort of the more monotonous task and then attempting to recognize some of the patterns as well. If you're trying to look at patterns or large volumes of data, then clearly you would need robust AI platform to do that. What humans are really good at is creativity and complex problem solving that an AI will continue to evolve in terms of the complexity of problems that it can effectively solve and will be able to solve in the future. So the way we address some of these issues issues is really there is a known set of problems which AI can actually effectively handle end to end. Right? And then so we, we create something called a self learning knowledge base where we start by saying, okay, here is a list of issues that we can solve on day one. And then let's say if we encounter a problem or an issue which is one of the issues that we already know, then we can, we can trigger the solution. If it's, if it's a new kind of issue, then our agenda AI framework kicks in and then we engage in a process of reasoning with the end user or the IT admin to really have a better understanding of what the problem actually looks like. We take a complex problem, we break it down into simpler set of problems and then we use.
- [16:38]
B
No, I like what you said there because I want to touch on that a little. It's the interaction between the human and the AI where the AI is actually prompting the human.
- [16:47]
A
Now that's great.
- [16:49]
B
I've seen more of this in these reasoning models where they're trying to get more information out of the user that maybe in the initial ask or the initial prompt that they gave the AI there just wasn't enough information or the user doesn't know how maybe even to describe the problem. That they're seeing. So now the AI is working with the human together. I think that there's a whole power in doing that that I think will unleash a lot of really cool new workflows and new ways of working.
- [17:25]
A
That is correct. And so if you think about where the gaps are in terms of effectively solving the problem, it starts by better defining what the problem is, right? And in many cases, AI may not be able to even recognize the problem. So if a human says, let's say my outlook is not starting, or said that there might be some other issues, then what it requires is a process of reasoning and engaging with the end user to take that problem, break it down into similar problems, get the user feedback with respect to the nature of problems. Then the AI proposes a series of solutions, and then one by one, you run through the process of experimentation to understand which of those possible solutions might be applicable in this current context. Because remember, a problem is not always a generic problem. It is a personalized problem. It requires a generic understanding and a user specific understanding as well. At that point of time, you take both of that, you run them through the reasoning approach, then you propose solution, and then you figure out which of the solution works. And that requires a verification of the solution as well. You ask the humans to verify if the solution works. If it does, then it becomes a known problem and a known solution. Then it gets added to the knowledge base. So next time you see a similar problem, then you know, okay, we have.
- [18:48]
B
Seen this problem, I've seen this before, something similar, right?
- [18:52]
A
So that's how the self learning models actually work.
- [18:56]
B
So that's pretty cool. And it shows the importance of humans. But just like what happened in the 90s and the early 2000s when companies were offshoring jobs and they were asking employees to train their replacements, right? How horrible is that? Right? Oh, we're gonna, we're gonna fly you to Costa Rica or we're gonna fly you to India or China and you're gonna train a team of people to do your job and then you're not going to have a job. Are people going to feel the same way about AI? Why would I give AI my information? Right? Because it's going to take over my job. Isn't that a fear? Have you heard that fear in some of the people you've been talking to?
- [19:48]
A
Yes, we do hear that fear. And I think what we have seen is because AI is here to stay and AI is going to enable significant transformations across our different kinds of industries and our life as well. I think the most effective way to think about AI is actually it's a platform or methodologies and framework that actually improve our life, make us more productive and actually make us more happier as humans. So essentially AI is really good at focusing on the mundane, low hanging stuff and over a period of time may be helping you either in terms of assisting your day to day work or augmenting your life in terms of making you more productive and happier. So you might want to allow AI to do the kind of tasks that don't really give you joy or pleasure or happiness. And humans can focus more on creativity and complex problem solving and building relationships that can actually provide joy, enrich your life and lead to longer sustained happiness. I think that's the way we do it.
- [21:06]
B
So I'm glad you brought that up because AI, you're right, AI is good at the mundane and that's where we should be focusing is. Look, I don't want to do triage because triage is hard and it requires a lot of information. I would rather do the complex, I'd rather be the specialist. I would rather handle complex problems and interact with people and help them solve problems because the gen is going to help identify what the problems really are. So it, but when push comes to shove, I don't need as many IT help desk people. Right? That's great.
- [21:52]
A
But I think the way to also think about it is the people who will remain will be way more productive and they will actually provide a much better customer experience because they are not only solving the problems effectively.
- [22:05]
B
No, no, I get that. But what about me as an individual? Right? If they come and slash 90% of my help desk and I was in the help desk, what am I going to do with my job now? Right? Help desk jobs are gone.
- [22:19]
A
Well, so there is. As economies have evolved over the last several hundred years or even several thousand years, I think we have gone through a similar structure of increasing productivity because of reskilling, because of different areas that humans want to focus on and eventually each of them. The revolution that we have gone through from an industrial standpoint has led to not only technology breakthroughs, but also humans becoming more productive and actually becoming happier by embracing the changes brought about by these technology revolutions.
- [22:56]
B
Well, and it means different jobs too.
- [22:59]
A
That is correct. That's correct. So what it means is that reskilling, either you focus on higher value added activities within the same sector or you look at different areas that you want to focus on. You know, one thing is clear, that now AI is going to help in a Couple of areas that it's just not replacing.
- [23:17]
B
It's not replacing everyone.
- [23:19]
A
Well, yeah, so it's probably replacing some of the tasks, but it's not going to replace the humans. Right. And so you're looking at automation, which is going to replace some of the sort of the lower level task. And over a period of time it's going to also assist in terms of helping humans do certain tasks better. Right. Where the human and the machines have to have some kind of an interplay and then it's going to augment humans or help them to look at kinds of areas that they never thought would be possible.
- [24:00]
B
No, no, I, I think this is really cool. Now, we haven't talked about prescriptive, not prescriptive, preventive, preventive stuff, right? We've talked about identifying problems and fixing problems in the interaction. So to me, this prevention thing is where the money really is, right? That's where I'm really going to save because that will increase productivity of the individual. They'll have less downtime and things like that. So tell me, how does that work? I've identified problems, you know, how do I move from problem to solution identification to prevention.
- [24:36]
A
Right. And so if you think about the entire value chain, you start by figuring out or trying to identify issues that you're able to detect and then diagnose. I think the diagnosis is really the key because that's what leads you to understanding what might be or what are the root causes of the problems here. And then once you start drawing a trend line of the different root causes, you start seeing enormous amounts of patterns across different devices and different kinds of issues. And then you extrapolate the patterns over a period of time. You also know which of the problems you have been able to solve from an automation standpoint. So think about the analogy we talked about, which is what are the parameters you need to track from a health standpoint point right in your blood testing, and then what might be the normal ranges. And then once you start seeing huge amounts of data with respect to that, you start seeing and which of those abnormalities lead to certain diagnosis because of correlations. So as you sort of start tracking a lot of the data across large swaths of populations segmented by different types and so on and so forth, you start getting valuable insights with respect to how some of those trends are going to manifest themselves in terms of these abnormalities and hence essentially the faults that could lead to these kinds of issues. And because you start also understanding what the root causes might be and what the solutions might be. A combination of extrapolating the trends and understanding the root cause and problems leads you to saying, okay, I see this trend happening. The fault is going to happen at a certain point of time. Now here is a preemptive action that you can take in order to prevent the fault from happening. So I think that that is where we're going.
- [26:21]
B
So that's where you're headed now. For example, fan failures in a certain model of laptop. It's no, we've seen it before preventively. So we're not losing downtime with a user. We can recall those back in, but send another one in its place or end of life. That model of laptop that I have in my thing. So those are the types of things that we could do. We could also push out patches before they're needed and things like that. Is that kind of the idea?
- [26:58]
A
Right. So if you look at the different categories of issues that you might see, you have application related issues, you have some hardware related issues, you have OS related issues as well, and then you have connectivity issues and so on and so forth. So we track a whole range of categories and within each category you have different subcategories and so on and so forth. Right. So the whole objective is to sort of make sure that things are taken care of before they actually happen, so that the downtime is minimized and then humans are more productive.
- [27:29]
B
So I love that. So your, your key measure of success is, is decreased downtime and increased productivity.
- [27:37]
A
That is correct. That's correct. So we have a huge range of sort of outcomes that we track. Right. One, of course, is what we. And these are all linked to the performance of the devices, number one. And the most critical of that is what is the unscheduled downtime that you have seen across these devices over a period of time? And there is of course a correlation between faults or the performance. While you might not have downtime, you will have degraded performance and that still impacts the productivity of the user. So our objective is to make sure that the performance metrics we are tracking do not fall into the abnormal range. And going back to the comparison we sort of outlined earlier, and then the downtime is.
- [28:35]
B
Suresh, when you guys engage with the company, what does that engagement look like? Id, are they reaching out to you when you come in? Do you do like an analysis of their current infrastructure? How? I mean, because people may be thinking, hey, this is the way it works. Right. It's just, that's just the way it is.
- [28:56]
A
Right.
- [28:57]
B
How do you guys engage? What's that look like?
- [29:00]
A
Right. So this is actually a huge problem for pretty much every Fortune 2000 company that we have gone after. So I personally have sold the previous version of the solution, which was predominantly more monitoring solution. And just stepping back where the market has been over the last couple of years is really focused more on monitoring the symptoms like they provide. What they sell is a thermometer. They're not really sort of going through the entire process, but even then there has been a huge demand for even accessing those kind of thermometers, what we call a basic monitoring platforms or reporting framework.
- [29:43]
B
There's lots of them out there essentially.
- [29:45]
A
Right. And so to your question, now that we have evolved and we are sort of in this era of dramatically different solutions which can go through the entire life cycle in terms of detection, diagnosis, remediation, prediction, we typically focus on large Enterprises, Fortune 2000 companies, where we see most of the pain happen. We also focus on a lot of the managed services providers who from an aggregated standpoint, oh yeah, they've got thousands.
- [30:20]
B
And thousands of advisors, right.
- [30:22]
A
Who would provide us a certain size. And then typically when we, or when they approach us, there is a certain self healing knowledge base that we have already put in place because we have looked at these patterns of issues across many other customers and many of a lot of that is relevant to a particular customer that we might start having conversations with. But then we also engage with them in terms of better understanding what the distribution of their issues looks like. So that we want to make sure that there is tangible value that is provided on day one by a combination of the knowledge base that we bring to the table. But also perhaps tweaking some of our solutions to better align with what their distribution looks like. And of course when we get data from them, if some of their issues tend to be very different, then that helps to train our models better so that we can better provide value.
- [31:19]
B
Your model is constantly being updated with new. New, right?
- [31:23]
A
That's correct.
- [31:25]
B
So that's really interesting because your, your model is becoming better and better based off of incidents that it's had in several different companies. So you get, that's correct best practices from all these different companies instead of it just being laser focused on one.
- [31:44]
A
That is correct. That's correct. And so, so the expertise we bring to a particular customer is multifold, right. To sort of look at a pattern across so many different devices that we have served across so many different customers, number one. Number two is if they provide us a more personalized or a tailored set of data which is more applicable to their context, we can take that in. But then the solution also benefits from our pattern recognition across so many different other devices that we have seen. We have to understand that the way that the models work is they take a complex problem, break it down into simpler set of problems, and then it's not that you're creating every solution from scratch because there is commonality.
- [32:33]
B
There's some commonality. It may get you 80% there. Right, right.
- [32:38]
A
Absolutely. The case. So it's going to be very challenging for any particular company to replicate this kind of a platform on their own because they don't have the benefit of a, the expertise and B, the kinds of data that we have access to across so many different customers.
- [32:55]
B
No, no, that's great. Hey Suresh, if people want to find out more about you or your company, where do they go to find out more information?
- [33:02]
A
Well, you can certainly Visit our website, eblessai.com or you can check us out on LinkedIn as well. We have a page and we have several different articles about how we have created impact for a variety of our customers. And then you can certainly reach out to me on my email suresh.nindagar.com all right, sounds great.
- [33:22]
B
Hey Suresh, this has been really informative. It's good to see that we have AI that is being used for practical things and freeing up people's time to be more creative and work on harder problems. So, hey, thanks for coming on the show.
- [33:36]
A
It's wonderful to have been here. Darren, thanks so much for inviting me. Foreign.
- [33:43]
B
Thank you for listening to Embracing Digital Transformation today. If you enjoyed our podcast, give it five stars on your favorite podcasting site or YouTube channel. You can find out more information about Embracing Digital transformation@embracingdigital.org Until next time, go out and embrace the digital revolution.