News & Events
September 1, 2022

The Role of Intelligent Automation Amid Global Slowdown

Share on TwitterShare on Twitter
Share on FacebookShare on Facebook
Share on LinkedinShare on Linkedin
Aliquam auctor diam.
Privacy Policy
Watch Now
Sign up
Watch Now

Interested in the future of intelligent automation (IA)? Join Jasper Masemann, Partner at HV Capital, and Manish Rai, Marketing VP at super.AI, for a LinkedIn Live discussion on the market's current state and where it's headed.

Amid global economic slowdown and an ongoing labor shortage, companies around the world are looking for novel, cost effective solutions to a seemingly infinite number of problems. KPMG estimates that the intelligent automation (IA) market will reach $231.9B by 2025, marking a 20X increase over a period of just seven years, partially in response to these trends. From Intelligent Document Processing (IDP) and Unstructured Data Processing (UDP) to process discovery, the value of intelligent automation continues to expand and evolve.

Catch a replay of The Role of Intelligent Automation Amid Global Slowdown for a discussion on topics spanning:

  • How enterprises will adjust their investment priorities amid current market conditions.
  • Whether a global slowdown will accelerate or decelerate intelligent automation investments.
  • Automation expansion from RPA and structured data to RPA + AI and unstructured data.
  • Valuable use cases for AI and intelligent automation given market conditions.
  • The emphasis on ROI for the next generation of artificial intelligence projects.

Full Transcript:

Manish Rai Hello, and welcome to the first in a series of AI for Business podcasts. Today we are going to talk about the role of intelligent automation amid the global slowdown. My name is Manish Rai. I am VP of marketing at super.AI.  I'll be your host. Just a quick overview of my background. I've been in Silicon Valley for 20 years and in intelligent automation last five years - three years at RPA leader Automation Anywhere, one year at Low-Code Automation leader Appian, and now at the leader in unstructured data processing (UDP) super.AI.  You can think os super.AI as the next generation intelligent document processing (IDP) solution that can not only process documents better than a first generation but can also process other data types - emails, images, videos, etc. I'm really pleased to have Jasper Masemann, a partner at one of the premier VC firms, HV Capital, out of Europe. So Jasper, could you please give us a quick overview of your background and interests?

Jasper Masemann Sure. I work at a 22-year-old German fund called HV Capital. I'm based in the Berlin office, which I actually opened seven years ago. We are originally from Munich, the very nice city of Octoberfest, which is coming back this year. And we started investing mainly in consumer applications and large e-commerce marketplaces. A couple of years ago, we started going into B2B.  I was a founder and entrepreneur before. I bootstrapped my company and successfully sold it. So I would say it was a classic Silicon Valley way of joining a VC firm. I joined here to focus on B2B software, which I would say is not at an infant stage but at an earlier stage than in Silicon Valley.  We invest in a lot of AI companies because we have quite a lot of very, very good AI talent in Germany and Eastern Europe. And we can compile that here in Berlin at cheaper salaries. But don't poach them. No. We invested in ten companies so far.  The largest one is from Israel called Verbit.AI, 2 billion valuation. I'm there for a board meeting in two weeks. Hopefully, see you Manish in Silicon Valley. Super.AI, we invested a year ago because you guys were at a very nice inflection point after building the product, finding product market fit, and having this platform ready to scale. And that's where we like to come in as investors together with NFX from Silicon Valley, pushing this new platform to the market. Thank you for inviting me.

Manish Rai: Thank you for being here. So one of the first things on people's minds is, you know, how are the investment priorities going to change with the slowdown? We have seen a dramatic change in condition. Nasdaq is down 30% from its peak, S&P down 20%, to Bitcoin down 70%. So all the asset classes are down, and it somewhat reminds me of what happened during the 2008 crisis. If you think about when the downturn happened, during the financial crisis of 2008, that was the beginning of the adoption curve for RPA because priorities started shifting from growth to cost saving. So the investment went more towards automation for cost savings, and that led to hyper-growth in RPA. So how do you see this cycle playing out in your mind?

Jasper Masemann Yeah, I mean, we've seen a couple of cycles being in the market for 22 years. Actually, we started in 2000 when there was this huge bubble burst, right? And we started a VC firm that was a bit boring at that time, I would say. But I think your comparison is pretty good. This is most similar to 2008. It's actually even better what we see because if you look at the balance sheets, the cash companies have accumulated from all the profits in the past. I mean, Apple is the best example. I guess all these companies have accumulated a lot of wealth - not in travel, obviously travel industry was hurting - but many industries went pretty well. That's one. And the second one is, yes, they're down, but they actually are more back to normal 2018, 2019 multiples. Valuations were normal things. I don't know if you want to evaluate a company 100 X the forward-looking revenue. It's probably never going to happen. At least you're not getting your money back. So this is actually a pretty good time right now from a valuation perspective, but also from a company perspective. But because everybody is expecting a certain type of recession, which always happens after overheating and everybody having a party, there is a certain hangover. People are now more focused on, hey, how can we use the cash to make our business ready for the future? And the best example is in Travel, where they see a rise in demand, but they don't have much money, so they go more for automation with a company that automates customer care, as an example, with AI.  And they [this AI company] has a spike in demand from all these companies. They don't have money, but they want to spend it on automation than on real people, which they cannot afford anymore. So this is a perfect time to actually have cost-saving automation efficiency improvement placed at customers. But you have to expect slower sales cycles. Whereas in the last year, everybody was going for digitization and wanted to have it as fast as possible. We see now budgets being pushed, for example, over the holidays more into September because people are saying, hey guys, we really have to think about this. We don't know how the recession will play out yet.

Manish Rai Yeah. I mean, so that brings me to what you mentioned - digitization - and that word is used very loosely. So do you mind sharing your definition of what you mean by digitization? People talk about digitization - it's been showing up in many surveys as being a top priority for companies for the last couple of years. What do you believe the next focus will be? Focus shifting from digitization to automation. So lets, for our audience's sake, define your view on what digitization is.

Jasper Masemann Yeah, I think there are two main trends and one enablement. So one is, many companies still run on bare metal or, let's say, very, very simple server infrastructure. And that makes it very difficult to install new things to manage them, even on the whole DevOps process. That has changed during the pandemic. You wanted remote access, you wanted people to work remotely. Even the whole API economy happened, so you would basically enable the usage of more software tools. And that's happened. That's the first digitization. So we have a company that does procurement - Sastrify. So they actually help companies find SaaS tools. And we see even small startups have 50 tools in place, the larger ones - 500. So that's a lot, and it's because it's easier to buy and use them and implement them in your current processes. And I'm talking about large corporations but also smaller companies here. So that's kind of the penetration of digital tools. And the second one is because now I have easier access to my workflows, my processes, all these kinds of things. And I have the awareness, and I have the IT department that's a bit more open because they're not so afraid of security anymore. They know how to address it. They would let more things work inside the company. And that leads to all these efficiency gains through automation, also through analytics for us, because you have to figure out stuff. But then you know, okay, there's something, I can monitor it, I can see the improvement constantly, and now I can automate it and make sure that the quality at the end is still there. Because one thing about automation is you're giving away control, right? Before, there were people that feels better. But as far as I can observe it, I'm fine. It's working.

Manish Rai: Great. Thank you for defining that because people have different views of digitization. I think what we saw in the early RPA adoption cycle, the focus was on structured data. And then, when I joined Automation Anywhere five years ago, it was the beginning of the shift of focus or expansion of focus into documents. Now, I was at the early stage of launching IQ Bot and helping define IDP -  intelligent document processing - as a category. So but what we saw is primarily people were so busy and focused on structured data. Some were coming up and wanting to tackle documents, but they were not able to take advantage of 90% of the data, which most analysts predict is unstructured. And only the innovators were looking at some unstructured data-driven use cases and trying to drive value for that. When do you see adoption cycles for broader use of AI in conjunction with RPA to handle about 90% of the data, both documents, and then moving beyond documents to other types? How do you see that evolving?

Jasper Masemann: I would say, I mean, you know that better than me. But the first wave in RPA was kind of a gold rush where the gold was in the river. So you could see it. Right. So was the easy thing, as you said, structured data. I know that data is structured. I feel safe, so I can write a script that I would program. It's a rules engine that wouldn't do anything wrong because it's not **** in and **** out. It's a good thing in that I can see, and something will happen. So that was fine. Obviously, the consultants that were implementing - so it wasn't your company, but it was more the consultants - saying we do everything with RPA. Then people realize for a period of time if the data is not good enough, then, you know, we have to add certain decision-making that's hyperautomation. But that's not even good enough because they often don't know what good input is. Right. And that would reduce RPA use cases. Numbers like 80% of the cases that are pitched for RPA are not working because of unstructured data and other things. And if you look at documents like how documents were automated in the past, you use templates, and you would say, this is how the document looks like. There is the price, there is the VAT. If that would move somewhere else, there's a problem. Now, the modern, more modern tools would even find it. But again, humans, they can. They know how price looks like. They have that kind of risk adjustment and say this is probably more the price than this one, and that's what AI is doing. But as we know, all the suppliers, if you're a large company like General Electric, you can say this has to be the invoice structure. But if you are not that big, and that's most of the companies out there, every different kind of documents are flying at you. So you have to have humans looking at them, and that's a boring task. You don't get the people anymore to do this kind of thing, and it's a very error-prone task. So this is clearly in document something and then think about financial services. Think about all the transactions out there that are definitely error-prone. They're rule engines, they are old. You want to make sure that those are working. And the more we digitize in the financial services and fintech, the more errors will happen. No human has the capability, or let's say we don't even have enough humans in the world, to check all this data. We need some kind of automation around, and you can't be the rule engine for that. And so all these trends that are happening cost saving, more efficiency gain, but also digitization, more data, you need certain automation around it that has to be as smart as a human, at least in some way.

Manish Rai Those are great insights. So as you remember, when I first joined the company [super.AI], our thesis was that documents is a space which is very saturated, and we will focus on images and other types of data. But what we started seeing is a lot of companies were coming to us even for document use cases because they said the first generation IDP companies overpromised and under-delivered to them, and some of them had proprietary models which became hard to manage. They work for some [document] types, not others. And they were black boxes, they couldn't do anything. And they loved that we have a flexible model where we can use the best model for any problem out there, and we can break it down. So that was one.  Second area where people said is, hey, they just tell me of my confidence level at the field level. Then I'm tinkering with the confidence levels to get the guaranteed output. And what they love is we allow them to pick. What are the trade-offs you want to make? What quality do you want? Do you want 99.9%, or are you okay with 98%? And your driver is cost or is it speed? And when they saw that,  people loved that concept. And the third area, they said, is for use cases like invoices and POs [purchase orders], because where it's semi-structured, and structure is changing so much, you're still only automating 60 to 70% of those. And I'm using humans for 30% of that. And the whole post-processing is not efficient because I have to source the humans myself, manage that workforce, and then it's not learning efficiently from those humans from exception management, and they loved that. Hey, do you have the option of crowdsourced resources that we can deploy for post-processing? Yeah.

Jasper Masemann: I mean, think about what's the reality. And the reality in the past was that's a consultant coming with a promise and say, I will automate this, and then they show you something, and it's not really automated. So there is post work again, another promise, another promise. And the nice thing about what you describe in the product and AI itself can show you over time how it's getting better. You can choose according to your budget, how good it should be, and how much you want to invest in certain quality levels. And I'm in control without having to do too much work. I think that's the nice one about its more transparency, more control, and obviously more cost saving.

Manish Rai Yeah. Yeah. No, those. Those are spot on. So moving on from this to the topic of use cases, as you're talking in the industry, what are some of the emerging use cases for AI? I can talk about our experience at super.AI and what we are seeing. Like I said, we didn't think documents would be something people will come to us for. And then we saw a wave of layoffs in high flier IDP companies lately, who had raised a lot of money and then didn't deliver. So we are seeing those sets of use cases. Another is transaction data enhancement. It could be banking firms and stuff where we have all had this problem. When we look at our credit card statement, and we see a random string with a $50 or €50 charge, and we are trying to figure out where did we spend that money? I can't remember. And that's a problem we are helping. Even a company like Walmart can have thousands and thousands of variations in how that shows up on a credit card bill. Those kinds of enhancements where it's a post-processing problem, you have to enhance the data and improve it. And the third is around redaction. So I think GDPR and strengthening of UK privacy laws, and all over the world, people are very conscious about their privacy, and people want to not only redact images but also videos, blur faces, remove brands and license plate numbers, etc. We are seeing an improvement in those. And finally, one other use case we have seen a lot around is visual inspection. So if you look at the supply chain part of it and in manufacturing, there are companies that specialize in visual inspection. And when you look at their processes, it's highly manual and very paper driven people taking pictures and coming back and somebody manually looking at that picture. And now we can see AI can analyze the pictures quite efficiently, identify cracks, rust, etc. So those are some of the use cases we have seen. What are you hearing about from some of the implementation partners you speak to?

Jasper Masemann Yeah, I think I would focus on two parts. One is the visual as you describe, and one is more around documents. And on the visual side, you could say it's very well advanced. So exactly what you describing wouldn't have been possible a couple of years ago. And now we can have more details. But I think what's more important is we can see more edge cases.  Because AI is a kind of statistics. So it's looking for patterns, repetitive things. So it was easier with faces and things with a lot of data. But inspection, that's a little bit more tricky. But the models have become better, more robust. And we know how to highlight variations there. So this is actually really upcoming, quality assurance on the industrial level where you still have humans doing that, where you still have a lot of costs and the quality control at the end. I’d imagine an iPhone, all the little details there. That's mostly humans doing that. So this is clearly something very, very important. And I'm not talking about self-driving cars. That's a different issue. It's really more on the visual quality control part. And then when you look at documents, I think there's still a lot around covering all the edge cases, what data to extract from the document. So there is still a lot of potential because you need to train every individual model here. And then for that, you need a very good training engine that is cost efficient and also where you can, as you describe that balance, the quality versus the investment. But then the second thoughts about NLU - natural language understanding - and then processing, which is now happening. So we actually understand what's in the contract. We can compare contracts, we can help lawyers, we can help even people saying, hey, we have so many contracts out there. What's the content of them? Let's compare them, not being a lawyer.  Or. Just email understanding for marketing, all these kind of things. This is really happening at scale. It's of good quality. Where you cannot do automation but augment, and I'm not saying replacing, but augment human beings because the models are robust and working. And the interesting thing is the first wave is usually products which have a very broad use case, self-driving cars, annotation, right, data annotation. But what every company out there really needs is an AI that works for them for their specific use case. And for that, you need a product that helps you. So you don't have to go into the Internet to figure out - what's the right AI model, what's the right data scientists, what's the right process to annotate it? I want this from the shelf. Kind of buying my own, I don't know, chainsaw. And then I can cut down the trees in my garden myself. But you want to make sure that this chainsaw is working. So this is what's now really the next question to apply AI in your own company so you can have the benefits and you don't have to buy a standard product because it doesn't work with the standard product because it's your data, your process, your customers, your company, and it needs certain customization.

Manish Rai Yeah, I think it brings me to another point. We saw a lot of statistics in the last few years that 80% of the IA projects fail or some large portion of them, 70 to 80% do. I'm not exactly sure how much, but Gartner put out statistics, and most of them are more like data science projects than actual products that give you business value. How do you see the AI and business landscape shifting from those early days when we had a high failure rate?

Jasper Masemann: Yeah, there's much more trust. The thing about what's the difference between AI and, let's say, deterministic computer science programming? I programmed something. I built a rule engine in whatever language, say Python, and I know that as a result - yes, I have some quality assurance - whatever I'm building, it's probably working over time.  With AI it's I get the data, and I'm trying to figure out if the pattern my hypothesis says is there will help me to make certain decisions. And the more data I have, the more secure I can be around it, and obviously, the more I know my business. This is not science, don't worry. I mean, everybody who's applying AI, as also I think is super.AI. They already know what the result is. They just have to do it at scale. That's not a research project. So the main thing about those AI projects in the past was people were trying to figure out if AI actually works with that data, what kind of model it could be, and what kind of team they would need. And now we know the models, we know the process for data labeling. We know how to deploy those models in production. We know what people to hire. So all done. And it's more around execution than experimentation.

Manish Rai: Yes. And those are great points. I think one thing our customers are also realizing is that our platform allows them to take any business problem where they can give unstructured data and create an AI application, which is enterprise-grade very quickly, and that gives them structured output. And so we become a step in the automation. If you're using RPA or low-code automation platform, you can simply use an API call to upload the data to us and then use another API call to download the processed data. And then you can go on and use RPA in conjunction, low-code, BPM, and all of those tools to fully automate the business process.

Jasper Masemann: And that feels more common than how you would work with a computer scientist, right? With an engineer. Because before you would speak to a data scientist. I would do some research. I would figure out the models. I don't know yet. And that's not what you're used to. Right. That's why many of those projects only had an innovation budget, very small, low risk took a lot of time. But if I know I can buy this from the shelf, and it's like buying, I don't know, snowflake infrastructure. Yeah. Then I'm fine that I can do it. I can scale it. I think that's also another important point. Nobody knew if they could scale those things because of all the processes, quality assurance, and deployment attached to it. It's very complex if you build it yourself, but you don't have to build it yourself.

Manish Rai: Right. So now**,** looking at one of the questions in everyone's mind**,** is that about three or four months back, right? Suddenly things came to a head with the war in Europe, the inflation hitting an all**-time high, the supply chain, the stock markets,** and all the asset values coming down quickly. And people were frozen. And I think a lot of my peers when I spoke to them, they were hearing the VCs were putting out theses from Y Combinator, yourself, and other companies, like, the party's over. Right? Like I said, now you need to have a hangover recovery plan, and you should have remedies. You should have, a lot of common things thrown around, a 24-36 month runway. And that was a very common advice a lot of VCs were giving to the companies. Fortunately, we were in a good spot because we raised funds last year with your help, and we already had that runway, so we didn't have to make much adjustments in response. But what do you see happening now that we have digested all this news for three months? We are seeing inflation stabilizing, beginning to come down, recession fears rising, and oil coming down. How do you see it all playing out over the next couple of years?

Jasper Masemann: I think that the two things to bear in mind, one is just saying, have 24-36 month runway. I'm not saying you said that, but for me, it just doesn't make sense. It's like telling a runner to carry as much water as possible. I mean, the question you should ask yourself is, where do you want to run? Right? How far is it? How much water do you need? So it's more around, Hey, do I want to reach profitability? How long would it take? Do I need some buffer, and do I want to reach a certain stage of my product?  And then I can scale it fast. I can raise the next round. How long will that take? And the truth is, it's not as easy as last year to raise money. So probably, to raise the same amount of money, you would have to reach more milestones or at a higher quality level. So probably it takes longer, and then maybe you need more runway. That would make sense, right? But it's more the question of what do I have to reach and how long does it take. And then the second one is you might want to take a look at your burn versus how you acquire customers. What kind of revenue you're giving if you get, I don't know how you would say it in English, but if you spend $5 and get $1 back from your customer, then you're doing something wrong, obviously. So those kinds of discussions are more important nowadays. And then the second one about the outlook is, yes, I wouldn't take too much risk on the go-to-market overspending right now because I don't know about maybe what my customer is doing in the next couple of months if they are all on hold if they are looking at the situation. I probably also do the same right now and have a good conversation with my customers.

Manish Rai: Those are great pieces of advice. So how are you changing your investing philosophy and thesis in this market? How has that evolved over the last year?

Jasper Masemann: Well, we got more active. We just had two, you know, three deals in the last days. Well, it's early stage. So it's different valuation levels. But also our growth team is very active because prices have come down. It's more interesting. It's more relevant. I mean, we have to return the money four or five, sixfold that our investors gave to us over time, over ten years. So that's good for us. It's a correction. It's tougher for my personal portfolio companies to raise money. But so far, I'm pretty happy that all of them, they never took, like, overvalued term sheets from, let's say, not-so-experienced investors. But I also know from some founders that last year they took kind of a 200 X revenue, multiple valuations, and they know they will never grow into this valuation for the next funding round, and they will have to probably accept a down round. Which is fine. Because if you think about it now, it's more realistic. It feels bad, but it's not bad because it's just a realistic value of your company. And you still got money last year, and you can use it for growth. So honestly, it's a good time if you realize that the last years were just a bit crazy.

Manish Rai: You're sounding more like Warren Buffett, who's delighted right now and making investments at that point.

Jasper Masemann: I bought a lot of American stocks in the last couple of weeks. Yes.

Manish Rai: Great. So well, time flew quickly. We're at the end of our allocated 30 minutes. So thank you. Yes. For your insights and joining us. And hopefully, we can do this again at some future time.

Read less
Share on TwitterShare on Twitter
Share on FacebookShare on Facebook
Share on GithubShare on Github
Share on LinkedinShare on Linkedin
Share on LinkedinShare on Medium
Button Text