Cognitive automation and artificial intelligence are core parts of the innovation strategy for many organizations. In this episode, we explore the impact of cognitive automation in the enterprise and share key lessons that you must understand in applications such as supply chain management and pricing and promotions for product distribution.

Our guest is Frederic Laluyaux, the President and Chief Executive Officer at Aera Technology, Inc. Mr. Laluyaux served as the Chief Executive Officer and President at Anaplan, Inc. Fred Laluyaux served as Senior Vice President of Operations of Armstrong Laing Group. He held several executive positions at SAP Business Objects. He served as Vice President of Sales and Services at ALG Software of Armstrong Laing Group since July 2006.

Transcript

This transcript has been lightly edited.

Introduction

Michael Krigsman: As we talk about artificial intelligence, machine learning, and terms like cognitive computing, do we know what they really mean and do we understand the impact on the enterprise? That's our topic today on CXOTalk. Fred Laluyaux is the CEO and the founder of Aera Technology.

Frederic Laluyaux: Aera Technology is a cognitive platform that enables what we call the self-driving enterprise. It's a platform for what we call cognitive automation, technology that understands how your business works, answers a lot of the questions that you have. It makes real-time recommendations on how to improve the operations, it predicts business outcomes, and it can take action autonomously. We built this platform over the last few years and we're rolling it out right now.

Michael Krigsman: I think, Fred, we need to try to unpack these terms, terms like cognitive computing, terms like cognitive automation. How do we begin understanding what this is?

Frederic Laluyaux: It sits really on a vision, which is, we're shifting from the era of people doing the work, the work being planning, optimizing, and running operations, business operations, in finance, in supply chain, in sales, in all the different functions of a company. They're running this operation. They're doing their job supported by data, by tools, by collaboration platforms.

We're getting to the point where now we're shifting from people doing the work supported by software to software, computers doing the work controlled by people. That's what we mean by cognitive automation. It's the automation and the augmentation of how decisions are being made and executed in an enterprise. We're using intelligence to actually automate and augment the decision-making process.

How is Cognitive Computing Different from Analytics and Reporting?

Michael Krigsman: Fred, is it a matter of, first, you aggregate data? Let me play devil's advocate or be a little bit facetious. How is this different from analytics and reporting?

Frederic Laluyaux: There are multiple levels of differences. The first thing is, those systems are real-time and always on. You think about analytics and reporting. You'll pull a report and you, as an analyst or as a manager, you'll analyze the data and start thinking about what decisions, what actions you need to take based on what you're analyzing. That's the whole point about getting access to those reports.

Here, the processing of collecting the data, aggregating the data, making sense of the data and executing a set of logical steps, projections, predictions, optimizations are actually done dynamically by the computer. Really, think about it as a giant brain that sits on top of your transactional systems and does the work that analysts would actually do, that managers would actually do. It goes all the way to making the decision and taking the action back into the transactional system.

One is a static report that you look at to make the decision. The other one is a dynamic decision-making system that will analyze and take the action.

What is the Self-Driving Enterprise?

Michael Krigsman: When you talk about being self-driving, what is that? What does that mean?

Frederic Laluyaux: Let me explain to you how it works. The system, end-to-end, the platform starts by crawling the transactional systems. We take the Google analogy, right? When you're crawling the Internet to create a hot replica of every single away-page into a single instance of the cloud so that you can then index, rank, and make that data accessible by a search algorithm. That's the way Google works. That's their breakthrough. A very smart idea.

What we did is we applied the same kind of concept to enterprise data, internal and external. We deploy our crawlers and we create a replica of the transactional data into a single instance of the cloud. Now, we do this across multiple types of ERPs, planning, and other types of transactional tools.

Once the data sits in our cloud, we harmonize it. We augment it. We derive from all these billions of rows of transactions, the business metrics that you need to actually understand your business. Think about a giant data layer. We call it a cognitive data layer that sits on top of all your transactional systems and brings you the ability to find, in a single instance of the cloud, all the information that you need to make a decision, to think through a decision-making process.

Now, once the data is in that foundation, that cognitive data layer, then you can apply, as you said, data modeling, artificial intelligence, statistical forecasting, optimization, a series of tools that allow you to actually run, dynamically, a process. That process says, "Hey, I found open orders without matching inventory. What do I do?" I need to go and look for excess inventory somewhere else. Oh, I can't find it. Now I need to look for production capacity somewhere. If I do, do I have the material?

Do you see how you can unfold that decision-making process? The reason why it works in real-time is that one hundred percent of the information that you need to make that decision sits in a normalized instance of the cloud. The crawlers work autonomously. They're constantly updating the cognitive data layer and the intelligence on top runs 24/7.

That's why we talk about self-driving is that the system will, autonomously, based on some criteria, take an action back into the ERPs. Let's say I'm going to upgrade my forecast for this product by 0.2% for this outlet for this period or, if the decision requires human supervision, it will generate a message, what we call a recommendation that will sit in your inbox. That in-box will say, "Hey, Michael. I recommend that you increase your forecast by 3% because I've done all this analysis, I've done all this work, and I'd like to get your supervision."

At which point you can say, "Hey, Aera. Yeah, let's actually take that action. Thank you." Or you say, "No, I don't want to follow Aera's recommendation because you missed something," in which case Aera will ask you, "What did I miss?" As a result, the system learns from your experience and your expertise and gets better over time.

The concept of self-driving is really to have a system that autonomously does the work from collecting, aggregating, augmenting the data all the way to processing the multiple steps that are required to make a decision. Fully self-driving is the ability of the system to take action right back into the transactional systems.

Michael Krigsman: Could we say that this is similar in a sense to Amazon giving you product recommendations only this is happening proactively and it's giving you recommendations about the next decision that you need to take or the next action that needs to be performed in a process?

Frederic Laluyaux: Somehow, you're absolutely right. It actually will give you. Where is cognitive automation in action today? What are the use cases that we can apply and deploy them for? Forecasting, demand forecasting, collecting all the data and helping to predict what is the forecast level for a product, that's one.

You have inventory optimization. You'd do promotion planning. All those different, complex use cases are there. One of them, you talk about Amazon, is around order management; helping a large, complex organization to predict what is the available to promise date for our complex order.

I think you had one of our clients a few weeks ago on the show. We deployed a technology pretty much for that purpose initially. What we do is we're able to give their customers, their clients a very accurate delivery date for their very complex order.

Now, the challenge here is, the data that is required to actually process this compute sits in 47 different ERPs. First, you have to collect the data, harmonize it, index it, augment it. Then you deploy the algorithms. Now we're able to tell the clients, in real-time, "Your order will be delivered that day."

It sounds like an easy thing to do because you take the Amazon example where we use that every day. When you put that at a scale of a very large enterprise with a lot of data complexity and multiple algorithms that have to be deployed, it's a very hard problem to crack.

What is Aera Technology Automating?

Michael Krigsman: It sounds like the key then is, after you're aggregating, collecting all of that data--I don't want to minimize the difficulty and the challenge of doing that--from the businessperson's point of view, we've got the data and now the system is making recommendations. Those recommendations have to be A) accurate and B) things that I would not have thought of myself because, if I can just do it in an easy way, I don't need that system.

Frederic Laluyaux: There are really two angles. There are two things that we're automating. The first thing is the expertise. The expertise is, you know how to think through a given problem. That can be modeled using our modeling environment, using data science.

The other thing that we need to capture is your experience. Algorithms are not going to get everything in the first place. The question that we have to crack is, how do I build that interface between the Aera brain, so to speak, and the users?

The way we've done this is by saying, "Hey, you know what? We're going to try false positives sometimes. We're going to send you a recommendation to deploy a 52-week calendar, a promotion plan, and you'll say, "You know what? For the third week of July, this promotion makes no sense because people are on vacation." Oh, so it makes no sense to promote this product in that region.

Well, maybe the algorithm missed that in the first place, but we're giving you the opportunity to make some correction and bring the information back to the algorithm that will then run again and, over time, get better and smarter. We have to digitize, basically, the expertise that you have to think through a problem but, also, the experience. That results in two things, Michael.

It results in a level of automation. I can actually do, 24/7, a lot of the work that you do not have the time to do but, also, augmentation because, over time, the recommendations that are delivered by the system are more accurate than what humans are able to do to deliver. There is really a concept of automation and a concept of augmentation. It's interesting to see how the system evolves over time.

You get immediately the effect of having clean data, single source, having deployed the most advanced algorithms and data modeling capabilities but you see how the accuracy of the recommendation increases over time because we don't have an issue with people leaving their job and taking their knowledge with us. Here, we build a permanent memory of all the decisions that are made in an enterprise on a given topic by all the different actors that are participating through that process. You would start aggregating those data points, you can make sense of it, and you can improve the quality.

I'll take a quick analogy, if you allow me, on the self-driving car. Self-driving cars have been programmed to drive a certain way on the road but, as you know, I think way more than 20 million miles, maybe around that of driving a fleet on the cars on the road to get those experience points. We're doing the same thing with not a self-driving car but with a self-driving enterprise. We'll, over time, deliver more accurate forecasts, more accurate recommendations on supply and demand balancing, more accurate recommendations for promotion planning, ATP, and so on and so forth. This is a live system that keeps learning from the way the users are telling us.

What is your Data Model for Cognitive Automation?

Michael Krigsman: You've got the data that you've collected and how do you interpret that data to come out with a model, so to speak, of the experience of the people? You've got data points, but you've got to create almost a three-dimensional abstraction of the people and their minds.

Frederic Laluyaux: It's actually a little more simple than that, Michael. We keep going back to, you get the data. Let me just pause one second here because that's really the biggest problem that we had to fix. If you want to digitize any kind of decision-making process, you need to have 100%, not 99%, 100% of the information/data that you need available, harmonized, with an indexed, understandable by the machine's data model.

The first problem is, how do I go from having 30, 40, 50 different ERPs? Sometimes, just one instance of an ERP doesn't change. We have to bring all these billions of transactions into the cloud and process it into that cognitive data layer.

Once the data is understood in that model, then I need to deploy the logical steps. Aera keeps track of where it is in the logical steps. If I go to you and I say, "I recommend that you shift inventory from this place to that place," so that you change the way your shipments are organized from this distribution center to another to see, I know exactly the context in which I'm asking you as Aera, that decision. I understand your business to risk your service levels. I understand the impact of the recommendation that I'm making, and I understand, when you grab the decision, if you decide yes, if you decide no, if you override some of the number, all of that is being captured.

Effectively, I'm creating the second level of data, which is this decision data. I understand what Michael decided to do at this point in time. Then, of course, it's another data set that I can superpose to my financial and operational data set and, therefore, start deriving some very interesting insight and learning how to either retrain you as a user and say, "You know what? You don't make an optimal decision here," or retrain the algorithm saying, "You know what? Michael is always right when he says no to that decision. Maybe I need to change the way the decision is made digitally in the system."

Michael Krigsman: As with other applications of machine learning, it sounds like the gathering of the data is the crucial piece and the algorithms are easier in comparison.

Frederic Laluyaux: I could not agree with you more. This is really the fundamental problem is collecting the data. Back to my point of being able to pull the data from an ERP, data lake, or whatever source it is. A lot of companies have built a data lake, data grids, data oceans. I'm hearing all sorts of words these days, but we can go straight into the ERP, whatever they are.

The first thing that we have to crack, really, is how do I pull that data without materially impacting the performance of the ERP because, of course, your clients would not be very happy if it takes the ERP down for half an hour to pull the data. We had to crack this. There is a lot of work that's happening there.

Then there is the second stage of logic, which is, how do I take this gibberish, I would say, transactional data and transform it into a clean data model? That's been years of work, of brute force, of mapping everything. There is limited intelligence in that process. It's a lot of human work that enables us to build this at scale.

Once it's done, it's done forever. That's the beauty of the model. With one of our clients, we're running 2,800 crawls a day. We're bringing in 1.1 billion rows of data into the system every day and it runs like a breeze. It just delivers those KPIs, those augmented KPIs that now feed the algorithm.

Now, the algorithm part is not trivial. Algorithms are only as good as the way the data has been prepared. The strength of the system is because the data is so clean and, with time series and real-time of date, we are actually able to prepare it very well so that the algorithms deliver a good prediction in that case or good optimization.

The second part is, how do you operationalize the result of a prediction or a forecast? How do I, in real-time, take that number that the algorithm has delivered and made something with it? You don't want it to go into a spreadsheet, then to a PowerPoint in front of users in a committee. We have found a way to say, "Hey, data comes in. The algorithm runs. A decision is made. It goes straight in front of you, Michael, or automatically gets executed." That's the self-driving concept. It goes really fast, so the data preparation is one thing but the execution, the operationalization of the output is really critical as well.

Michael Krigsman: We have a question from Twitter and Zachary Jeans asks, "Are you able to use Aera, the technology of Aera, in the running of your business? If so, do you have any examples of that?"

Frederic Laluyaux: We call that project Aera on Aera. It's going to make one of our guys super happy that the question was asked. We're literally doing it right now, to be candid. We've been working really hard for the last three years to support some of the largest companies in the world.

The technology that we've built really has been designed for massive scale to be deployed. I think you had J&J, Merck, and Reckitt Benckiser on your show. They're all using our technologies.

What Culture Changes are Necessary for Companies to Adopt Cognitive Automation?

Michael Krigsman: We have a question from Twitter from Sal Rasa who says, "What about the cultural shifts that are necessary for an organization to leverage this type of cognitive automation technology?

Frederic Laluyaux: We're at the beginning of the journey. We're only a year in with some customers being truly live with this system, with the system having somehow either taken over an entire process or really being interacting with the users. My perspective is not long enough to draw some firm conclusions. That's the excitement about what we're doing today is we're experimenting. We're working with our clients in that experimentation.

What I can tell you and what surprised me is that the aspect of automation came stronger than I expected. In other words, what we realized, if you think of the company that has thousands of planners that are basically operating their business, the supply chain, manufacturing, finance, coordinating everything, they're completely overwhelmed. The idea of giving them more data, more tools, more computation capabilities, more collaboration, faster everything, it's reached a point of no return and people are stressed and tired of being asked to do more when they see the competition with digital natives who are actually running a lot faster than them.

They really welcome the technology that we provide because it helps them get through a lot of the work that they can't otherwise get through during a given day. When the system is running 24/7 and doing 80% or 90% of the work, and you get to the office in the morning, you can see how this digital assistant, basically, has performed work for you, real work, and taking actions. It's a real relief, so now you can focus on what you're really good at and what you're uniquely positioned to do, designing the network, designing some plans as opposed to writing their execution. There's been a real strong interest in that automation part as well as the augmentation because running some of these complex decisions is very repetitive and very difficult.

The question that you ask, if I can talk about one more thing, is super interesting in one way. We have a client who is running their entire forecast process in what they call touchless forecasting. In other words, the entire processing is done with Aera. There is no human intervention. The numbers that are called, the forecast numbers that are called, are 100% called by era.

Now, from a cultural perspective, who owns the number? Who owns Aera? Who owns the number? The business now is going to execute against a plan that's been designed by a computer.

The initial reaction was like, "Well, that's not my number." What they discovered is that the accuracy of the forecast, after you run it for a few months, you realize that the accuracy is exceptional. Then, suddenly, everybody aligns toward it.

Yeah, all these changes are happening. Globally, right now, I think the system is pretty much welcome by the users because it really helps them.

What is the Aera Technology User Interface?

Michael Krigsman: From the point of view of the end-user, what does the system look like? What's the user interface?

Frederic Laluyaux: There are four things that you have to crack if you really want to build a cognitive automation platform.

  • The first thing is the data, and we've talked about it.
  • The second thing is the science. How do I digitize the decision-making process?
  • The third is process. How do I embed this digital brain inside my organization and work with the users?
  • The fourth pillar that we work so hard on is change. How do I build a user interface and an ability for the system to interact with the users in real-time in such a hopefully pleasing and easy way?

We built a series of tools. I'm going to use the word "cognitive" a lot, so I'm sorry about that in advance. It's called a cognitive workbench. What it does, basically, think about it as your email, your Gmail. We have different skills. Aera has different skill sets that are working for you 24/7.

It delivers those very clean recommendations just like a message in your inbox saying, "Michael, I recommend you change this number. Michael, I recommend you change that promotion." You right-click on it and say, "Tell me why," or you say, "Do it," or you say, "Don't do it."

We have to really build that new tool that allows you to interact with the system. That interaction, actually, is real-time because the recommendation that Aera will make at any point in time might change.

I come to you at 2 o'clock on Tuesday saying, "I recommend you change your shipment structure for this product for this customer," but maybe at 4 o'clock the business context has changed. Aera has captured some external signals and internal signals, and that recommendation either changed or is now made obsolete before you even touched it.

Meaning, we have to be able to get to you 24/7 if you want to, so we built this very cool app where Aera can speak to you. Literally, it's like Alexa or Siri. You get in your car in the morning and you say, "Aera, what are the open actions for me or for my team?" Aera will tell you, "You've got 24 open actions that are open recommendations for an impact of $47,000. This is what they are. Do you want to do it?" You can literally use the voice to interact with the system just like you would do with Siri or Alexa to turn off the lights in your house or put on some music.

The same kind of interaction has built. It's a hit. People really enjoy that.

Michael Krigsman: I suppose it's a hit because, if it's giving accurate results and, from the user perspective, the way to access those results, it's pretty simple.

Frederic Laluyaux: There's no way to lie about this. Aera will make a recommendation to do something and it will calculate and will tell you exactly this is the financial, operational, service level risk. This is the impact of that recommendation and this is the timeframe. You will check over time, was it right or was it wrong?

The system, you build that trust because there is no subjective aspect about it. It's fully objective. Was the recommendation right? Did the recommendation work?

Talk about a change in adoption. The best email I get is, every Friday, this customer says, "This week, we had 147 recommendations for this specific," whatever, "department, unit from Aera for a total value of $547,000 out of which 97 were accepted. The rest were objected. This is why." You can literally measure the impact of that system in real-time, every week, every day. Again, that's a very different way of thinking about how you're running a business but it's absolutely objective.

What is the Role of Trust in Cognitive Computing?

Michael Krigsman: This concept of trust, please elaborate on that. That seems like a crucial point to me.

Frederic Laluyaux: If I come to you and I say I recommend you do something that impacts the way you're performing at work, you'll challenge me and say, "Hey, hold on, Fred. Can you show me where the data is coming from? Can you explain to me what is the logic that you applied? Did you check with so-and-so that they were okay with that? Did you measure the impact of that decision on the rest of the impacted ecosystem?" You're going to challenge me.

Then, after a while, when I come to you, if I'm always giving you the right answers, then you'll say, "Fred, yeah, I got it. I trust you. You can go ahead and run this for me."

The way the system runs initially is, you can define the threshold. At first, there are a lot of recommendations that the users have to get through and say, "Hey, I agree with Aera on this. I don't agree with Aera on that. You might run the system in parallel until you realize that, you know what? The accuracy of the recommendations delivered by Aera is better than what we've been able to achieve manually, which makes perfect sense because the system works with a lot more data and can manage a lot more complexity. It's your way of thinking that's been digitized, so you shouldn't be surprised there, and it works 24/7.

When you see the numbers, you start trusting the numbers. Every time, every new client, we're going through that process of, "Whoa. What is that system telling me?" Then you build that trust. Without the trust, the automation is very limited. You need trust to actually deliver automation.

With automation, you actually build what I call augmentation. You interact more with the system. The system gets smarter over time and it's a virtuous loop, if you want.

We, as a company, can now monitor the performance of every skill that we deploy with our clients in real-time. We actually have this concept of AAR, Aera Accepted Recommendation and Aera Automated Recommendation. We monitor this in real-time.

It's not like a software that you just say, "Hey, Michael. I've installed it and call this number when you have a problem." We're actually in this continuous engagement with our clients saying, "Hey, the accuracy went down this week for this specific region and for this specific forecasting process. Let's look into it together." Of course, clients look into it and we're here to support. You look at this as real-time engagement and the trust builds over time.

Actually, if I say one more thing on this, I was surprised because I was expecting trust to take more time to be picking up. I think I underestimated the amount of pain that is going on right now in large enterprises when they're trying to simply keep up.

We got to a point where we're expecting folks in planners and others to actually become as good as computers and that's not the right thing to do. We've increased the cadence, so to speak, to a point that's not sustainable. We need that digital relief.

What Will Cognitive Automation and AI do to Human Labor and Jobs?

Michael Krigsman: Let's pick up on this topic of augmentation. We have a question or comment from Zachary Jeans again. It's an interesting one. He says, "With the Terminator movie fresh in our minds, do Aera customers or prospective clients have concerns about automation and the whole concept of the robots taking over our jobs?"

Frederic Laluyaux: Of course, and we're now touching on something a little broader, but it's unavoidable. We have a client who has a very colorful way. This is a show in the morning. I'm not going to use that word. Let's say he calls it bad jobs. He says, "Look. There is a lot of bad stuff that people have to do that can be automated. All that is to reposition people's attention, work, and effort into the value-added stuff that computers don't do very well." Asking someone to repeat the same kind of processing over and over, year after year, is not interesting.

There is a level of automation, but the automation is in our life everywhere. What we're seeing right now happening is what I call the center of the pyramid. You've got your factories and the shop floor, and then you have your CEO up there. In the middle, there are a lot of repetitive operations.

If you think of what happened on the shop floor, we went from people doing the work supported by big machines to machines doing the work controlled by people. We're just bringing that concept up. I think it's creating a lot more opportunities for people to work on interesting stuff.

Yes, of course, there is a level of automation. As I said before, the reaction that we see from those operators is like, "Thank God you're helping me here because I don't have to spend six hours a day pushing data from one tool to an Excel spreadsheet to this to that or running after people to get an approval and coming home feeling that I only covered 20% of what I was supposed to cover."

This is a helping tool. This is a tool that takes away a lot of that repetitive work but, also, delivers augmentation. It's not just about doing what you were doing, but it's doing it in a way that's more efficient where computers can actually "beat the human" and we should welcome that.

Michael Krigsman: It's removing a lot of repetitive labor, essentially.

Frederic Laluyaux: Yeah, absolutely and, also, enabling things that would not otherwise be possible. You were asking the question of augmentation. Let me give you a very simple example.

If you think about promotion planning, you create promotion plans so that, when you go to a store, you get the buy one get one free or you see the product in front of the shelf. That has to be planned months in advance. That budget that's allocated to a promotion plan is the second-largest spend for consumer packaged goods companies.

Now, think about the work that really has to be done to adjust to the digital natives, the marketplaces that are actually buying products, doing promotions on the fly, and shipping that to your home in real-time. You cannot ask the account managers who are responsible to build these promotions to constantly monitor every single feed and adjust. There is a lag time, actually, with your supply chain.

Here, you have a system that literally, every day, can read through billions of point of sale data, merge it with Nielsen Elasticity data, look I real-time at the levels of your supply chain, predict what those levers are going to be, and optimize, basically, that supply and demand. It makes no sense to make a promotion if you can't supply and this is very complex, so you want people to actually tell the system, "This is the way I want you to think about it, but then please run this 24/7."

Our brains are not meant to analyze data in real-time across multiple time horizons and we're not computers. We need those computers to do the work for us.

Why is Cognitive Automation Important to Business?

Michael Krigsman: We have another question from Twitter, which I think relates to this, which is, "Can you kind of summarize the relevance of this kind of technology to businesspeople?" To put it another way, a business leader; why should a business leader care about this?

Frederic Laluyaux: Think about the disruption that Amazon has brought to the world of retail, consumer packaged goods, and so on and so forth. Every leader that I meet that can be disrupted by their technology and by their organization is reaching out to us and saying, "The foundations, the fundamental pillars of our organization are being threatened by this world that's moving very, very fast."

These digital disruptors think about everything as a piece of technology, as a software, and we're still structured in this big old pyramid on top of 47 different ERPs. We're trying to ask our people to run faster and faster and make more accurate decisions. We're trying to bring the decision-making process closer to the point of impact and closer to real-time.

They've reached a peak. They know that the relative and the absolute performance of a lot of their functions are degrading rapidly as a result of rotations in the workforce and a lot of different factors. They know that, as I said, relative and absolute performance is degrading and that, if they don't start building that digital layer that allows them to catch, really, and anticipate and react to the digital disruption that's pushed on them by the digital natives, they're going to be in trouble.

There is a high level of relevancy. If I candidly tell you what surprised me when we launched Aera was how relevant the topic was with C-level executives in the largest companies in the world. I was very proud. It was two and a half years ago we launched a concept of the self-driving enterprise, the cognitive operating system. I thought, "Wow, we're up there."

The execs that we talked to didn't say, "Wow. Congratulations, guys." They said, "Where have you been? We've been waiting for this for a long time. You, as an industry, keep telling us that you're going to make our people better? That's not the point, guys. We don't want that. we want to free our people from doing a lot of the bad work that they shouldn't be doing and focus on engaging with the clients. Engaging with the community and doing all this kind of intelligence stuff."

The reaction is there. The appetite for that kind of technology is clearly there. The relevancy is higher than I've ever experienced in my career.

How Should Business Leaders Adopt Cognitive Automation in their Organizations?

Michael Krigsman: Fred, as we finish up, what advice do you have for business leaders who are listening to this and saying, "Yeah, want to do this"? How should companies prepare for adopting these kinds of technologies and the changes that it may bring?

Frederic Laluyaux: My advice is always the same; "Jump in." I'll loop back to something we talked about, Michael, a little bit earlier, which is starting to create that digital memory of how decisions are made and executed in your enterprise is the key to having the algorithms get smarter over time. You have to start building that data set.

The early adopters that have been doing this for 12, 18 months already see an impact on the quality, on the accuracy of the algorithm. There's really a virtual circle when you get going.

My advice is, take one process. Take one function. Logistics, supply chain, take whatever you want, but get started. Start learning. Start operating with that support in mind. Think about Aera or cognitive automation as a supporting platform. If you wait, well, that data collection process will actually be delayed and it will take much longer.

Now, if you have that system running 24/7, you get more accurate. You run a lot faster. You increase your agility. You become a lot more competitive, I mean, significantly more competitive. You can adjust your pricing level, your supply level, and so on and so forth. Your competitors who are still analog are not able to do that at the same speed.

My advice is, speak one topic, start deploying the technology and learn from it. As far as preparation, there's not much because we're built in a way that allows us to plug into any landscape. You can have 40, 50 different ERPs that are not talking to each other. We take care of that.

For us, the vision has always been to enable non-digital native companies to actually operate as fast and as efficiently as digital native companies that were born in the last 20 years. For that, without asking them to rethink what I call their bedrock of ERPs, their fundamental transactional landscape because, if we ask them to touch that, they're not going to make it. It's just too big of a transformation. We had to actually enable our technology to plug on their diverse landscape as opposed to asking them to come to our technology.

Long story short, I would say that my advice is, start now. I can see an increasing interest. We're doing pilots in many of the largest companies in the world and it's really critical to get that process started early so that you can build that intelligence relative to how decisions are being made and executed in the company early.

Michael Krigsman: Are there things that a company has to do regarding the data collection? Do they have to change their operations in any way to start gathering the data?

Frederic Laluyaux: No. No, no. That's exactly a critical point, which is, our technology will plug on top of their ERP, whatever they are, and understand the mapping and do all this stuff. You don't have to. There is a bit of work and adjustments. It's never as easy as it sounds but, in a matter of weeks, you actually have that cognitive data layer built on top of your ERPs or, as I said before, your data ocean, links, swamp, whatever you want to call it. No, there is not a lot of preparation.

The preparation is in the validation of the metrics that are calculated by Aera and it really is on the cognitive automation and the cognitive augmentation. When you deploy a skill, you want to make sure that it's adjusted to the way you operate. There is work that needs to be done there.

Michael Krigsman: Ultimately, then, it's giving feedback into the system about the results that have been achieved so that the system can correct itself for the future.

Frederic Laluyaux: Correct, yeah, and it's the system that corrects itself, but it's also sometimes the user that can correct themselves when you actually see this is how you make these kinds of decisions in that type of context for this business value over time. You may decide to change the way you actually think as well. It's a system adjusting and sometimes it's the user readjusting to this new light that we shed on how decisions are made in a company.

Michael Krigsman: Okay, well, unfortunately, we are out of time. It's been a very fast 45 minutes. Fred, thank you very much for taking your time to be with us today.

Frederic Laluyaux: It's been a real pleasure. Thanks for having me, Michael. Thank you.

Michael Krigsman: We've been speaking with Fred Laluyaux, who is the founder and CEO of Aera Technology. Before you go, please subscribe on YouTube and hit the subscribe button at the top of our website and we'll send you great information. Thank you so much, everybody, and I hope you have a great day. Come back next time and we'll see you again soon. Bye-bye.​

This transcript has been lightly edited.

Introduction

Michael Krigsman: As we talk about artificial intelligence, machine learning, and terms like cognitive computing, do we know what they really mean and do we understand the impact on the enterprise? That's our topic today on CXOTalk. Fred Laluyaux is the CEO and the founder of Aera Technology.

Frederic Laluyaux: Aera Technology is a cognitive platform that enables what we call the self-driving enterprise. It's a platform for what we call cognitive automation, technology that understands how your business works, answers a lot of the questions that you have. It makes real-time recommendations on how to improve the operations, it predicts business outcomes, and it can take action autonomously. We built this platform over the last few years and we're rolling it out right now.

Michael Krigsman: I think, Fred, we need to try to unpack these terms, terms like cognitive computing, terms like cognitive automation. How do we begin understanding what this is?

Frederic Laluyaux: It sits really on a vision, which is, we're shifting from the era of people doing the work, the work being planning, optimizing, and running operations, business operations, in finance, in supply chain, in sales, in all the different functions of a company. They're running this operation. They're doing their job supported by data, by tools, by collaboration platforms.

We're getting to the point where now we're shifting from people doing the work supported by software to software, computers doing the work controlled by people. That's what we mean by cognitive automation. It's the automation and the augmentation of how decisions are being made and executed in an enterprise. We're using intelligence to actually automate and augment the decision-making process.

How is Cognitive Computing Different from Analytics and Reporting?

Michael Krigsman: Fred, is it a matter of, first, you aggregate data? Let me play devil's advocate or be a little bit facetious. How is this different from analytics and reporting?

Frederic Laluyaux: There are multiple levels of differences. The first thing is, those systems are real-time and always on. You think about analytics and reporting. You'll pull a report and you, as an analyst or as a manager, you'll analyze the data and start thinking about what decisions, what actions you need to take based on what you're analyzing. That's the whole point about getting access to those reports.

Here, the processing of collecting the data, aggregating the data, making sense of the data and executing a set of logical steps, projections, predictions, optimizations are actually done dynamically by the computer. Really, think about it as a giant brain that sits on top of your transactional systems and does the work that analysts would actually do, that managers would actually do. It goes all the way to making the decision and taking the action back into the transactional system.

One is a static report that you look at to make the decision. The other one is a dynamic decision-making system that will analyze and take the action.

What is the Self-Driving Enterprise?

Michael Krigsman: When you talk about being self-driving, what is that? What does that mean?

Frederic Laluyaux: Let me explain to you how it works. The system, end-to-end, the platform starts by crawling the transactional systems. We take the Google analogy, right? When you're crawling the Internet to create a hot replica of every single away-page into a single instance of the cloud so that you can then index, rank, and make that data accessible by a search algorithm. That's the way Google works. That's their breakthrough. A very smart idea.

What we did is we applied the same kind of concept to enterprise data, internal and external. We deploy our crawlers and we create a replica of the transactional data into a single instance of the cloud. Now, we do this across multiple types of ERPs, planning, and other types of transactional tools.

Once the data sits in our cloud, we harmonize it. We augment it. We derive from all these billions of rows of transactions, the business metrics that you need to actually understand your business. Think about a giant data layer. We call it a cognitive data layer that sits on top of all your transactional systems and brings you the ability to find, in a single instance of the cloud, all the information that you need to make a decision, to think through a decision-making process.

Now, once the data is in that foundation, that cognitive data layer, then you can apply, as you said, data modeling, artificial intelligence, statistical forecasting, optimization, a series of tools that allow you to actually run, dynamically, a process. That process says, "Hey, I found open orders without matching inventory. What do I do?" I need to go and look for excess inventory somewhere else. Oh, I can't find it. Now I need to look for production capacity somewhere. If I do, do I have the material?

Do you see how you can unfold that decision-making process? The reason why it works in real-time is that one hundred percent of the information that you need to make that decision sits in a normalized instance of the cloud. The crawlers work autonomously. They're constantly updating the cognitive data layer and the intelligence on top runs 24/7.

That's why we talk about self-driving is that the system will, autonomously, based on some criteria, take an action back into the ERPs. Let's say I'm going to upgrade my forecast for this product by 0.2% for this outlet for this period or, if the decision requires human supervision, it will generate a message, what we call a recommendation that will sit in your inbox. That in-box will say, "Hey, Michael. I recommend that you increase your forecast by 3% because I've done all this analysis, I've done all this work, and I'd like to get your supervision."

At which point you can say, "Hey, Aera. Yeah, let's actually take that action. Thank you." Or you say, "No, I don't want to follow Aera's recommendation because you missed something," in which case Aera will ask you, "What did I miss?" As a result, the system learns from your experience and your expertise and gets better over time.

The concept of self-driving is really to have a system that autonomously does the work from collecting, aggregating, augmenting the data all the way to processing the multiple steps that are required to make a decision. Fully self-driving is the ability of the system to take action right back into the transactional systems.

Michael Krigsman: Could we say that this is similar in a sense to Amazon giving you product recommendations only this is happening proactively and it's giving you recommendations about the next decision that you need to take or the next action that needs to be performed in a process?

Frederic Laluyaux: Somehow, you're absolutely right. It actually will give you. Where is cognitive automation in action today? What are the use cases that we can apply and deploy them for? Forecasting, demand forecasting, collecting all the data and helping to predict what is the forecast level for a product, that's one.

You have inventory optimization. You'd do promotion planning. All those different, complex use cases are there. One of them, you talk about Amazon, is around order management; helping a large, complex organization to predict what is the available to promise date for our complex order.

I think you had one of our clients a few weeks ago on the show. We deployed a technology pretty much for that purpose initially. What we do is we're able to give their customers, their clients a very accurate delivery date for their very complex order.

Now, the challenge here is, the data that is required to actually process this compute sits in 47 different ERPs. First, you have to collect the data, harmonize it, index it, augment it. Then you deploy the algorithms. Now we're able to tell the clients, in real-time, "Your order will be delivered that day."

It sounds like an easy thing to do because you take the Amazon example where we use that every day. When you put that at a scale of a very large enterprise with a lot of data complexity and multiple algorithms that have to be deployed, it's a very hard problem to crack.

What is Aera Technology Automating?

Michael Krigsman: It sounds like the key then is, after you're aggregating, collecting all of that data--I don't want to minimize the difficulty and the challenge of doing that--from the businessperson's point of view, we've got the data and now the system is making recommendations. Those recommendations have to be A) accurate and B) things that I would not have thought of myself because, if I can just do it in an easy way, I don't need that system.

Frederic Laluyaux: There are really two angles. There are two things that we're automating. The first thing is the expertise. The expertise is, you know how to think through a given problem. That can be modeled using our modeling environment, using data science.

The other thing that we need to capture is your experience. Algorithms are not going to get everything in the first place. The question that we have to crack is, how do I build that interface between the Aera brain, so to speak, and the users?

The way we've done this is by saying, "Hey, you know what? We're going to try false positives sometimes. We're going to send you a recommendation to deploy a 52-week calendar, a promotion plan, and you'll say, "You know what? For the third week of July, this promotion makes no sense because people are on vacation." Oh, so it makes no sense to promote this product in that region.

Well, maybe the algorithm missed that in the first place, but we're giving you the opportunity to make some correction and bring the information back to the algorithm that will then run again and, over time, get better and smarter. We have to digitize, basically, the expertise that you have to think through a problem but, also, the experience. That results in two things, Michael.

It results in a level of automation. I can actually do, 24/7, a lot of the work that you do not have the time to do but, also, augmentation because, over time, the recommendations that are delivered by the system are more accurate than what humans are able to do to deliver. There is really a concept of automation and a concept of augmentation. It's interesting to see how the system evolves over time.

You get immediately the effect of having clean data, single source, having deployed the most advanced algorithms and data modeling capabilities but you see how the accuracy of the recommendation increases over time because we don't have an issue with people leaving their job and taking their knowledge with us. Here, we build a permanent memory of all the decisions that are made in an enterprise on a given topic by all the different actors that are participating through that process. You would start aggregating those data points, you can make sense of it, and you can improve the quality.

I'll take a quick analogy, if you allow me, on the self-driving car. Self-driving cars have been programmed to drive a certain way on the road but, as you know, I think way more than 20 million miles, maybe around that of driving a fleet on the cars on the road to get those experience points. We're doing the same thing with not a self-driving car but with a self-driving enterprise. We'll, over time, deliver more accurate forecasts, more accurate recommendations on supply and demand balancing, more accurate recommendations for promotion planning, ATP, and so on and so forth. This is a live system that keeps learning from the way the users are telling us.

What is your Data Model for Cognitive Automation?

Michael Krigsman: You've got the data that you've collected and how do you interpret that data to come out with a model, so to speak, of the experience of the people? You've got data points, but you've got to create almost a three-dimensional abstraction of the people and their minds.

Frederic Laluyaux: It's actually a little more simple than that, Michael. We keep going back to, you get the data. Let me just pause one second here because that's really the biggest problem that we had to fix. If you want to digitize any kind of decision-making process, you need to have 100%, not 99%, 100% of the information/data that you need available, harmonized, with an indexed, understandable by the machine's data model.

The first problem is, how do I go from having 30, 40, 50 different ERPs? Sometimes, just one instance of an ERP doesn't change. We have to bring all these billions of transactions into the cloud and process it into that cognitive data layer.

Once the data is understood in that model, then I need to deploy the logical steps. Aera keeps track of where it is in the logical steps. If I go to you and I say, "I recommend that you shift inventory from this place to that place," so that you change the way your shipments are organized from this distribution center to another to see, I know exactly the context in which I'm asking you as Aera, that decision. I understand your business to risk your service levels. I understand the impact of the recommendation that I'm making, and I understand, when you grab the decision, if you decide yes, if you decide no, if you override some of the number, all of that is being captured.

Effectively, I'm creating the second level of data, which is this decision data. I understand what Michael decided to do at this point in time. Then, of course, it's another data set that I can superpose to my financial and operational data set and, therefore, start deriving some very interesting insight and learning how to either retrain you as a user and say, "You know what? You don't make an optimal decision here," or retrain the algorithm saying, "You know what? Michael is always right when he says no to that decision. Maybe I need to change the way the decision is made digitally in the system."

Michael Krigsman: As with other applications of machine learning, it sounds like the gathering of the data is the crucial piece and the algorithms are easier in comparison.

Frederic Laluyaux: I could not agree with you more. This is really the fundamental problem is collecting the data. Back to my point of being able to pull the data from an ERP, data lake, or whatever source it is. A lot of companies have built a data lake, data grids, data oceans. I'm hearing all sorts of words these days, but we can go straight into the ERP, whatever they are.

The first thing that we have to crack, really, is how do I pull that data without materially impacting the performance of the ERP because, of course, your clients would not be very happy if it takes the ERP down for half an hour to pull the data. We had to crack this. There is a lot of work that's happening there.

Then there is the second stage of logic, which is, how do I take this gibberish, I would say, transactional data and transform it into a clean data model? That's been years of work, of brute force, of mapping everything. There is limited intelligence in that process. It's a lot of human work that enables us to build this at scale.

Once it's done, it's done forever. That's the beauty of the model. With one of our clients, we're running 2,800 crawls a day. We're bringing in 1.1 billion rows of data into the system every day and it runs like a breeze. It just delivers those KPIs, those augmented KPIs that now feed the algorithm.

Now, the algorithm part is not trivial. Algorithms are only as good as the way the data has been prepared. The strength of the system is because the data is so clean and, with time series and real-time of date, we are actually able to prepare it very well so that the algorithms deliver a good prediction in that case or good optimization.

The second part is, how do you operationalize the result of a prediction or a forecast? How do I, in real-time, take that number that the algorithm has delivered and made something with it? You don't want it to go into a spreadsheet, then to a PowerPoint in front of users in a committee. We have found a way to say, "Hey, data comes in. The algorithm runs. A decision is made. It goes straight in front of you, Michael, or automatically gets executed." That's the self-driving concept. It goes really fast, so the data preparation is one thing but the execution, the operationalization of the output is really critical as well.

Michael Krigsman: We have a question from Twitter and Zachary Jeans asks, "Are you able to use Aera, the technology of Aera, in the running of your business? If so, do you have any examples of that?"

Frederic Laluyaux: We call that project Aera on Aera. It's going to make one of our guys super happy that the question was asked. We're literally doing it right now, to be candid. We've been working really hard for the last three years to support some of the largest companies in the world.

The technology that we've built really has been designed for massive scale to be deployed. I think you had J&J, Merck, and Reckitt Benckiser on your show. They're all using our technologies.

What Culture Changes are Necessary for Companies to Adopt Cognitive Automation?

Michael Krigsman: We have a question from Twitter from Sal Rasa who says, "What about the cultural shifts that are necessary for an organization to leverage this type of cognitive automation technology?

Frederic Laluyaux: We're at the beginning of the journey. We're only a year in with some customers being truly live with this system, with the system having somehow either taken over an entire process or really being interacting with the users. My perspective is not long enough to draw some firm conclusions. That's the excitement about what we're doing today is we're experimenting. We're working with our clients in that experimentation.

What I can tell you and what surprised me is that the aspect of automation came stronger than I expected. In other words, what we realized, if you think of the company that has thousands of planners that are basically operating their business, the supply chain, manufacturing, finance, coordinating everything, they're completely overwhelmed. The idea of giving them more data, more tools, more computation capabilities, more collaboration, faster everything, it's reached a point of no return and people are stressed and tired of being asked to do more when they see the competition with digital natives who are actually running a lot faster than them.

They really welcome the technology that we provide because it helps them get through a lot of the work that they can't otherwise get through during a given day. When the system is running 24/7 and doing 80% or 90% of the work, and you get to the office in the morning, you can see how this digital assistant, basically, has performed work for you, real work, and taking actions. It's a real relief, so now you can focus on what you're really good at and what you're uniquely positioned to do, designing the network, designing some plans as opposed to writing their execution. There's been a real strong interest in that automation part as well as the augmentation because running some of these complex decisions is very repetitive and very difficult.

The question that you ask, if I can talk about one more thing, is super interesting in one way. We have a client who is running their entire forecast process in what they call touchless forecasting. In other words, the entire processing is done with Aera. There is no human intervention. The numbers that are called, the forecast numbers that are called, are 100% called by era.

Now, from a cultural perspective, who owns the number? Who owns Aera? Who owns the number? The business now is going to execute against a plan that's been designed by a computer.

The initial reaction was like, "Well, that's not my number." What they discovered is that the accuracy of the forecast, after you run it for a few months, you realize that the accuracy is exceptional. Then, suddenly, everybody aligns toward it.

Yeah, all these changes are happening. Globally, right now, I think the system is pretty much welcome by the users because it really helps them.

What is the Aera Technology User Interface?

Michael Krigsman: From the point of view of the end-user, what does the system look like? What's the user interface?

Frederic Laluyaux: There are four things that you have to crack if you really want to build a cognitive automation platform.

  • The first thing is the data, and we've talked about it.
  • The second thing is the science. How do I digitize the decision-making process?
  • The third is process. How do I embed this digital brain inside my organization and work with the users?
  • The fourth pillar that we work so hard on is change. How do I build a user interface and an ability for the system to interact with the users in real-time in such a hopefully pleasing and easy way?

We built a series of tools. I'm going to use the word "cognitive" a lot, so I'm sorry about that in advance. It's called a cognitive workbench. What it does, basically, think about it as your email, your Gmail. We have different skills. Aera has different skill sets that are working for you 24/7.

It delivers those very clean recommendations just like a message in your inbox saying, "Michael, I recommend you change this number. Michael, I recommend you change that promotion." You right-click on it and say, "Tell me why," or you say, "Do it," or you say, "Don't do it."

We have to really build that new tool that allows you to interact with the system. That interaction, actually, is real-time because the recommendation that Aera will make at any point in time might change.

I come to you at 2 o'clock on Tuesday saying, "I recommend you change your shipment structure for this product for this customer," but maybe at 4 o'clock the business context has changed. Aera has captured some external signals and internal signals, and that recommendation either changed or is now made obsolete before you even touched it.

Meaning, we have to be able to get to you 24/7 if you want to, so we built this very cool app where Aera can speak to you. Literally, it's like Alexa or Siri. You get in your car in the morning and you say, "Aera, what are the open actions for me or for my team?" Aera will tell you, "You've got 24 open actions that are open recommendations for an impact of $47,000. This is what they are. Do you want to do it?" You can literally use the voice to interact with the system just like you would do with Siri or Alexa to turn off the lights in your house or put on some music.

The same kind of interaction has built. It's a hit. People really enjoy that.

Michael Krigsman: I suppose it's a hit because, if it's giving accurate results and, from the user perspective, the way to access those results, it's pretty simple.

Frederic Laluyaux: There's no way to lie about this. Aera will make a recommendation to do something and it will calculate and will tell you exactly this is the financial, operational, service level risk. This is the impact of that recommendation and this is the timeframe. You will check over time, was it right or was it wrong?

The system, you build that trust because there is no subjective aspect about it. It's fully objective. Was the recommendation right? Did the recommendation work?

Talk about a change in adoption. The best email I get is, every Friday, this customer says, "This week, we had 147 recommendations for this specific," whatever, "department, unit from Aera for a total value of $547,000 out of which 97 were accepted. The rest were objected. This is why." You can literally measure the impact of that system in real-time, every week, every day. Again, that's a very different way of thinking about how you're running a business but it's absolutely objective.

What is the Role of Trust in Cognitive Computing?

Michael Krigsman: This concept of trust, please elaborate on that. That seems like a crucial point to me.

Frederic Laluyaux: If I come to you and I say I recommend you do something that impacts the way you're performing at work, you'll challenge me and say, "Hey, hold on, Fred. Can you show me where the data is coming from? Can you explain to me what is the logic that you applied? Did you check with so-and-so that they were okay with that? Did you measure the impact of that decision on the rest of the impacted ecosystem?" You're going to challenge me.

Then, after a while, when I come to you, if I'm always giving you the right answers, then you'll say, "Fred, yeah, I got it. I trust you. You can go ahead and run this for me."

The way the system runs initially is, you can define the threshold. At first, there are a lot of recommendations that the users have to get through and say, "Hey, I agree with Aera on this. I don't agree with Aera on that. You might run the system in parallel until you realize that, you know what? The accuracy of the recommendations delivered by Aera is better than what we've been able to achieve manually, which makes perfect sense because the system works with a lot more data and can manage a lot more complexity. It's your way of thinking that's been digitized, so you shouldn't be surprised there, and it works 24/7.

When you see the numbers, you start trusting the numbers. Every time, every new client, we're going through that process of, "Whoa. What is that system telling me?" Then you build that trust. Without the trust, the automation is very limited. You need trust to actually deliver automation.

With automation, you actually build what I call augmentation. You interact more with the system. The system gets smarter over time and it's a virtuous loop, if you want.

We, as a company, can now monitor the performance of every skill that we deploy with our clients in real-time. We actually have this concept of AAR, Aera Accepted Recommendation and Aera Automated Recommendation. We monitor this in real-time.

It's not like a software that you just say, "Hey, Michael. I've installed it and call this number when you have a problem." We're actually in this continuous engagement with our clients saying, "Hey, the accuracy went down this week for this specific region and for this specific forecasting process. Let's look into it together." Of course, clients look into it and we're here to support. You look at this as real-time engagement and the trust builds over time.

Actually, if I say one more thing on this, I was surprised because I was expecting trust to take more time to be picking up. I think I underestimated the amount of pain that is going on right now in large enterprises when they're trying to simply keep up.

We got to a point where we're expecting folks in planners and others to actually become as good as computers and that's not the right thing to do. We've increased the cadence, so to speak, to a point that's not sustainable. We need that digital relief.

What Will Cognitive Automation and AI do to Human Labor and Jobs?

Michael Krigsman: Let's pick up on this topic of augmentation. We have a question or comment from Zachary Jeans again. It's an interesting one. He says, "With the Terminator movie fresh in our minds, do Aera customers or prospective clients have concerns about automation and the whole concept of the robots taking over our jobs?"

Frederic Laluyaux: Of course, and we're now touching on something a little broader, but it's unavoidable. We have a client who has a very colorful way. This is a show in the morning. I'm not going to use that word. Let's say he calls it bad jobs. He says, "Look. There is a lot of bad stuff that people have to do that can be automated. All that is to reposition people's attention, work, and effort into the value-added stuff that computers don't do very well." Asking someone to repeat the same kind of processing over and over, year after year, is not interesting.

There is a level of automation, but the automation is in our life everywhere. What we're seeing right now happening is what I call the center of the pyramid. You've got your factories and the shop floor, and then you have your CEO up there. In the middle, there are a lot of repetitive operations.

If you think of what happened on the shop floor, we went from people doing the work supported by big machines to machines doing the work controlled by people. We're just bringing that concept up. I think it's creating a lot more opportunities for people to work on interesting stuff.

Yes, of course, there is a level of automation. As I said before, the reaction that we see from those operators is like, "Thank God you're helping me here because I don't have to spend six hours a day pushing data from one tool to an Excel spreadsheet to this to that or running after people to get an approval and coming home feeling that I only covered 20% of what I was supposed to cover."

This is a helping tool. This is a tool that takes away a lot of that repetitive work but, also, delivers augmentation. It's not just about doing what you were doing, but it's doing it in a way that's more efficient where computers can actually "beat the human" and we should welcome that.

Michael Krigsman: It's removing a lot of repetitive labor, essentially.

Frederic Laluyaux: Yeah, absolutely and, also, enabling things that would not otherwise be possible. You were asking the question of augmentation. Let me give you a very simple example.

If you think about promotion planning, you create promotion plans so that, when you go to a store, you get the buy one get one free or you see the product in front of the shelf. That has to be planned months in advance. That budget that's allocated to a promotion plan is the second-largest spend for consumer packaged goods companies.

Now, think about the work that really has to be done to adjust to the digital natives, the marketplaces that are actually buying products, doing promotions on the fly, and shipping that to your home in real-time. You cannot ask the account managers who are responsible to build these promotions to constantly monitor every single feed and adjust. There is a lag time, actually, with your supply chain.

Here, you have a system that literally, every day, can read through billions of point of sale data, merge it with Nielsen Elasticity data, look I real-time at the levels of your supply chain, predict what those levers are going to be, and optimize, basically, that supply and demand. It makes no sense to make a promotion if you can't supply and this is very complex, so you want people to actually tell the system, "This is the way I want you to think about it, but then please run this 24/7."

Our brains are not meant to analyze data in real-time across multiple time horizons and we're not computers. We need those computers to do the work for us.

Why is Cognitive Automation Important to Business?

Michael Krigsman: We have another question from Twitter, which I think relates to this, which is, "Can you kind of summarize the relevance of this kind of technology to businesspeople?" To put it another way, a business leader; why should a business leader care about this?

Frederic Laluyaux: Think about the disruption that Amazon has brought to the world of retail, consumer packaged goods, and so on and so forth. Every leader that I meet that can be disrupted by their technology and by their organization is reaching out to us and saying, "The foundations, the fundamental pillars of our organization are being threatened by this world that's moving very, very fast."

These digital disruptors think about everything as a piece of technology, as a software, and we're still structured in this big old pyramid on top of 47 different ERPs. We're trying to ask our people to run faster and faster and make more accurate decisions. We're trying to bring the decision-making process closer to the point of impact and closer to real-time.

They've reached a peak. They know that the relative and the absolute performance of a lot of their functions are degrading rapidly as a result of rotations in the workforce and a lot of different factors. They know that, as I said, relative and absolute performance is degrading and that, if they don't start building that digital layer that allows them to catch, really, and anticipate and react to the digital disruption that's pushed on them by the digital natives, they're going to be in trouble.

There is a high level of relevancy. If I candidly tell you what surprised me when we launched Aera was how relevant the topic was with C-level executives in the largest companies in the world. I was very proud. It was two and a half years ago we launched a concept of the self-driving enterprise, the cognitive operating system. I thought, "Wow, we're up there."

The execs that we talked to didn't say, "Wow. Congratulations, guys." They said, "Where have you been? We've been waiting for this for a long time. You, as an industry, keep telling us that you're going to make our people better? That's not the point, guys. We don't want that. we want to free our people from doing a lot of the bad work that they shouldn't be doing and focus on engaging with the clients. Engaging with the community and doing all this kind of intelligence stuff."

The reaction is there. The appetite for that kind of technology is clearly there. The relevancy is higher than I've ever experienced in my career.

How Should Business Leaders Adopt Cognitive Automation in their Organizations?

Michael Krigsman: Fred, as we finish up, what advice do you have for business leaders who are listening to this and saying, "Yeah, want to do this"? How should companies prepare for adopting these kinds of technologies and the changes that it may bring?

Frederic Laluyaux: My advice is always the same; "Jump in." I'll loop back to something we talked about, Michael, a little bit earlier, which is starting to create that digital memory of how decisions are made and executed in your enterprise is the key to having the algorithms get smarter over time. You have to start building that data set.

The early adopters that have been doing this for 12, 18 months already see an impact on the quality, on the accuracy of the algorithm. There's really a virtual circle when you get going.

My advice is, take one process. Take one function. Logistics, supply chain, take whatever you want, but get started. Start learning. Start operating with that support in mind. Think about Aera or cognitive automation as a supporting platform. If you wait, well, that data collection process will actually be delayed and it will take much longer.

Now, if you have that system running 24/7, you get more accurate. You run a lot faster. You increase your agility. You become a lot more competitive, I mean, significantly more competitive. You can adjust your pricing level, your supply level, and so on and so forth. Your competitors who are still analog are not able to do that at the same speed.

My advice is, speak one topic, start deploying the technology and learn from it. As far as preparation, there's not much because we're built in a way that allows us to plug into any landscape. You can have 40, 50 different ERPs that are not talking to each other. We take care of that.

For us, the vision has always been to enable non-digital native companies to actually operate as fast and as efficiently as digital native companies that were born in the last 20 years. For that, without asking them to rethink what I call their bedrock of ERPs, their fundamental transactional landscape because, if we ask them to touch that, they're not going to make it. It's just too big of a transformation. We had to actually enable our technology to plug on their diverse landscape as opposed to asking them to come to our technology.

Long story short, I would say that my advice is, start now. I can see an increasing interest. We're doing pilots in many of the largest companies in the world and it's really critical to get that process started early so that you can build that intelligence relative to how decisions are being made and executed in the company early.

Michael Krigsman: Are there things that a company has to do regarding the data collection? Do they have to change their operations in any way to start gathering the data?

Frederic Laluyaux: No. No, no. That's exactly a critical point, which is, our technology will plug on top of their ERP, whatever they are, and understand the mapping and do all this stuff. You don't have to. There is a bit of work and adjustments. It's never as easy as it sounds but, in a matter of weeks, you actually have that cognitive data layer built on top of your ERPs or, as I said before, your data ocean, links, swamp, whatever you want to call it. No, there is not a lot of preparation.

The preparation is in the validation of the metrics that are calculated by Aera and it really is on the cognitive automation and the cognitive augmentation. When you deploy a skill, you want to make sure that it's adjusted to the way you operate. There is work that needs to be done there.

Michael Krigsman: Ultimately, then, it's giving feedback into the system about the results that have been achieved so that the system can correct itself for the future.

Frederic Laluyaux: Correct, yeah, and it's the system that corrects itself, but it's also sometimes the user that can correct themselves when you actually see this is how you make these kinds of decisions in that type of context for this business value over time. You may decide to change the way you actually think as well. It's a system adjusting and sometimes it's the user readjusting to this new light that we shed on how decisions are made in a company.

Michael Krigsman: Okay, well, unfortunately, we are out of time. It's been a very fast 45 minutes. Fred, thank you very much for taking your time to be with us today.

Frederic Laluyaux: It's been a real pleasure. Thanks for having me, Michael. Thank you.

Michael Krigsman: We've been speaking with Fred Laluyaux, who is the founder and CEO of Aera Technology. Before you go, please subscribe on YouTube and hit the subscribe button at the top of our website and we'll send you great information. Thank you so much, everybody, and I hope you have a great day. Come back next time and we'll see you again soon. Bye-bye.​