Hello, everyone. Good morning, afternoon, and evening, wherever you are. Welcome to the Aera Technology Automation Summit. I'm Laurent Lefouet, chief strategy officer at Aera Technology, and I'm thrilled to have you join this discussion about the art and science of decisions.
Decisions are at the heart of quality automation, and as we are contemplating building and guiding models to make decisions for us, we thought important to reflect on how we are making decisions, what we can learn from it, what we want to keep or change.
So joining me today for this discussion are two senior professionals. I want to introduce you to, starting with Helen Davis. Good morning, Helen.
You are today vice president, beauty and personal care, and also Planning Excellence North America at Unilever, a group you've joined with over 25 years of experience across all aspects of supply chain and around the world.
You started a career in juice manufacturing at the Minute Maid company before moving to Coca-Cola, where you took several roles in manufacturing, logistics, planning. These roles took you to Middle East India, China, Germany, and back to the U.S. to lead their refreshment supply chain.
With all this experience abroad, we can imagine you've seen different ways of getting to decisions, as well as a lot of similarities between people, irrespective of their origins.
You then moved to Reckitt to lead their North American supply operations. And this is where you pioneered digital transformation, leveraging AI and machine learning, and also where we met for the first time, by the way.
Just before Unilever, you were also senior vice president of global manufacturing for Estée Lauder, where you led that transformation for the factory of the future.
Needless to say, you are very familiar with the need to think and act fast in an industry where speed is of essence. Hence, your particular interest on AI technology fits in this context. So we'll get back to this in a second. Welcome, and thanks for being with us today, Helen.
Thank you very much.
On the other side of the table in another part of the world, please welcome Hugh Burgin. Hugh, you're a principal in the technology consulting services practice at Ernst & Young, and the America's leader for data analytics and artificial intelligence within the EY Microsoft Services Group.
You have over 20 years of experience providing data analytics, machine learning, and digital technology services to clients around the world. Leading projects on marketing and sales optimization, natural language processing as well, intelligent forecasting and scenario modeling, for example.
You spend also significant time helping clients migrate their data platform to benefit from more modern technologies. Your perspective on how technologies are used today to changing decisions, and your feedback on their impacts, will be very precious. Thanks for being with us.
Thank you for having me.
An introduction: I want to share with you about a book I read, and I recommend all decision makers and data and analytics professionals to read. It's not only insightful, but fun.
This book, Thinking, Fast and Slow from Daniel Kahneman, is about decades of research he conducted with his fellow Amos Tversky about the yin and the yang of our thinking process. The main thesis is developing the dichotomy between two modes of thoughts, which it refers as System 1 and System 2. Simple names. That's what we are using while making decisions.
System 1 is our fast and instinctive thinking mode, where emotions dominate. While System 2 is our slower, deliberate, and more logical way of thinking where reason dominates.
For example, if I ask you two plus two, you immediately thought four without any effort, right? You probably couldn't even refrain from thinking four. It's your System 1. You didn't do the math. You didn't even consider the calculation. You did it by association, using your memory. But if I ask you 23 multiplied by 17. Now you pause and think first, "I don't really have to do this, right?"
That's System 1 telling you that I'm not really asking, and there is no reward for the effort. But if you decide to do it anyway, then you will have to invoke your System 2. Focus on the operation, and as you do so, it'll prevent you from actively listening to this conversation. So please don't do it.
The point is, we have two ways for making decisions, but they are not equal. System 1 can be turned off. It will be always prompt at propelling unsolicited opinions to your mind, while System 2 requires actually deliberate effort and will to start actually to kick in. So guess who's in control of most of our decisions ... and with what consequences?
Helen, every day bring flows of planning work and unplanned events and business space requires to think and act fast. Information is not easily available. In your role as supply chain leader, what are the decisions you and your team have to make? And how is it working?
Yeah, it's a great question. Because I think I can relate to the Level 1 decision makers, because that's on a daily basis happening, where you have to make quick decisions about, Do I need to produce this? Do I need to air-freight something in to produce? Is it raw impact materials? How do I mitigate any service risk? Can I take something off quality hold to make those service requirements? And what's the priority between customers?
Those tactical decisions are being made every day by my team and myself. There's longer-term decisions out there as well, right? Do I have enough capacity for the future? What does my growth look like? But those are more deliberate discussions that we have with analytics.
Especially when COVID hit, I would say the volatility of it made people have to make decisions pretty fast, and there's a lot of data actually in my company. But is it the right data for that specific moment in time is the question. How does one person triangulate this plethora of data properly to make the right decision? That's where the crux and the difficulty is, if we're not using some kind of automation.
What tools are your team using usually when they have to make those decisions to get the data, to do the analysis? Do they have the time to do so?
Well, we are using Aera Technologies today, but not end-to-end completely, as of yet. If we're not using Aera, then it is based on data points. Maybe it's a customer calling and saying, "Hey, I don't have enough products." Or if you're at the logistics center, you may make the decision that "I don't have enough space. I'm going to move this product to a different location."
And those decisions have consequences. They're looking at a static point in time, and maybe not at the entire universe of data. And the consequence could be you've serviced the wrong customer. You've made your network imbalanced. You've made too much product and you're sitting on a lot of inventory.
So those decisions at a static moment in time when people are looking at data that way, and not in a dynamic way or getting clear direction, can have some negative consequences.
I remember also conversation we had years ago about the people churn. People are actually in the job. They have a lot of experience, they've been there for a while. Then they are moving on to another role in the company. And then you have someone different, someone else with another experience. How does that affect actually the decisions, and the quality of these decisions? And can either way to mediate that?
Yes, I was reading somewhere about millennials versus Generation X. I'm Generation X. We stay with companies for a very long time. Although this is my fourth company, so I'm probably not the usual Generation X.
But with some of the millennials that are coming in, job hopping is the norm. So you have to take time, usually, and train somebody in. If you're in a planning role, that may take three months to get pretty well-versed in what's happening. But of course, you won't have the experience of the planner that's been sitting at that desk for 10, 15 years and has the network and the relationships. Those don't come overnight.
How I've mitigated that in the past is actually using cognitive automation so that how do we take a person that's been there for one day, and make them as good as the person that's been there for 20 years with the same kind of knowledge?
Well, you push that knowledge to them. And instead of them digging for the knowledge, the knowledge comes to them and says, "Hey, you're going to have this issue. You better take this action, or it takes the action for you." Suddenly the turnover rate that you have becomes a non-issue, because you have this continuity through cognitive automation, versus through people.
Yeah, you can crystallize the experience, the memory, and the knowledge in the system collectively. And you can learn from actually the group, not just actually an individual.
That's right. Yeah, that's a good point.
Hugh, a lot of investment and expectations are poured today into advanced analytics in most organizations. How is it working for them? And what can we learn from this first project at scale?
Sure. Well, Helen and her team are obviously making a lot of decisions. You just heard her reference probably 10 different decisions they're making on a day-to-day basis.
Operational teams and leadership teams don't need data and insights just to have insights. They need insights that matter, so that they can make decisions. So that they can actually drive an action, or drive a decision to improve the business or to operate the business.
So often in the past, people have been flooded with data, flooded with insights. But they weren't always the insights that really mattered to make the decision they needed to make. And so I think organizations are getting better and better at that over time, focusing on what they really need.
But from an advanced analytics perspective, like you said, the rise of cognitive automation and machine learning and AI, it used to be a black box. It used to be an experiment that would occur in the basement. And it's now proven.
I mean, companies are having success leveraging these types of advanced analytics all over their organization. Whether it's supply chain and forecasting, or whether it's marketing optimization or inventory optimization; there's many, many examples of tens or hundreds of millions of dollars of value that companies are getting out of these types of advanced analytics.
So the question becomes less about, "Will it work?" But, "What can we do to make it work? How can we embrace it as an organization and give it the chance to succeed?" Rather than question it and treat it like something that's not part of our lives every day.
So the science works, but there is a risk that we don't make it work if we don't approach a problem or the project carefully, it means.
Helen, you actually implemented these projects in your experience. What did you see? I mean, the before and after; how did you manage these transitions? And did you face any challenges?
Yeah, absolutely. I think the first skill set I tried to put in place was forecasting. And I think the immediate go-to was, "There's a problem with the cognitive automation. It's doing the wrong thing. This is why everything's wrong with my forecast."
But we actually locked the demand planners out of the system for those specific SKUs. Because as far as change management, it was a bit difficult because it was the first time for them to see this type of technology.
And it was nice that we locked them out because they were able to see, by exception reporting, what may be an issue. They could question it with the data scientist, but they couldn't actually change it. And what we saw over time is the system was actually right.
Once they started seeing the system was actually right and forecasting properly, much better than they would've done because they would've put their hands on it and changed it, then they started believing. That was great to be the first step, because then when we roll out the next skillset, they were more apt to believe in the system the second time around as we started rolling out more and more.
And I would say that they became huge proponents. They could focus their time on other business needs and strategic needs, and not on these tactical things. So overall, we became a much better supply chain because of it.
That's a question that can be for both of you, in fact. Why do we ... and I say we because as people, tend to challenge what machines or what algorithms ... because it's not the machine, it's really the models are telling us, which is based on science.
Even though we are facing the numbers, facing a reality, we are still actually trying to push back. Why is that, in your opinion?
Well, we did a survey this past year on people's trust that they have in AI or cognitive automation services. And half of the people said they had trust issues, or lack of trust, in the underlying data and the underlying inputs to the machine. And it's a challenge.
There's lack of trust in the inputs that are going into the machine. Then there's lack of trust on the outputs in terms of the decisions that are being made, and the calculations that are being made. So there's a real need, I think, for organizations to put a proactive plan in place around, or a framework in place around, how are they going to monitor these risks to their organization.
How can they proactively assess those risks and determine the impact? What are the guardrails they can put in place to help give people the trust to make those decisions, and get the benefits out of these types of capabilities?
Once we manage to have this trust and we enter more into a co-operation with the machine where the machine actually ... but I say cooperation, because the machine does not collaborate with us. The machine works for our goal, not the machine's goal.
What is, let's say, the next step for people? What is changing for people in their day-to-day's work?
Yeah, I mean, what I would say is operationally, people have an ability to focus on really the decisions that are going to make the biggest impact on the business. And to Helen's point earlier, rather than be bogged down by the day-to-day smaller decisions, they can focus on those decisions that are really going to move the needle for their business.
So I think organizationally, people are really having to rethink how do we structure our teams as they relate to data and analytics and AI? Do we have a centralized organization that owns everything? Do we have these capabilities embedded throughout each of our teams? How do we train our people to give them the skills, either technically to build these solutions, or the skills to know how to run them?
So modernizing those workforce skill sets and structuring the organization to be nimble enough. And the demand for these types of cognitive automation and AI capabilities is far exceeding the supply of people to do the work.
And so, how can you as an organization continue to own that intellectual property and own that differentiator, but still be able to do your day job? It's tough. It's tough. There's not an easy button for that.
Yeah. It's a lot about the work, which is not done today because of the lack of time and resources. And that wouldn't make economical sense to have 10,000 planners.
I think we did the calculation once. If you have 1,000 product that you have to deploy in 100 DCs, and you have 1,000 planners, and only 27 seconds that you have for each planner to think about the combination of [inaudible 00:18:27] location.
And of course, you don't work this way as a person. But to give you an idea, it's 27 seconds to say, "Okay, that's that quantity of product that I want there." Of course, you do it by the aggregated level, and you do your Excel spreadsheet to break it down. That's another debate, but that gives you an idea of the scale of the problem.
So with cognitive automation, Helen, the promise is to go beyond better insight as you described, especially as people are already overwhelmed by information. The objective is really to automate the end-to-end decision-making process down to the execution. So is automation, this automating decision, important for you?
Absolutely. I mean, if we look at the volatility that we see in the market, I was just looking at some of the new players that are coming into the market. You have all these little small brands coming in, and they're quick and they're agile.
Then if you're coming in from a bigger company, it's a little bit difficult. How do you act like a small fish when you're a big fish? You have to be agile in your decision making and be quick. And with all the data points that you have today, you can't be quick in the decision making without automation.
And I would say, being in a supply chain is a ripple effect. If your customer or consumer is pulling a lot of product out suddenly, your reaction to that is usually quite slow if you're not automated. You make calls, you email people, you have meetings, and you say, "Okay, do I adjust my production schedule? I need to order raw and pack materials to get this going."
But if that's all automated, if those checks are happening and say, "Let me put it in your purchase order for you. Let me order your raw and pack for you. You don't have enough. You have to change your production schedule," and it changes it for you.
All those meetings and those tactical discussions go away. And as Hugh said, you get to focus on what's important for the business, which is probably innovations and other things that are really driving the business growth, versus these tactical little decision-making meetings that happen every day.
What do you expect cognitive automation to change if you project yourself in a few years, and it's deployed broadly? Across not only supply chains potentially, but also think procurement, think financing, marketing. What would the future look like?
From my point of view, you'll have a bit of a split in organization. You'll have a potentially an organization that's really data scientists and really focused on driving and improving what we already have.
You'll maybe have still a subset of people that are a bit tactical, looking at things and have things arranged just to clean off data. But then you'll have this group of people that's really strategic, more strategic facing, so that we can push out new innovations.
I think most CPG companies were pushing out maybe 6 or 8% of their volume as new innovations. It's going up to 30, 40% now, so it's really important to get those insights, understand what the customer and consumer wants, and focus our business there. So I think that three-tiered approach ... then obviously, we'll be quick about it.
Then putting innovations out in the marketplace will become much faster. Pulling them out of the marketplace if they're not working will become much faster.
And I think the speed and agility will be there for the bigger companies, as well as the smaller companies. Today, the smaller companies are fast because when you've got 10 people in your company, you can be fast. When you've got 10,000, you're not so fast.
You've been also working a lot with manufacturing. In manufacturing, it's a lot about industry for point zero and the smart factories, digital factories.
Where do you see actually what cognitive automation can bring to that shop floor, more fairly operational processes? Which, by the way, are less digital and more real work. Because this is where you make the product that you have to move out, right? It's physical. Where can cognitive automation help here?
Well, I think there's a couple of areas. If we look inside the four walls of the factory, automation can help there. Because you can actually project if your lines are going to go down, if there's going to be a failure.
Now, most people have a maintenance system that has all the historic data in it that you can cover with cognitive automation right away. You can also connect it to sensors on your lines, and understand that you're having minor stops, and something might happen.
If you're into batch management; a lot of companies do batch recipes. And they're not always right when you make it the first time. You have to adjust some things. But augmentative automation could take the history and say, "Hey, the last hundred times you made this, you added another bag of citric acid. So I'm just going to change your recipe for you and adjust it." I think those are the things that you can do inside the four walls.
I mean, there's a lot of other things, like predictive financials and things in the back office that are happening. But to connect the dots, I use an example of connecting the dots when the customer pulls too much inventory out.
But if a supplier or your manufacturing plant go down, there should also be a ripple effect upstream to say, "You need to change now your customer orders. You need to manage upstream because you're not going to get the product you need, because your supplier or manufacturer went down." And I think that by connecting the dots, the supply chain will become very important.
Thank you. The question coming to my mind as we think about automating more and more of these processes. Hugh, for you: Should we consider machines to become flawless or to be flawless? What are the potential risks there?
Well, no, I don't think we should think of them as flawless. But I do think they can serve a purpose. I think of it as when you think about a narrow decision or a specific decision that needs to be made, it needs to be made hundreds of times or thousands of times. It's less about being flawless, but it's more about being quicker and better.
And if you can make those decisions fast and you can make those decisions more and more accurate over time, then you're going to see a significant improvement.
I think it's also about purposeful design, as you think about how you design a process and you think about embedding automation, or embedding machine learning into that process. Really being purposeful about what's the experience of the people around that process.
What's the input to that process? How does that output get absorbed by the organization, or absorbed by the operation? And really designing it in a purposeful way so that you are maximizing the value you can get out of it?
Whether it's incorrect 15% of the time or what have you, you're still designing for that, and designing to get the most value you can out of it.
With automated decisions, I'm assuming that most of them are actually getting you better and faster. Helen, in your opinion, who's responsible then, if it's the machine making the decision? IT? Who owns the KPI?
This is a matter of opinion, of course. In my opinion is, functionally, my team still owns the operational KPIs. If service is the main KPI, then they're responsible for service.
Now that may mean that they need a different skill set in order to operate. For example, they may need to know a little bit of data science. They may need to know some other data points in order to make those decisions, and understand the root cause of issues.
I think it's really important that we always maintain a relationship with the customer. And if you can't understand what's happening with your KPIs, you can't really talk to the customer about what's driving issues. I think that never goes away. It's just another skill set that we'll need to add to the player's portfolio.
So the ownership stays where it is today, but the profile of people will have to change up. Actually all the training there, we have to bring more science or understanding of these models. They would be responsible for them, as they're guiding them, is your perspective. And probably it will have some impacts also on how we are organizing the organization.
I think Hugh saw the fact that we're going to have, or you said [inaudible 00:27:21] the organization with people more strategic thinking, then you are going to have data scientists to operate these models. There will be potentially some organizational impacts as well.
So to wrap up on our conversation, what would be your practical advice from your experience to start with a cognitive automation journey?
Yeah ... Helen, do you want to go first?
No, go ahead, Hugh.
My point of view is that cognitive automation and AI and machine learning are proven, as we talked about earlier. They do work. They can drive competitive differentiation. But they are not easy. And they do fail a lot of the time.
So the question is what as an organization can we do to enable these things, to have the chance to succeed in our company? Because we know they're working in other places. They're working in parts of our company, or they're working in our competitors'; how can they be successful here?
I think one of the keys is executive sponsorship, C-level sponsorship, senior level sponsorship. Because you have to create a culture where you embrace it, and you give it a chance to be successful.
If there's not that top-down sponsorship, people like Helen who are embracing it, then people easily treat it as a proof of concept. "We did this little study, it was cool. We'll push it aside. And now we'll go back to our real job."
And I think that top-down sponsorship is really a key ingredient to making these things successful in companies like the ones we're talking about today.
Yeah. I would only add to that I totally agree with what Hugh said. I've been lucky; I've had executive sponsorship at the companies that have implemented cognitive automation. I think it's also important to have frontline ownership, and involve the frontline in the process. It's not like building this out and saying, "Okay, now go use it." But involve them in every step of the way.
The IT partnership is really important. Fixing things within 24 hours if they are failing, because that's part of change management to ensure people are not getting disgruntled. And I think Hugh's point around failure ... failure is an option, right? And we have to make it an option.
I remember when I was doing the forecasting piece and we lost our forward demand. We had dropped off after a month. So we only had a month's worth of demand, and everything fell off. We couldn't go to SnOP, I made the decision to cancel everything. And we had to do manual work to get it back into the system.
But those are the type of failures you have to be willing to take in order to make this change. Because once we made the change and everything was on autopilot ... my goodness, my demand planners loved it, because they weren't battling every day and fighting with data.
All right. Where there is a will, there is a way, and that goes for the sponsorship. And we need also the adoption from the frontline people to embrace it as well, so that we make their life easier. In fact, the objective is not to make it more complex.
I think that concludes this discussion very well. I want to thank you both for being with me today. Well, I'm looking forward actually continuing this discussion and journey and implementation of projects with you, Helen, with you Hugh. And I wish you all and everyone a very nice day. Thank you.
Thank you, Laurent.