What is cognitive technology? Fred Laluyaux, CEO of Aera Technology, tells CXOTalk about AI, machine learning, and the self-driving enterprise that brings data from multiple systems to make decision-making faster and more reliable.
Beyond ERP: Cognitive Technology and the Self-Driving Enterprise
What is cognitive technology? Fred Laluyaux, CEO of Aera Technology, tells CXOTalk about AI, machine learning, and the self-driving enterprise that brings data from multiple systems to make decision-making faster and more reliable.
“What we’re building is this new generation of applications that are following the same characteristic, real-time and always on, connected outside and in, thinking in autonomous. That’s the high-level principle of Aera – very smart software that has some level of autonomy,” Laluyaux says.
“The nature of a business is fairly complex. We are realizing, basically, the promise of the ERP from 30 years ago by actually truly bringing all this transactional data into a single level. You could say that, in a way, we are fulfilling, finally, the promise of the ERP by removing the complexity of these different systems, bringing all this data into the cloud, into a normalized environment, and having a single, logical process, which we call the decision process mapping. We can map how decisions are being made around demand optimization, planning, and so on and so forth, and automate that in real time. In a way, we’re actually fulfilling that initial promise.”
Laluyaux is an entrepreneur who founded his first company at age 23. Before launching Aera, he was the CEO of Anaplan, which he grew from 20 to 650 employees, and a $1B+ valuation. Before that, he held several executive positions at SAP, Business Objects, and ALG software.
Transcript
Michael Krigsman: I'm Michael Krigsman, an industry analyst and the host of CxOTalk. I'm speaking with Fred Laluyaux, who is the CEO of Aera Technology. Hey, Fred, how are you? Thanks for taking a few minutes with us.
Fred Laluyaux: I'm great. Thanks for having me today.
Michael Krigsman: Fred, tell us about Aera Technology and, also, I'm really interested in why you joined the company as CEO.
Fred Laluyaux: Aera is the cognitive technology that enables what we call the self-driving enterprise. What we're building is this new generation of applications that are following the same characteristic, real-time and always on, connected outside and in, thinking in autonomous. That's the high-level principle of Aera, so very smart software that has some level of autonomy.
Michael Krigsman: You're a veteran of enterprise software. You're a serial entrepreneur. You've worked for some of the largest organizations in the enterprise software world. How is this different from a transactional ERP system, for example?
Fred Laluyaux: You have to realize the problem we're tackling. I mean, why are we doing what we're doing and what are building that's so hard that something is about to change?
Think about the world of enterprise software, 40 years of investment in transaction automation and ERPs on top of which we have invested tons of money in business intelligence, planning, and optimization software. Look forward 40 years into this journey, and the reality of our customers is what? They spend two weeks to try to get an answer to a question that could be as simple as, when is this order going to be ready to be delivered to that customer? Why is that? Why is it so complicated?
Most of the companies we're talking to have dozens. Some go through more than 100 different ERP systems. I'm talking about very large companies. Those systems don't talk to each other. There is no cognitive; there is no intelligence that connects all this transactional data, so you end up with very big silos. You end up with very heavy hierarchies. When you want to make a decision, it's very difficult to measure the impact of that decision on financial, operational, and risk metrics. That's the reality of information workers today.
I joke, but I call our generation the laptop generation. Why? I was born and raised in my career with a laptop, and I use this device 15 hours a day to try to provide basic answers to some fairly simple questions. That's the problem that we need to tackle, and we find a way to do it by leveraging what we call Internet scaled technology. That brings together all of this transactional data coming from all those different data sources, blend it with external sources as well, and bring all of that into what we call this cognitive data layer, which is an intelligent informational layer that connects all the different dimensions in my business and gives me real-time visibility into my business operations. On top of which I can now build a series of skills that are leveraging AI and machine learning or rule-based modeling with very cool UI and NLP to have these real-time interactions with the data.
Michael Krigsman: You're aggregating data across disparate systems.
Fred Laluyaux: Yes. Yes.
Michael Krigsman: Then you are using cognitive technology to, can we say, operate on that data?
Fred Laluyaux: Yeah.
Michael Krigsman: What are you doing to that data?
Fred Laluyaux: Yeah, so it's exactly right. The first part of the problem, again going back to our self-driving analogy, right? The first thing is, how do I get my data to fuel the brain? That's where everything breaks in the enterprise world because I'm using, again, dozens of different systems, ERPs and other systems of records.
The first thing I need to do is translate all this data into a common language. Once I have this single instance that pools in real time through thousands of scrolls every day [and] pull all this data into that single layer, I can start doing any kind of business planning, optimization, visualization that I need to do on an ongoing basis, so very pragmatic applications. Right now, we're deploying our technology available to promise a use case. A large pharma company wants to be able to tell their customers where their orders will be available to deliver.
We're working with other companies on on-shop availability issues. We're working with troubleshooting issues. We're working with demand forecasting. We're looking for supply/demand matching optimization, all those pragmatic use cases in the world of supply chain.
The reality of our customers, they come to us and say, "Look. We have $10 billion stuck in inventory right now. We need to reduce that number to $8 billion." The traditional stack of ERP plus applications plus people doesn't allow us to get through that wall. We need to rethink the architecture. We need to rethink our app to have them running intelligently in real time if we want to be able to get to the next level of performance optimization.
Michael Krigsman: Why can't they use their existing system to get this data? Isn't that what the existing system is designed for?
Fred Laluyaux: Yeah. You look at a supply chain. You look at the manufacturing process grows across GOs. You will end up with open orders with no matching inventory. Great. Very simple problem. I need to find inventory to convert that open order into revenue.
But, how do I do this when I have 54 different ERP systems? That's not simple. And, I need to go and say, "Okay. Predict my demand," so I use our algorithms to do demand predictions. When I've got that number, I have to look at, "Do I have excess inventory somewhere?" Well, that's not a simple answer to get when you have all these different systems.
I assume that I don't have excess inventory. Now I'm increasing production capacity. Well, I can make the same product in multiple areas. What are the cost implications? What are the SLA implications?
The nature of a business is fairly complex. We are realizing, basically, the promise of the ERP from 30 years ago by actually truly bringing all this transactional data into a single level. You could say that, in a way, we are fulfilling, finally, the promise of the ERP by removing the complexity of these different systems, bringing all this data into the cloud, into a normalized environment, and having a single, logical process, which we call the decision process mapping. We can map how decisions are being made around demand optimization, planning, and so on and so forth, and automate that in real time. In a way, we're actually fulfilling that initial promise.
Michael Krigsman: Fred, okay, so you are pulling together data from these multiple systems, but there's that AI, machine learning, cognitive layer. Where does that come into play?
Fred Laluyaux: We're bringing a massive amount of data, billions of rows of data, into that environment. We tie the data together into what we call this--as I said earlier--cognitive layer where the data is highly connected and tied, so I can look at a customer, know which product they bought, know which components go in the product, know where I sold the component from, and so on and so forth. That entire value chain is now connected. There is AI in that process.
Also, as I mentioned earlier, we're building lots and lots and lots of data that needs to be then analyzed and optimized. We run predictions, so demand forecasting. We're going to do a lot of applying the standard algorithms to that.
Where it gets really interesting, in my opinion, is when I combine that cognitive data layer with what I called usage data. Remember that the whole idea is to optimize how decisions are being made. If I come to you with Aera, and I say, "Michael, I've identified open orders in Brazil that have no matching inventory, but there is good news. There is excess inventory in Mexico, and Aera has calculated that if you move that inventory to Brazil, you'll be able to convert another $20 million of revenue. Would you like to do this?"
That's a decision I'm putting in front of you as the system is learning how you think and what decisions you make. That decision is made in the context of financial, operational, risk, and environment that I also, as Aera, understand. I understand the context of your decision. I am now capturing the decision that you are making and, over time, I'll be able to measure the impact of that decision, how it was intended, but also unintended consequences. I'm merging now two data sets: one that's really purely cognitive, transaction data that gives the context of your business and allows me to identify opportunities for optimization through our recommendations, but also, I learn, just as self-driving cars running around the streets to learn, I'm learning from the decisions that you're making. That is pure AI NML.
We're also using those technologies to engage with our users. The goal for us is to become the platform where decisions are actually being made. For that, I need to be real time. I need to be highly usable. I need to have an Alexa or a Siri type usability, but in the context of very complex business decisions. It's pretty much everywhere, and we could not do what we do without these technologies.
Michael Krigsman: As the user engages with the system, over time, the system starts to see the patterns, the type of decisions that the user is trying to make, things like that.
Fred Laluyaux: That's exactly right, and this is not something that -- so we start just like the self-driving car. We start teaching Aera using some process mapping, very basic stuff, if/then statements in a fairly complex tree, but our ability to look at multiple systems and so on and so forth. You start with that, but then you learn, as people make decisions, what decisions are just the right ones.
It doesn't mean that you always make the right decisions. We're able to measure the impact of that decision over time and get back to you and say, "Hey, Michael. Be careful. Every time you're trying to optimize for this, you actually dig a hole here." That's, again, this whole algorithm-based.
Michael Krigsman: Thus, the self-driving enterprise because you've got the data from across the enterprise in these various systems that you're bringing together.
Fred Laluyaux: For the system to learn or any intelligence to learn, you need to have a memory. If you think about what we fail as enterprises for the last 40 years is to build that memory. Every time, every year, we start a new cycle. This planner moves from that job to that job. The sales manager goes from that function to that function or that region to that region, and you lose that memory. So, you're restarting constantly.
What I think we need to create in order to really get to the next level of performance optimization in the broader sense of the word is building that permanent memory over time. Three years later, when a planner goes from one product to another, you don't leave with your memory. You leave your memory behind. That's fundamentally important if we want to have learning systems that become a lot more intelligent.
Michael Krigsman: Right. Okay. Fred Laluyaux, CEO of Aera Technology. Thank you so much for spending a few minutes with us.
Fred Laluyaux: Thanks for having me today. Thank you.
Published Date: Feb 26, 2018
Author: Michael Krigsman
Episode ID: 507