#1 Demystifying Hyperautomation PART 1

Hello and welcome to another episode of the Process and Automation Podcast with the Automation Guys. Today's podcast is all about Hyperautomation. You're going to demystify Hyperautomation and peel off Hyperautomation bit by bit so you get a good understanding of what technologies are within hyperautomation, what it is, what you can do with it, what you actually get out of it from a business perspective. So, Arno, what do you think? What is this hyperautomation? Why is everyone going crazy on this new buzzword?

I think there is actually quite a lot of confusion when people talk about this term hyperautomation. Out there. I think if you ask 10 different professionals what they think it means, even if they have some subject knowledge, you'll probably get 10 different answers. You know, for me, I guess hyperautomation can be something which people view, you know, in, in context of the things they've read about it, maybe what they've heard their colleagues talk about, maybe what they've seen, what the analysts are talking about. But in essence, for me, hyperautomation is getting automations into your business at a very, very rapid pace. So from ideation, where we look at a problem, to actually execution and production— productionization of that And that should be something that should be very fast. And in order to achieve that, that's where you would utilize or even exploit some of the, um, the progressive automation technologies. And those include what I would consider the, the three pillars of hyperautomation myself. Which is artificial intelligence. That's obviously a big topic on its own, but it's basically having a neural network that could make cognitive decisions on behalf of a user or react on a user's input and provide some guidance as to, you know, what decisions should be made and make recommendations. So AI is quite a big thing. It's quite a big buzzword at the minute. AI is very easily accessible. It's very commoditized. So you could have as much AI as you need by doing subscriptions, or you could build your own AI models. So that, that's quite a big, big topic in itself. And I think that will only grow in the next couple of years. And then also the concept of process mining. Now, process mining is actually a way that you could look inside your business and discover things that is broken. So you could discover processes that's broken by looking at these, at your systems, and the process mining software will sort of kind of steer you in a direction to tell you, well, you know, I think this process here, it's taking very long, you know, that's a good candidate to do a bit of investigation. Okay. I guess the third pillar is what we called, or what we call, robotic process automation. And that's, that's literally training a piece of software to behave like a human, teaching it to click on applications, teaching it to log on to different systems and do stuff like a human would do that, but obviously very accurate and, you know, up to 20, 30 times faster than a human. So I think if you combine all of those things together, find the right problems to solve, because the technology is out there. All you need to do is find those problems, apply those technologies rapidly, then you officially on your hyperautomation journey. And, you know, you could stretch that out as long as you want to. Because the benefits that it will start to provide you would obviously unlock more of your staff's time. They can contribute more towards your hyperautomation program, and you can have more initiatives. I mean, what's your view, Sascha?

I think, I think the big, the big difference to, to other automation topics— so obviously automation is out there for a long time, so hyperautomation is now, now really I think putting all these things together to really achieve this end-to-end kind of processing much better than just these individual things we looked at before. So now when you think about process automation, an invoice comes in, so first of all you need to OCR it. So this was in the past maybe just OCR and then someone did something with it, but now with all the disciplines of Hyperautomation, what we classify as hyperautomation these days, we can truly achieve end-to-end processing in the organization. So really going from modern systems into legacy systems, so there's no barrier anymore with RPA for us. So I think, yeah, combining all these things together, that really is the game changer for lots of businesses. We had lots of success just with workflow, just with AI, but now combining all these things together and then using, for example, process mining to actually figure out that you have made it all better for the business is quite fascinating. Yeah, what do you think, I know, are sort of the best use cases? So I just mentioned scanning OCR. Now, do you have any sort of particular use cases businesses would use HyperAutomation for?

I think every business can potentially benefit from it. I think HyperAutomation is sort of a a culture, a cultural thing within inside a business. It's a way, a way to think about how you run your business. And I think this, this, this, there's no, there's no one-size-fits-all for this. It can be as small as it needs to be on day one, or it could be really large and very transformational, you know, towards the, the maturity of, of your hyperautomation life cycle. In terms of use cases, I think if you look at businesses where you need to connect journeys together. So if you look at finance, for instance, there's a lot of processes within finance. If you take your invoice example, where it's, you know, accounts payable, you would typically, you know, the process that precedes that is going to be supplier onboarding. So supplier onboarding, there's going to be a lot of things in there that's perhaps manual where you onboard a supplier. There might be some contractual things you need to do and some legal due diligence you need to do. So it is a bit like you know, when, when you look at one particular area within finance, it's so connected with other things that, um, that is, you know, other dependencies that you need to, to really choose, you know, how much do I want to bite off initially, um, and how ready is my business to start embracing this. Now again, for the, the finance example, it could be as simple as you know, you've got a shared inbox and everybody has seen these shared inboxes across the business where, you know, suppliers email invoices into. And, you know, a typical scenario would be that somebody looks at the invoice, they recognize that this needs to, you know, be approved. They send it to that particular person for approval based on the invoice amount. They need to make sure that the customer account and all the invoice details are, are correct. Um, so there's a lot of things in there that needs manual checking. So I think that's a perfect example of where you could plug a tiny bit of automation in. Um, I wouldn't call it hyper automation at that point, but I think as you, as you start evolving that process, and rather than somebody, uh, looking at that inbox Perhaps you train a robot to monitor the inbox, to listen out for invoices. When that invoice is received, you could then send that to another robot that does the OCRing of the invoice. Yeah, you know, once you've got the OCR back, perhaps you could have an artificial intelligence model that makes sure that the the unstructured data, because of course every invoice looks different for every supplier. But you could deploy a bit of artificial intelligence to extract that structured data from an unstructured document and then pass that, you know, along the next step of the process where another robot might decide, does this need to go to human for review? Or is it below the threshold? Can I just automatically approve this and submit that to finance for paying?

Yeah, you mentioned unstructured data. I think that was for many— I think when people started with RPA, we were not talking about AI back then and other things. So the unstructured data was really a big, big problem because the RPA could do all these things, but only if they understand the data. So if the data is already there in a good structure, then the bots can really do good stuff. But yeah, what are you doing with these invoices? They all— always contracts, yeah, by nature they all unstructured and then quite useless to a bot in itself. So yeah, that was very important. So when OCR came back, right, so a couple of years ago, to become very, very sexy again. Yes, really, really.

Yeah, and I guess intelligent OCR.

Yeah, exactly.

I think if you, if you look at, if you look at what RPA vendors do and, you know, the sort of things that I guess is in their roadmap, you'll see a lot of them are actually investing in artificial intelligence because they know that that's the next step. The next step is to take robotic sequences where you train a robot, a software bot, to do stuff, um, and, and really enhance that by giving it some cognitive behavior so it can actually understand the difference between you know, an O and a zero on an invoice and make a decision to say, well, this was OCR'd wrong. This is definitely a zero. It's not an O. You can make that decision and the process doesn't break down there where, you know, you've got a piece of data which is compromised, you know, the whole process at that point. And you need to pull a human into that to make a decision to say, well, this is a zero because I can see that. So a lot of the vendors are investing heavily in AI, acquiring AI, incorporating AI with inside their platforms, because that is such a key element of hyperautomation or the hyperautomation story.

Yeah, actually, you mentioned earlier the invoice scanning. I think a good example we had recently as well was just pure customer support. So people think, okay, all that stuff is all very fancy and only for really big use cases, but this was just customer support. And if it's a big customer support customer, obviously it makes huge sense to be more efficient, even if it's just a minute here and there for every customer request. So yeah, so what we did there, whatever email comes in, whatever inquiry comes in, either chat or email or any other channel, So the message was automatically identified, understood by the AI, by some natural language processing to understand what is actually the context of the document. And yeah, and then based on that, but now after understanding all these emails, we, yeah, we could just change customer data if it was, for example, a request to change the name after getting married. So that kind of stuff, and even automatically responding meaningful, not just a typical one, hey, thank you, we have received your message, more like a proper meaningful email back to customers with the right information. So these are very simple things where RPA can then make these changes in all these customer systems. Some, some big companies have hundreds of legacy systems out there, and RPA can do all that stuff in a minute or so.

Yeah.

And yeah, so that was quite, quite impressive. Everyone was really, really amazed what it could do.

So for our listeners that has never looked at RPA or know what RPA stands for, and I know we've been talking about software bots But how would you describe that in a very, very layman's term? Like, what— how do you visualize that over a podcast recording for perhaps somebody that listens to this and say, you know, I want this thing, this RPA thing, as part of this hyper automation? What does it stand for and what does it do?

Yeah, so RPA in itself, yeah, that stands for robotic process automation. And when we talk about the bot which is, which is a central part of it, is it is really capturing or learning what a human is doing on, on, on the user interfaces. So this is, this is a key bit. So really capturing, learning what someone is doing, and then this, this will be all put into the system and it can be repeated by, by the bot. In the same way again and again and again. So obviously there's a bit more intelligence going in there, but effectively it's recording what a normal person would do day in, day out, and then just repeats that autonomously at any time. So obviously then, yeah, you can run it 24/7 and you can have another sort of a colleague bot coming in to do exactly the same as what someone else has learned or the system has learned. And then suddenly you have two bots doing exactly the same. Yeah, you can scale up sort of more and more, but effectively it is useful for capturing these user activities. Someone is doing in the office high-volume work, very repetitive work, very rule-based work. So it's a very, very standardized structured work. So that can be then used by the bots to run it again and again.

Yeah, I mean, I've heard an analogy which I think is quite a good one. I think most people have heard about Excel macros.

Oh yeah, yeah.

And I think that's probably the closest you can get to drawing some sort of comparison with a software bot. Whereby you could see it as a macro, but the macro is not just restricted to Excel. It can actually interface and do things in virtually any other application. So if you ever created a macro, you would know that you could record a macro. It will run and crunch some numbers for you so you don't have to do that, and it does that lightning fast. So, so I think this, this is a good way for me to, to describe it, where it's, it is sort of a macro, and it is all rules-based, but it's very accurate, and it can reach into virtually anything as long as you tell it to do that. And that's what we call the training. So when, when you look at a software bot, like a human, you still have to train it. Now, within Site, the, the software bot itself, when you create it, there is— in most software, robotic software, there's, there's like a training module, and that's kind of where you assemble your macro. Now, of course, the macro itself is only as good as the implementation, or the bot will only be as good as the training. So, so you have to be quite specific. Now I guess when we talk about intelligent bots, I think that's where we actually look at these macros or these robot training sequences and that the training you give these robots and give them decision-making capabilities. So for example, if, if something fails, it's trying to do something. It could make a decision based on cognitive models because it might look at the failure that happened in the past. It knew what was, what was done with that particular failure. It spotted something that's very similar. And then statistically, because of the similarities between the two cases, you could train that bot almost to react the same, right? And I think that's where the kind of the link with the AI or artificial intelligence comes in, in my view. So it's this kind of this evolution, isn't it, Sascha, of how we kind of look at RPA today? But it's really a kind of an old concept.

Yeah, it goes back, I think the first one came up in the '50s really as an RPA. So I can't remember the guy who invented it really, but yeah, and then before became like a really hot thing. Yeah, I think it was 2012 or so, it came, started to become a bit more, more exciting again, and especially over the last couple of years. Crazy, crazy stuff. So, so we talked about the, the RPA element and then in the AI. Most people think with the AI, so you have to teach it a lot, so similar like with RPA, we have to train the system a little bit, but with all the AI, so lots of people say it will take always ages to train it to be effective. And what do you think, how can we make the most use of AI? In these scenarios, in those scenarios?

Yeah, I mean, I think AI, you, you need to make a decision for your particular use case or your particular scenario. In, in my view, what we tell our customers is, um, there are a lot of commoditized AI out there that could help you with Computer vision, and computer vision is, is an AI capability where you pass it in an image of somebody's face and it tells you if that person is happy or sad. You pass it in an image of a car that's had a crash and it tells you the damage, what it thinks the damage to the car is. Um, you know, you send it a picture of a structure and it tells you something about that structure. So a lot of that computer vision, you know, that's one example of a commoditized AI. So that could be something that fits nicely into a particular use case, especially insurance, where you could use AI to do some preprocessing of insurance claim, right? And almost, you know, let's look at an example where, You've got insurance on your mobile phone. You dropped your phone, the phone is damaged. You raise a claim, you take a picture of the damage on the phone. That goes through the AI model. It determines, yep, this is a phone that has clearly been damaged. It's within sight of parameters of the claim. The claim is up to date, we can automatically pay that insurance or the insurance claim. We can settle that claim. For the end user, that's great. I'm in distress, my mobile phone doesn't work. I think most people will go into panic stations. But that's that type of journey you want to create using that quick decision-making. I mean, that's just one example of sort of commoditized AI.

Yeah, yeah, they actually, um, yeah, available, isn't it, from, uh, from the, from the big vendors, isn't it, like Google or Amazon and Microsoft, isn't it? So they, they make computer vision easy accessible, um, yeah, and then, yeah, yeah, so that the learning has been done already by someone, isn't it? So we can just get it, get it, get it in there and have immediate benefits. Yeah, it sounds really cool.

And I mean, I think with, with commoditized AI, the beauty of it is the fact that it, it's priced reasonably. So, you know, different providers obviously provide, you know, different structures. So, you know, it is literally pay as you use. And also it supports continuous learning. So if you do do specific things and you see your model perhaps don't predict things the way it should, then, you know, you could provide feedback and that artificial intelligence engine would take that on board and it would make it better. I think the other option, so if you feel, well, I don't want my data in Microsoft Cloud or IBM Watson. You know, I want to do this myself. You know, there's a lot of sort of AI, even open source AI, or, you know, within sight that the big vendors like Microsoft, Oracle, Java, these guys, that you can create your own models and you could train your own models by providing it data. And creating those neural networks to make the decisions. And a good example of where we used this before was for taking unstructured data where it was card cancellations that was sent through a spreadsheet. And by nature of a spreadsheet, there's always a lot of mistakes in it. So we created, by using historic data, we created a model that could predict what the intention was for a particular card. So if a card number comes in, even if there's a mistake in the number, it can sort of predict what what the actual number is. And it would then give you that structured data you need in order to down the line actually pass that to a robot. Now, that was something very specific, a specific use case. There wasn't a commoditized artificial intelligence model for that. And in that particular instance, you know, we decided, well, we can develop that from scratch because that was such a typical component, you know, with, with, within inside that particular use case.

So that's it for today. If you like the podcast, we would really appreciate if you could rate this with 5 stars on iTunes. What can we do better? What topics do you like us to talk about? Feel free Feel free to connect with us on LinkedIn, Twitter, and Clubhouse, and visit theautomationguides.net to secure your free automation ebook. Thank you very much for listening.

Let's automate it.