Deloitte: Digital Supply Networks @ the Cognitive Automation Summit

By
Panel of Thought Leaders
30m
Deloitte: Digital Supply Networks @ the Cognitive Automation Summit

How customers think about cognitive automation, and how it will be used in the future of supply chain.

Laurent Lefouet:

Good morning, afternoon and evening. Welcome to the Aera Cognitive Automation Summit today. I'm Laurent Lefouet, Chief Strategy Officer at Aera Technology and I'm thrilled to have you in this session with our partners, where they will share the experience with customers and how they think about cognitive automation and how it will be shaping the future of supply chains. Joining us today are two pioneers from Deloitte coming from both the US and Europe. I'm saying pioneers because they believe their role is not only to bring subject matter expertise in their field of digital supply chains and they clearly have a lot but they think their mission is also to bring new ideas, new technologies, new experience that matters to their customers.


And this is how our path crossed a few years back and we've been working together since the beginning of this Aera journey. As you listen to them you will learn how they re-imagined agility for their customers. You will get the first glance at cognitive automation in the context of planning and fulfillment and how the cognitive control towers enable customers to then drive execution. This presentation is based on existing projects we have with mutual customers and this session are very short so without further ado, please welcome Rafael Calderon, head of Global Synchronized Planning & Fulfillment market at Deloitte here in the US and Kevin Overdulve, head of EMEA Logistics & Distribution practice and of Deloitte's cognitive control towers offering in Europe. Gents, it's still great to have you here. Thanks for joining us today.


Rafael Calderon:

Thank you. A pleasure to be here always with our partners from Aera to talk about exciting things so let's just see you here. I wanted to start the conversation today, the dialogue today with what we're seeing in the market in terms of the last three, four years in terms of transitioning of supply chains into this digital world. And the concept that we called in at Deloitte some years ago was the concept of Digital Supply Networks. And it was really a concept that came about from a shift, a structural shift that we started to see in the market around how we traditionally had thought about supply chain as a very linear flow and how in supply chain typically the way that supply chain came about was adding pieces to that supply chain and with it systems and processes that all were predicated on the very linear relationship between the pieces of supply chain.


Over time what started to happen was with the advent of the cloud, with the advent of computing power, the ability to do things in memory, the ability to do the things remotely and access much more computing power. What we find is a tremendous opportunity to process and do things that we couldn't do before. The amount of advancement that we've had in terms of algorithms and math and the ability to process large amounts of information have really created a tremendous opportunity for supply chain to really advance its capabilities in ways that it couldn't do before. What we try to do on the right side of the page is to really describe how we think about supply chain in this new era. In this new era we don't really have those barriers anymore because we're working on single data models across the full value chain. We don't need to transact over different systems and different types of structures, instead we can transact over a common data layer and we can actually imagine scenarios and decisions that include all variables across the whole ecosystem.


Not just between our company but also the ecosystem partners that we transact whether it's a customer or it's a supplier, is extending those avenues and being able to think about a single set of information that we can transact against. We now are able to process a tremendous amount of information and be able to bring that information to bear in terms of the decision that we make. We're also able to combine the physical and the digital loop and we are able to take sensors, IOT, 5g, different kinds of things to enable a tremendous amount of unstructured and structured data together to be able to then process that information and analyze that information using techniques such as Artificial Intelligence, machine learning and be able to really understand patterns that we couldn't do before and reach levels of performance that supply chain could never get to before.


So in this new world of DSN or Digital Supply Networks, we really are looking at a much more interactive set of interactions between the different parts of the supply chain. We're looking at a set of entities that can interact at any point where consumer information can be captured and immediately made available to the product development team, something we couldn't do in the past. Or when information is available based on the conditions of the supply chain as they change, that information is available to every single node in that supply chain and everybody's able to use that information to their advantage. The world of today looks different, feels different and that's hence why we call this concept of Digital Supply Networks.


Laurent Lefouet:

But this is clearly a radical shift in how supply chains are organized and our supply chain orchestration is structured. What are the barriers your clients have to face as they go through this transition?


Rafael Calderon:

It's a great question Laurent. I think the first... Always the barrier that everybody thinks about is data. And yes, absolutely data and the ability to create the data that we need to be able to do things is always an important part. But if you actually imagine from my vantage point, the biggest barrier is us. Is the organizational dynamics. The eagles are the humans that get in the middle of the advancement of technology. There is so much that goes along in a multi matrix organization around my control power, my decision rights. Whether this is my territory. This is my system versus your system. This is my project versus your project. It just totally gets in the way. So to me the whole organizational dynamics and the politics between the different parts of the organization that to me is the biggest barrier to overcome.


Because at the end of the day we're trying to align a lot of different agendas to a very common set of objectives. In the past it was easy for all of us to transact in our silo because technologically or informationally we couldn't get to a point where we were more integrated. Now that we're more integrated we have to bring those barriers down. We have to think about whether it is no longer your demand planning model or my demand planning model between retailers and consumers, it doesn't matter anymore. At the end of the day it doesn't matter what hands on the queue are there but it's more around what is that integrated decision flow for the whole company. I would say that's probably the one thing that we see as the biggest thing to figure out in terms of the advancements in the DSN space.


Laurent Lefouet:

So what you're saying is that there is a technical problem with the data but there is more of an organizational problem which is to align people around a set of KPIs that would be common, right? And because if you actually drive behavior with KPIs you have to make sure that there is no conflict between different objectives across different teams as you put this model in place.


Rafael Calderon:

Exactly. That's spot on. It's all about the incentives, right? It's all about now that we have the possibilities of being able to work like this... I can give you an example, right? In the area of planning, right? We used to work with demand planners, material planners, supply planners. In the world of tomorrow, we can work with a network banner or a connected planner where I can actually have all the information in front of me and make all decisions simultaneously. What does that mean in terms of the change that I'm about to undertake in terms of my role? What are my KPIs? How do I change that? How do I change those behaviors? And more importantly, how do we get the humans to see this huge opportunity technology as a true enhancement of their capability and not as an enemy?

The philosophy of the company in terms of how they think about digital is really important because what we don't want is for the human capital resource in the company to see this whole technology as a replacement of them or as a threat. It has to be a marriage of the two and in tremendous synchronicity where we can find the weight for computers to be good at what they do and humans to be good at what they do. That together is a one plus one equals three equation.


With this what we wanted to do was to illustrate... I'm more kind of one click down in terms of how we think about DSN and the concept of agility is really important here. Because when we think about things like COVID and others there's a tremendous need for companies to be truly integrated, where information is made available and the frequency of information increases and the need for me to act on a smaller piece of time becomes critical. In a world pre COVID, we had a lot of stability in many of the supply chains where there's a repeatable pattern, we got a lot of statistics and we're able to leverage history to be able to make decisions. In the world of COVID things are changing week by week. All that information that we had in the past that we've collected for years becomes useless.


My demand focus all of a sudden went from great accuracy to no accuracy because the drivers and calls of fact completely changed. The way that people behaved had completely changed, right? How do we adapt the supply chain to that? How do we build a supply chain around that? What we see here in the page is trying to depict this concept of the integrated supply chain in terms of what we call a cognitive automation model built in a layer that connects the whole company. On the top you see capabilities. Each of these capabilities is represented with hexagons and the different colors talk about the different parts of supply chain. The language that we're using here is a new language that we are developing with the ACM which is eventually going to replace the score model. Deloitte is working with the ACM to replace the score model for the digital era.


What you see on the screen is a little bit of that. It's a little bit of that language. And what we're doing is we're actually thinking about this capabilities as a set of things that we do in the company. The supply chain is at the core and it connects with commercial and one side and finance and the back office on the other. And then it extends to the data and the partners connecting the full ecosystem. In the middle of what we see is a layer. Is a layer of connectivity that has three fundamental things. One is a digital twin which is an exact replica of the full value chain that can tell us exactly how the movie's playing. That means virtually I know exactly what's happening in that supply chain and I can make decisions, vet decisions in a digital way before I vet them physically.


In the middle I have my network intelligence which is all the math and algorithms that I can leverage for the solutioning of any type of optimization problem. And then the automated workflow which is as I get good at this, I can start to automate a lot of the things that the humans do so that I can start to work on the more complex problems and automate all the stuff that I do [inaudible 00:10:33] day in day out. The idea here is with a copy of the supply chain and the value chain, the math to act on that data and the ability to automate, I can create almost an enhancement of the human where if I'm an executive, this is a entity that is helping me make better decisions, act better and be able to simulate things in a matter of time that I would never be able to do with me or my team.


It basically enhances the human component of the enterprise by a factor of 10 and enables us to do things that we couldn't do before. This brain or this layer then acts, takes information from all the systems of the company and then gives back information eventually helping take over those systems to try to automate and become a cognitive enterprise that is truly performing at a level that we couldn't do before. That is the concept that we were trying to illustrate here in terms of the agility and what our vision is in terms of how companies are going to evolve to this cognitive company that we're displaying in the page.


Laurent Lefouet:

In the context where we are today and with the black Swan like the COVID, massive one, where you mentioned the AI models or the planning models that you created are falling short of actually enabling organizations to anticipate what's going on. How does a digital supply network and what you just described is helping companies respond to these kinds of shocks that are unpredictable by nature?


Rafael Calderon:

The first thing is by enabling a lot of sources of information across the value chain we're able to sense what's happening, right? So think about IOT. If I take IOT and I display it from a warehouse, trucks, factories, products and embedding information I can now capture a lot of information about my ecosystem and I can have that information available. The next thing is what do I do with my response? If I'm able to take in what's happening in the market, able to identify a change in a pattern, a change in a demand but forecast in a particular city. A change in a movement or a forecasted problem because I see something going on that's going to create a problem for me.


If I have access to all that information, I'm not able to respond a lot faster. How do I configure my whole supply chain to be able to take advantage of the frequency and amount of information that I'm now able to capture with this concept of DSN and IOT? If I can capture a lot more information, what is the type of math and the type of integration that I need to put in place to truly be able to say, "Information available make it usable to every node so they can make the right decisions to be able to prevent a failure and to be able to then drive a better service at a lowest cost." It's able to be much more predictive and much more active than what we had in the past with the configuration that we had before the DSN concept.


Laurent Lefouet:

In a summary, if before you had AI primarily focused on planning and predicting what's going to happen in the case of shock, physically it's about really sitting on top of the signals and enabling an AI to augment the responsiveness of the entire organization by reacting to that event very quickly, being connected to data and being connected to people.


Rafael Calderon:

Exactly. It's basically low on replacing the physical effort, the manual effort with computers to really help us put in automatic a lot of things that in the past we would have to just manage by sheer power, by sheer force in terms of putting a number of people to monitor things. It's creating a tremendous capability in terms of the ability to sense and respond in a lot faster and more effective ways than we had in the past.


Laurent Lefouet:

Any example you could share with us where you actually did implement something similar?


Rafael Calderon:

Yeah and I'll pass it over to Kevin to talk about a real example, right? To hopefully bring this to life for everybody in a real tangible way so Kevin.


Kevin Overdulve:

Yes. Thanks Rafa. And thanks Laurent and the Aera team for inviting us for this session. We talked in the previous pages a little bit about agility and responsiveness and how do you actually take real-time information from execution and use that in order to drive decisions. The example that I would like to present to you today is a case that we developed together with Aera just recently for a large beverage company. The key issues of this beverage company were that on the one hand they saw that their costs of sales in terms of logistics was fairly high. On the other hand they saw they had a fairly high loss rate so they were basically missing out on sales because of a supply chain performance which was not up to par. One of the key reasons was that they were not able to leverage the data they had sitting across a multitude of different systems in a smart way to drive day-to-day decision-making.

Obviously, if you're aware of the concept of a control tower this will start to ring a bell but control tower can be defined in a number of different ways. And yes, a control tower in a traditional definition is a capability that would continuously pull in information from execution, compare that versus basically what you have planned for and start to notify, give alerts on certain exceptions, deviations from plan taking place. And that's a basis that's where you need to start. For this company we took it one step further. Obviously with those notifications, with those anomalies in the supply chain you want to use that information in order to drive a certain decision almost on an ongoing basis. Next to that typical control tower functionality, we developed what we would call a cognitive skill in order to automate a certain decision-making process that people today in the organization are doing on a day-to-day basis but where they are typically lagging behind and running behind effects. On the left side of the page you actually see those different levels of what we would call a modern control tower.


Obviously you start with your data layer. You have certain descriptive and reactive components to allow you to react immediately to anything happening in that supply chain which you haven't planned for. And that could be on the hourly basis or could be on a half hour or even a minute basis. And then as you go up into the maturity curve, you enter into the space of diagnostics, predictive, prescriptive. All of those components that typically we run through if we are about to make a certain decision which can also be automated of course. And as you grow and adopt more of this cognitive automation into the processes and decision-making on a day-to-day basis, at a certain point in time you will move much more towards a control tower, a modern control tower which has the capability to self-heal or self synchronize the supply chain on a day-to-day basis. To give you some examples of the functionalities that we developed, on the bottom and the traditional control tower view we developed a capability actually infused by the entire COVID wave which we called a rapid response control tower and we'll go into a little bit more detail later on.


But basically the idea was in a matter of three to four weeks we developed a capability bringing together information across manufacturing execution systems across a dozen of different WMS systems, a DMS system, ERP, SAP APO. Basically bringing all of that information together to allow people to have sort of an iPhone notification type of application but then are targeted towards their supply chain with all of the details behind it of course. From a cognitive skill perspective we focused on their out of stock problem. Whenever an out of stock risk has been identified by the solution, the cognitive skill starts running through a series of different actions that can be taken in order to avoid that stock-out or in order to mitigate the negative impacts of that stock-out.


And you see there on the page that we looked at two potential actions: either inventory deployments and can I move product that's sitting in another warehouse to the warehouse that is basically destined to fulfill the order? Or production scheduling. Can I do an update of my production schedule? And with that actually avoid the stock-outs update space. What we also developed was a decision board. The decision board is basically a monitoring tool which looks at how well are the recommendations provided by the solution adopted. And to give you one simple example, the inventory deployments skill after just two sprints of three weeks and the recommendations coming out of that were accepted for more than 70%. Basically 70% of the work that is currently being done by a user after six weeks could be fully automated because everybody was agreeing upon those recommendations.


Laurent Lefouet:

It looks like things went pretty fast because I can see this is the work done in the first 100 days. You mentioned the two sprints of three weeks.


Kevin Overdulve:

Yes.


Laurent Lefouet:

First question is what were some of the learnings that you took along the way in implementing this tower?


Kevin Overdulve:

It indeed went pretty fast. Actually the 100 days that you see on the screen involves a discovery phase, a design phase together with the users then three sprints of development. And then even at the end of it another three weeks where we handed the solution over to the business and they actually started using it as if it was real life, thereby confirming the benefits coming out of the solution in this cognitive automation. I think what's important in order to run this in 100 days is to on the one hand be laser focused on a certain scope element and obviously needs to be something that adds value and can be done and implemented even with a certain amount of data available and the accuracy of the data. But continuing to stay laser focused on what really adds value, what is really differentiating versus what they have and we don't want to recreate certain reports people already have, is I would say one key learning thing.


I think the second key learning which is important, kind of across the board if you look at cognitive automation, is to get people alone and educate people to see how this concept kind of revolutionizes the way how work is being done. To reference point earlier on, it's not always easy to kind of get people away from, "This is the way... How I do it right now. This is my territory. I don't want to give that up." But cognitive automation really has the potential to kind of embed certain digital agents into your company that really take over all of that work that humans shouldn't be doing but are doing way too much today.


Laurent Lefouet:

And in this specific example how the organization responded to this new way of working.


Kevin Overdulve:

Actually from the start we had discussions on how to position cognitive automation as a capability in the company. The response eventually was very positive in the sense that even before we were starting to put a design on paper, people were already buying into that vision and into the way this capability can be positioned as a complimentary thing next to I would say their planning systems and all of their transactional automation systems. That was a very important change component right from the start.


Laurent Lefouet:

You mentioned that you had a more detailed example of the thing that you've built as part of the tower.


Kevin Overdulve:

Yes. Exactly. The first one that I would like to deep dive on is our rapid response control tower. You might think this is an ordinary dashboard but that's definitely not true in the sense that you obviously see here information which could also appear in a dashboard looking back two or three weeks after the fact. What do you see here? Is all information that's being uploaded on a 10 minute basis.


We focus on five adherence KPIs is how we call it from the left to the right. Am I getting in what I was expecting to get in order to steer my feeling line? Am I feeling what I was expected to feel? Am I shipping out what I was expected to ship out or what I plan to ship out? How are my sales consuming the forecast that I had and how am I doing in terms of stock position across my network? You also see... It's even a little bit but you see orange and red color coding which is basically to what extent this KPI, this adherence KPI is meeting or not meeting the thresholds they had set forth. This is for example the manager's view. Basically a manager could log in at every time of the day and basically track how well these plans versus the execution KPIs were performing in the organization.


At one level lower we had transactional exception reports where the planners were actually getting notifications whenever a shipment was not arriving, whenever the production line was down and hence certain orders could not be fulfilled. Whenever a retailer dumped a massive order on them at the end of the month and they weren't expecting that or they weren't able to accommodate that. So that's the level of lower this is the middle level. We also had one executive level on top of that which was basically giving a retrospect of the different supply chain issues occurring in that supply network of this client and how they were evolving on a day-to-day basis. The executives could basically focus on one or two issues that were pending for five, six, seven days and take the right action in order to make sure that the next day wasn't occurring on their control tower.


And obviously you need to have the technology in order to be able to continue to load this execution information and continuously compare that with the plan. That's of course where the power of the data crawlers functionality of Aera comes in handy. This is visibility not visibility traditional reporting style but visibility as if you had an iPhone notification which was currently or continuously giving you updates and things going wrong in your supply chain. If we then enter into the next layer we start talking about that stock-out skill. That stock-out skill is linked to this control tower. As the control tower gives you information on how well our stock position is versus our plan, it also gives you all the details of that stock position at skew level, at skew location level even. Based on this view, actually Aera started to populate an entire list of potential stock-out risks by... As I mentioned before, combining all of that execution and planning information into one source of truth.


I'm now stepping into the shoes of a planner at this company. In the morning that planner would log into Aera and would open up their preventive stock-out skill and they have a list of 97 pending recommendations, 260 more are completed already. They can filter or they can sort their lists of recommendations from high to low and total benefits or the margin contribution at risk. And they could select for example, the third one to have a deeper look into what does this recommendation actually entail? When clicking on this, this green below opens and Aera actually immediately gives you the recommendation. In this case Aera recommends transferring a certain amount of pallets of a certain skew from warehouse A to warehouse B because at warehouse B there's... Or that skews the risk of a stock-out in two days from now. You also see here all the details of this recommendation and you see the planning book here and you basically see the reason why Aera has recommended to do a stock transfer order for the skew from location A to location B.


Laurent Lefouet:

[crosstalk 00:28:38] Just giving one question is in that example, the stock-outs it's a prediction of stock-out, it's not an actual stock-out, correct?


Kevin Overdulve:

Exactly. We're continuously... We're basically looking at an evolving picture of stock-out risks over the next seven to 14 days. And as for execution I'll always deviate from that plan. You see that there are quite some line stoppages or sales are not under control versus forecasts. Certain in-bounds of raw materials are not there. They're basically taking all of that information into account to predict at skew location level when you have a risk of a stock-out. Even more, we have developed the machine learning model which does this stock-out risk identification. It takes into account 32 different features, elements that have an influence on a stock-out risk and also takes into account external information like weather information, in order to basically build an engine that is better able to predict whether a stock-outs actually has a higher risk of taking place or not. And the good thing is that obviously compared to deterministic rules this machine learning model gears you towards those skews that will really be in stock-outs and not the ones that you think will be in stock-out based on a stem for business rule.


Laurent Lefouet:

That is obviously a lot of capabilities that you build in there and probably not enough time in this session to go through all of them. What would be your... Let's say conclusion and recommendation for your customers for the people that... Who are watching a decision?


Kevin Overdulve:

My initial recommendation would be... And something that I hear a lot from clients was also mentioned earlier on. We don't have the right master data, we don't have the right accuracy of data. I think traditionally even companies who struggle with their data would already benefit a lot from just having all of that information connected to each other in a virtual DNA with a machine basically picking out the elements, pointing them towards the elements that they need to look at and they need to make decisions on.


And I think all of this can actually be realized in a fairly short amount of time. I would always advise a client to immediately start thinking about what is my biggest issue and what is the typical decision making process around that issue and how do I move and make a first move or step towards cognitive automation. And then think through what kind of visibility do I need? What kind of data do I need? Develop such a use case, bring it into production. Actually, sometimes even scale that geographically or to other functions and then start tackling a new use case but always from the issue back. I've seen too many companies starting on a journey where they decided, "Let's boil the ocean right from the start." And then you get stuck at a certain point in time and you're not able to deliver at the speed and value that you want to deliver.


Laurent Lefouet:

[crosstalk 00:32:07] business problem.


Rafael Calderon:

One thing to add to what Kevin said is not all data is created equal. When we say statements like, "I have bad data." We all have that data. Now having bad data it's like having lower back pain. Everybody has some type of lower back pain. The question is there's elements of data that are actually very useful to create this MBP. Take a little piece of data as Kevin is suggesting, create that demonstration of the use case and then use that to garner the support and get people excited around the story and the narrative to then get the support that you need to go make the additional investment. An important aspect of DSN is to move quickly, right? Is to move quickly and I think Kevin that's kind of what you guys are experiencing in your particular project.


Kevin Overdulve:

Exactly. The best situation you can basically be in is that you indeed pick a pilot, you operationalize your pilot solution and you kind of get that ROI of six to nine months max and you start using the benefits you're reaping from that capability in order to fund the rest of your journey. But it's a continuous journey. As you've implemented two or three or four different use cases, you'll probably see things that you couldn't have imagined before. Things that with the knowledge you have today you're just not able to see issues that you don't know about yet. That journey is also kind of an evolving thing.


All right. And then just closing off, I think when we look at these digital pilots and especially if we talk about implementing cognitive automation, we'd like to use these three steps in order to again, keep that laser focus. It's important to start thinking big. Why? As I mentioned before, one of the success criteria of the projects that we've done together with Aera is to position cognitive automation very well. There needs to be some kind of a dream and a longer term vision first and everybody needs to buy into that vision and then you can actually start delivering upon that vision. Start small around choosing what pilot that is on the one hand valuable and so your lower back pain as Rafael mentioned it. And on the other hand is also feasible with the amount of information and the systems that you have and you would be surprised how much is actually feasible in those small pilot setups.


But then of course with the pilot you won't be heading towards the speed to value that you like to head towards. The last phase scaling fast is hyper important. We typically speak about a couple of different ways how you can start scaling and you can either go very broad and start adding use cases or you can go very far and scale things geographically or for example in logistics across the number of modes of transport. Or you can go deep and further invest in taking advantage of machine learning and adding advanced analytics because that has kind of a ripple effect and a bigger effect on your short term ROI.


Rafael Calderon:

One thing to add to the point on scaling fast is what we've seen with our clients is the importance of changing the way that we measure projects. Incorporate setups. Typically we'd take a project, we measure its ROI and we try to create a business case around it. In the world of DSN we have to think about almost having the private equity mentality. There's a number of things that we're trying and we're measuring the ROI of the whole portfolio. Some things are going to fail. Some things are going to work. What's the overall return of that total portfolio? Just taking that step and thinking about [it] like that just opens the aperture and provides the right flexibility to be able to have the right level of incentive for your executives to want to make those bets and try those things. But if you measure everything on a single project basis, it's going to be very difficult for people to take a chance and that's part of what we see. Sometimes companies struggle, you have to change the mindset, take a PEA mentality and really try to think about this as a portfolio of projects in a total ROI.


Laurent Lefouet:

Very clear. Amazing journey so far. You've been with us working together since the beginning. Very thrilled to see how this is unfolding and actually becoming real with multiple customers that we have in process today. Thank you for joining us in this first Cognitive Automation Summit. This year it's online. I'm sure that one day it will be on site and I'm sure that there will be a lot of people coming your way on stage to ask questions, raise their hand and so on. I'm looking forward to this moment. Meanwhile they can of course reach out to you, ask questions and I'm looking forward to our next conversation.


Get social with us!

Follow us on LinkedIn: Cognitive Automation Community
Follow us on Twitter: CognitiveAutomation

By
Panel of Thought Leaders
,

Kevin Overdulve & Rafael Calderon Director of Logistics & Distribution, and Global Synchronized Planning & Fulfillment Leader

Published:
September 24, 2020
Share:

Deloitte: Digital Supply Networks @ the Cognitive Automation Summit

How customers think about cognitive automation, and how it will be used in the future of supply chain.

Laurent Lefouet:

Good morning, afternoon and evening. Welcome to the Aera Cognitive Automation Summit today. I'm Laurent Lefouet, Chief Strategy Officer at Aera Technology and I'm thrilled to have you in this session with our partners, where they will share the experience with customers and how they think about cognitive automation and how it will be shaping the future of supply chains. Joining us today are two pioneers from Deloitte coming from both the US and Europe. I'm saying pioneers because they believe their role is not only to bring subject matter expertise in their field of digital supply chains and they clearly have a lot but they think their mission is also to bring new ideas, new technologies, new experience that matters to their customers.


And this is how our path crossed a few years back and we've been working together since the beginning of this Aera journey. As you listen to them you will learn how they re-imagined agility for their customers. You will get the first glance at cognitive automation in the context of planning and fulfillment and how the cognitive control towers enable customers to then drive execution. This presentation is based on existing projects we have with mutual customers and this session are very short so without further ado, please welcome Rafael Calderon, head of Global Synchronized Planning & Fulfillment market at Deloitte here in the US and Kevin Overdulve, head of EMEA Logistics & Distribution practice and of Deloitte's cognitive control towers offering in Europe. Gents, it's still great to have you here. Thanks for joining us today.


Rafael Calderon:

Thank you. A pleasure to be here always with our partners from Aera to talk about exciting things so let's just see you here. I wanted to start the conversation today, the dialogue today with what we're seeing in the market in terms of the last three, four years in terms of transitioning of supply chains into this digital world. And the concept that we called in at Deloitte some years ago was the concept of Digital Supply Networks. And it was really a concept that came about from a shift, a structural shift that we started to see in the market around how we traditionally had thought about supply chain as a very linear flow and how in supply chain typically the way that supply chain came about was adding pieces to that supply chain and with it systems and processes that all were predicated on the very linear relationship between the pieces of supply chain.


Over time what started to happen was with the advent of the cloud, with the advent of computing power, the ability to do things in memory, the ability to do the things remotely and access much more computing power. What we find is a tremendous opportunity to process and do things that we couldn't do before. The amount of advancement that we've had in terms of algorithms and math and the ability to process large amounts of information have really created a tremendous opportunity for supply chain to really advance its capabilities in ways that it couldn't do before. What we try to do on the right side of the page is to really describe how we think about supply chain in this new era. In this new era we don't really have those barriers anymore because we're working on single data models across the full value chain. We don't need to transact over different systems and different types of structures, instead we can transact over a common data layer and we can actually imagine scenarios and decisions that include all variables across the whole ecosystem.


Not just between our company but also the ecosystem partners that we transact whether it's a customer or it's a supplier, is extending those avenues and being able to think about a single set of information that we can transact against. We now are able to process a tremendous amount of information and be able to bring that information to bear in terms of the decision that we make. We're also able to combine the physical and the digital loop and we are able to take sensors, IOT, 5g, different kinds of things to enable a tremendous amount of unstructured and structured data together to be able to then process that information and analyze that information using techniques such as Artificial Intelligence, machine learning and be able to really understand patterns that we couldn't do before and reach levels of performance that supply chain could never get to before.


So in this new world of DSN or Digital Supply Networks, we really are looking at a much more interactive set of interactions between the different parts of the supply chain. We're looking at a set of entities that can interact at any point where consumer information can be captured and immediately made available to the product development team, something we couldn't do in the past. Or when information is available based on the conditions of the supply chain as they change, that information is available to every single node in that supply chain and everybody's able to use that information to their advantage. The world of today looks different, feels different and that's hence why we call this concept of Digital Supply Networks.


Laurent Lefouet:

But this is clearly a radical shift in how supply chains are organized and our supply chain orchestration is structured. What are the barriers your clients have to face as they go through this transition?


Rafael Calderon:

It's a great question Laurent. I think the first... Always the barrier that everybody thinks about is data. And yes, absolutely data and the ability to create the data that we need to be able to do things is always an important part. But if you actually imagine from my vantage point, the biggest barrier is us. Is the organizational dynamics. The eagles are the humans that get in the middle of the advancement of technology. There is so much that goes along in a multi matrix organization around my control power, my decision rights. Whether this is my territory. This is my system versus your system. This is my project versus your project. It just totally gets in the way. So to me the whole organizational dynamics and the politics between the different parts of the organization that to me is the biggest barrier to overcome.


Because at the end of the day we're trying to align a lot of different agendas to a very common set of objectives. In the past it was easy for all of us to transact in our silo because technologically or informationally we couldn't get to a point where we were more integrated. Now that we're more integrated we have to bring those barriers down. We have to think about whether it is no longer your demand planning model or my demand planning model between retailers and consumers, it doesn't matter anymore. At the end of the day it doesn't matter what hands on the queue are there but it's more around what is that integrated decision flow for the whole company. I would say that's probably the one thing that we see as the biggest thing to figure out in terms of the advancements in the DSN space.


Laurent Lefouet:

So what you're saying is that there is a technical problem with the data but there is more of an organizational problem which is to align people around a set of KPIs that would be common, right? And because if you actually drive behavior with KPIs you have to make sure that there is no conflict between different objectives across different teams as you put this model in place.


Rafael Calderon:

Exactly. That's spot on. It's all about the incentives, right? It's all about now that we have the possibilities of being able to work like this... I can give you an example, right? In the area of planning, right? We used to work with demand planners, material planners, supply planners. In the world of tomorrow, we can work with a network banner or a connected planner where I can actually have all the information in front of me and make all decisions simultaneously. What does that mean in terms of the change that I'm about to undertake in terms of my role? What are my KPIs? How do I change that? How do I change those behaviors? And more importantly, how do we get the humans to see this huge opportunity technology as a true enhancement of their capability and not as an enemy?

The philosophy of the company in terms of how they think about digital is really important because what we don't want is for the human capital resource in the company to see this whole technology as a replacement of them or as a threat. It has to be a marriage of the two and in tremendous synchronicity where we can find the weight for computers to be good at what they do and humans to be good at what they do. That together is a one plus one equals three equation.


With this what we wanted to do was to illustrate... I'm more kind of one click down in terms of how we think about DSN and the concept of agility is really important here. Because when we think about things like COVID and others there's a tremendous need for companies to be truly integrated, where information is made available and the frequency of information increases and the need for me to act on a smaller piece of time becomes critical. In a world pre COVID, we had a lot of stability in many of the supply chains where there's a repeatable pattern, we got a lot of statistics and we're able to leverage history to be able to make decisions. In the world of COVID things are changing week by week. All that information that we had in the past that we've collected for years becomes useless.


My demand focus all of a sudden went from great accuracy to no accuracy because the drivers and calls of fact completely changed. The way that people behaved had completely changed, right? How do we adapt the supply chain to that? How do we build a supply chain around that? What we see here in the page is trying to depict this concept of the integrated supply chain in terms of what we call a cognitive automation model built in a layer that connects the whole company. On the top you see capabilities. Each of these capabilities is represented with hexagons and the different colors talk about the different parts of supply chain. The language that we're using here is a new language that we are developing with the ACM which is eventually going to replace the score model. Deloitte is working with the ACM to replace the score model for the digital era.


What you see on the screen is a little bit of that. It's a little bit of that language. And what we're doing is we're actually thinking about this capabilities as a set of things that we do in the company. The supply chain is at the core and it connects with commercial and one side and finance and the back office on the other. And then it extends to the data and the partners connecting the full ecosystem. In the middle of what we see is a layer. Is a layer of connectivity that has three fundamental things. One is a digital twin which is an exact replica of the full value chain that can tell us exactly how the movie's playing. That means virtually I know exactly what's happening in that supply chain and I can make decisions, vet decisions in a digital way before I vet them physically.


In the middle I have my network intelligence which is all the math and algorithms that I can leverage for the solutioning of any type of optimization problem. And then the automated workflow which is as I get good at this, I can start to automate a lot of the things that the humans do so that I can start to work on the more complex problems and automate all the stuff that I do [inaudible 00:10:33] day in day out. The idea here is with a copy of the supply chain and the value chain, the math to act on that data and the ability to automate, I can create almost an enhancement of the human where if I'm an executive, this is a entity that is helping me make better decisions, act better and be able to simulate things in a matter of time that I would never be able to do with me or my team.


It basically enhances the human component of the enterprise by a factor of 10 and enables us to do things that we couldn't do before. This brain or this layer then acts, takes information from all the systems of the company and then gives back information eventually helping take over those systems to try to automate and become a cognitive enterprise that is truly performing at a level that we couldn't do before. That is the concept that we were trying to illustrate here in terms of the agility and what our vision is in terms of how companies are going to evolve to this cognitive company that we're displaying in the page.


Laurent Lefouet:

In the context where we are today and with the black Swan like the COVID, massive one, where you mentioned the AI models or the planning models that you created are falling short of actually enabling organizations to anticipate what's going on. How does a digital supply network and what you just described is helping companies respond to these kinds of shocks that are unpredictable by nature?


Rafael Calderon:

The first thing is by enabling a lot of sources of information across the value chain we're able to sense what's happening, right? So think about IOT. If I take IOT and I display it from a warehouse, trucks, factories, products and embedding information I can now capture a lot of information about my ecosystem and I can have that information available. The next thing is what do I do with my response? If I'm able to take in what's happening in the market, able to identify a change in a pattern, a change in a demand but forecast in a particular city. A change in a movement or a forecasted problem because I see something going on that's going to create a problem for me.


If I have access to all that information, I'm not able to respond a lot faster. How do I configure my whole supply chain to be able to take advantage of the frequency and amount of information that I'm now able to capture with this concept of DSN and IOT? If I can capture a lot more information, what is the type of math and the type of integration that I need to put in place to truly be able to say, "Information available make it usable to every node so they can make the right decisions to be able to prevent a failure and to be able to then drive a better service at a lowest cost." It's able to be much more predictive and much more active than what we had in the past with the configuration that we had before the DSN concept.


Laurent Lefouet:

In a summary, if before you had AI primarily focused on planning and predicting what's going to happen in the case of shock, physically it's about really sitting on top of the signals and enabling an AI to augment the responsiveness of the entire organization by reacting to that event very quickly, being connected to data and being connected to people.


Rafael Calderon:

Exactly. It's basically low on replacing the physical effort, the manual effort with computers to really help us put in automatic a lot of things that in the past we would have to just manage by sheer power, by sheer force in terms of putting a number of people to monitor things. It's creating a tremendous capability in terms of the ability to sense and respond in a lot faster and more effective ways than we had in the past.


Laurent Lefouet:

Any example you could share with us where you actually did implement something similar?


Rafael Calderon:

Yeah and I'll pass it over to Kevin to talk about a real example, right? To hopefully bring this to life for everybody in a real tangible way so Kevin.


Kevin Overdulve:

Yes. Thanks Rafa. And thanks Laurent and the Aera team for inviting us for this session. We talked in the previous pages a little bit about agility and responsiveness and how do you actually take real-time information from execution and use that in order to drive decisions. The example that I would like to present to you today is a case that we developed together with Aera just recently for a large beverage company. The key issues of this beverage company were that on the one hand they saw that their costs of sales in terms of logistics was fairly high. On the other hand they saw they had a fairly high loss rate so they were basically missing out on sales because of a supply chain performance which was not up to par. One of the key reasons was that they were not able to leverage the data they had sitting across a multitude of different systems in a smart way to drive day-to-day decision-making.

Obviously, if you're aware of the concept of a control tower this will start to ring a bell but control tower can be defined in a number of different ways. And yes, a control tower in a traditional definition is a capability that would continuously pull in information from execution, compare that versus basically what you have planned for and start to notify, give alerts on certain exceptions, deviations from plan taking place. And that's a basis that's where you need to start. For this company we took it one step further. Obviously with those notifications, with those anomalies in the supply chain you want to use that information in order to drive a certain decision almost on an ongoing basis. Next to that typical control tower functionality, we developed what we would call a cognitive skill in order to automate a certain decision-making process that people today in the organization are doing on a day-to-day basis but where they are typically lagging behind and running behind effects. On the left side of the page you actually see those different levels of what we would call a modern control tower.


Obviously you start with your data layer. You have certain descriptive and reactive components to allow you to react immediately to anything happening in that supply chain which you haven't planned for. And that could be on the hourly basis or could be on a half hour or even a minute basis. And then as you go up into the maturity curve, you enter into the space of diagnostics, predictive, prescriptive. All of those components that typically we run through if we are about to make a certain decision which can also be automated of course. And as you grow and adopt more of this cognitive automation into the processes and decision-making on a day-to-day basis, at a certain point in time you will move much more towards a control tower, a modern control tower which has the capability to self-heal or self synchronize the supply chain on a day-to-day basis. To give you some examples of the functionalities that we developed, on the bottom and the traditional control tower view we developed a capability actually infused by the entire COVID wave which we called a rapid response control tower and we'll go into a little bit more detail later on.


But basically the idea was in a matter of three to four weeks we developed a capability bringing together information across manufacturing execution systems across a dozen of different WMS systems, a DMS system, ERP, SAP APO. Basically bringing all of that information together to allow people to have sort of an iPhone notification type of application but then are targeted towards their supply chain with all of the details behind it of course. From a cognitive skill perspective we focused on their out of stock problem. Whenever an out of stock risk has been identified by the solution, the cognitive skill starts running through a series of different actions that can be taken in order to avoid that stock-out or in order to mitigate the negative impacts of that stock-out.


And you see there on the page that we looked at two potential actions: either inventory deployments and can I move product that's sitting in another warehouse to the warehouse that is basically destined to fulfill the order? Or production scheduling. Can I do an update of my production schedule? And with that actually avoid the stock-outs update space. What we also developed was a decision board. The decision board is basically a monitoring tool which looks at how well are the recommendations provided by the solution adopted. And to give you one simple example, the inventory deployments skill after just two sprints of three weeks and the recommendations coming out of that were accepted for more than 70%. Basically 70% of the work that is currently being done by a user after six weeks could be fully automated because everybody was agreeing upon those recommendations.


Laurent Lefouet:

It looks like things went pretty fast because I can see this is the work done in the first 100 days. You mentioned the two sprints of three weeks.


Kevin Overdulve:

Yes.


Laurent Lefouet:

First question is what were some of the learnings that you took along the way in implementing this tower?


Kevin Overdulve:

It indeed went pretty fast. Actually the 100 days that you see on the screen involves a discovery phase, a design phase together with the users then three sprints of development. And then even at the end of it another three weeks where we handed the solution over to the business and they actually started using it as if it was real life, thereby confirming the benefits coming out of the solution in this cognitive automation. I think what's important in order to run this in 100 days is to on the one hand be laser focused on a certain scope element and obviously needs to be something that adds value and can be done and implemented even with a certain amount of data available and the accuracy of the data. But continuing to stay laser focused on what really adds value, what is really differentiating versus what they have and we don't want to recreate certain reports people already have, is I would say one key learning thing.


I think the second key learning which is important, kind of across the board if you look at cognitive automation, is to get people alone and educate people to see how this concept kind of revolutionizes the way how work is being done. To reference point earlier on, it's not always easy to kind of get people away from, "This is the way... How I do it right now. This is my territory. I don't want to give that up." But cognitive automation really has the potential to kind of embed certain digital agents into your company that really take over all of that work that humans shouldn't be doing but are doing way too much today.


Laurent Lefouet:

And in this specific example how the organization responded to this new way of working.


Kevin Overdulve:

Actually from the start we had discussions on how to position cognitive automation as a capability in the company. The response eventually was very positive in the sense that even before we were starting to put a design on paper, people were already buying into that vision and into the way this capability can be positioned as a complimentary thing next to I would say their planning systems and all of their transactional automation systems. That was a very important change component right from the start.


Laurent Lefouet:

You mentioned that you had a more detailed example of the thing that you've built as part of the tower.


Kevin Overdulve:

Yes. Exactly. The first one that I would like to deep dive on is our rapid response control tower. You might think this is an ordinary dashboard but that's definitely not true in the sense that you obviously see here information which could also appear in a dashboard looking back two or three weeks after the fact. What do you see here? Is all information that's being uploaded on a 10 minute basis.


We focus on five adherence KPIs is how we call it from the left to the right. Am I getting in what I was expecting to get in order to steer my feeling line? Am I feeling what I was expected to feel? Am I shipping out what I was expected to ship out or what I plan to ship out? How are my sales consuming the forecast that I had and how am I doing in terms of stock position across my network? You also see... It's even a little bit but you see orange and red color coding which is basically to what extent this KPI, this adherence KPI is meeting or not meeting the thresholds they had set forth. This is for example the manager's view. Basically a manager could log in at every time of the day and basically track how well these plans versus the execution KPIs were performing in the organization.


At one level lower we had transactional exception reports where the planners were actually getting notifications whenever a shipment was not arriving, whenever the production line was down and hence certain orders could not be fulfilled. Whenever a retailer dumped a massive order on them at the end of the month and they weren't expecting that or they weren't able to accommodate that. So that's the level of lower this is the middle level. We also had one executive level on top of that which was basically giving a retrospect of the different supply chain issues occurring in that supply network of this client and how they were evolving on a day-to-day basis. The executives could basically focus on one or two issues that were pending for five, six, seven days and take the right action in order to make sure that the next day wasn't occurring on their control tower.


And obviously you need to have the technology in order to be able to continue to load this execution information and continuously compare that with the plan. That's of course where the power of the data crawlers functionality of Aera comes in handy. This is visibility not visibility traditional reporting style but visibility as if you had an iPhone notification which was currently or continuously giving you updates and things going wrong in your supply chain. If we then enter into the next layer we start talking about that stock-out skill. That stock-out skill is linked to this control tower. As the control tower gives you information on how well our stock position is versus our plan, it also gives you all the details of that stock position at skew level, at skew location level even. Based on this view, actually Aera started to populate an entire list of potential stock-out risks by... As I mentioned before, combining all of that execution and planning information into one source of truth.


I'm now stepping into the shoes of a planner at this company. In the morning that planner would log into Aera and would open up their preventive stock-out skill and they have a list of 97 pending recommendations, 260 more are completed already. They can filter or they can sort their lists of recommendations from high to low and total benefits or the margin contribution at risk. And they could select for example, the third one to have a deeper look into what does this recommendation actually entail? When clicking on this, this green below opens and Aera actually immediately gives you the recommendation. In this case Aera recommends transferring a certain amount of pallets of a certain skew from warehouse A to warehouse B because at warehouse B there's... Or that skews the risk of a stock-out in two days from now. You also see here all the details of this recommendation and you see the planning book here and you basically see the reason why Aera has recommended to do a stock transfer order for the skew from location A to location B.


Laurent Lefouet:

[crosstalk 00:28:38] Just giving one question is in that example, the stock-outs it's a prediction of stock-out, it's not an actual stock-out, correct?


Kevin Overdulve:

Exactly. We're continuously... We're basically looking at an evolving picture of stock-out risks over the next seven to 14 days. And as for execution I'll always deviate from that plan. You see that there are quite some line stoppages or sales are not under control versus forecasts. Certain in-bounds of raw materials are not there. They're basically taking all of that information into account to predict at skew location level when you have a risk of a stock-out. Even more, we have developed the machine learning model which does this stock-out risk identification. It takes into account 32 different features, elements that have an influence on a stock-out risk and also takes into account external information like weather information, in order to basically build an engine that is better able to predict whether a stock-outs actually has a higher risk of taking place or not. And the good thing is that obviously compared to deterministic rules this machine learning model gears you towards those skews that will really be in stock-outs and not the ones that you think will be in stock-out based on a stem for business rule.


Laurent Lefouet:

That is obviously a lot of capabilities that you build in there and probably not enough time in this session to go through all of them. What would be your... Let's say conclusion and recommendation for your customers for the people that... Who are watching a decision?


Kevin Overdulve:

My initial recommendation would be... And something that I hear a lot from clients was also mentioned earlier on. We don't have the right master data, we don't have the right accuracy of data. I think traditionally even companies who struggle with their data would already benefit a lot from just having all of that information connected to each other in a virtual DNA with a machine basically picking out the elements, pointing them towards the elements that they need to look at and they need to make decisions on.


And I think all of this can actually be realized in a fairly short amount of time. I would always advise a client to immediately start thinking about what is my biggest issue and what is the typical decision making process around that issue and how do I move and make a first move or step towards cognitive automation. And then think through what kind of visibility do I need? What kind of data do I need? Develop such a use case, bring it into production. Actually, sometimes even scale that geographically or to other functions and then start tackling a new use case but always from the issue back. I've seen too many companies starting on a journey where they decided, "Let's boil the ocean right from the start." And then you get stuck at a certain point in time and you're not able to deliver at the speed and value that you want to deliver.


Laurent Lefouet:

[crosstalk 00:32:07] business problem.


Rafael Calderon:

One thing to add to what Kevin said is not all data is created equal. When we say statements like, "I have bad data." We all have that data. Now having bad data it's like having lower back pain. Everybody has some type of lower back pain. The question is there's elements of data that are actually very useful to create this MBP. Take a little piece of data as Kevin is suggesting, create that demonstration of the use case and then use that to garner the support and get people excited around the story and the narrative to then get the support that you need to go make the additional investment. An important aspect of DSN is to move quickly, right? Is to move quickly and I think Kevin that's kind of what you guys are experiencing in your particular project.


Kevin Overdulve:

Exactly. The best situation you can basically be in is that you indeed pick a pilot, you operationalize your pilot solution and you kind of get that ROI of six to nine months max and you start using the benefits you're reaping from that capability in order to fund the rest of your journey. But it's a continuous journey. As you've implemented two or three or four different use cases, you'll probably see things that you couldn't have imagined before. Things that with the knowledge you have today you're just not able to see issues that you don't know about yet. That journey is also kind of an evolving thing.


All right. And then just closing off, I think when we look at these digital pilots and especially if we talk about implementing cognitive automation, we'd like to use these three steps in order to again, keep that laser focus. It's important to start thinking big. Why? As I mentioned before, one of the success criteria of the projects that we've done together with Aera is to position cognitive automation very well. There needs to be some kind of a dream and a longer term vision first and everybody needs to buy into that vision and then you can actually start delivering upon that vision. Start small around choosing what pilot that is on the one hand valuable and so your lower back pain as Rafael mentioned it. And on the other hand is also feasible with the amount of information and the systems that you have and you would be surprised how much is actually feasible in those small pilot setups.


But then of course with the pilot you won't be heading towards the speed to value that you like to head towards. The last phase scaling fast is hyper important. We typically speak about a couple of different ways how you can start scaling and you can either go very broad and start adding use cases or you can go very far and scale things geographically or for example in logistics across the number of modes of transport. Or you can go deep and further invest in taking advantage of machine learning and adding advanced analytics because that has kind of a ripple effect and a bigger effect on your short term ROI.


Rafael Calderon:

One thing to add to the point on scaling fast is what we've seen with our clients is the importance of changing the way that we measure projects. Incorporate setups. Typically we'd take a project, we measure its ROI and we try to create a business case around it. In the world of DSN we have to think about almost having the private equity mentality. There's a number of things that we're trying and we're measuring the ROI of the whole portfolio. Some things are going to fail. Some things are going to work. What's the overall return of that total portfolio? Just taking that step and thinking about [it] like that just opens the aperture and provides the right flexibility to be able to have the right level of incentive for your executives to want to make those bets and try those things. But if you measure everything on a single project basis, it's going to be very difficult for people to take a chance and that's part of what we see. Sometimes companies struggle, you have to change the mindset, take a PEA mentality and really try to think about this as a portfolio of projects in a total ROI.


Laurent Lefouet:

Very clear. Amazing journey so far. You've been with us working together since the beginning. Very thrilled to see how this is unfolding and actually becoming real with multiple customers that we have in process today. Thank you for joining us in this first Cognitive Automation Summit. This year it's online. I'm sure that one day it will be on site and I'm sure that there will be a lot of people coming your way on stage to ask questions, raise their hand and so on. I'm looking forward to this moment. Meanwhile they can of course reach out to you, ask questions and I'm looking forward to our next conversation.


Get social with us!

Follow us on LinkedIn: Cognitive Automation Community
Follow us on Twitter: CognitiveAutomation

Article
Pascal Bornet
Intelligent Automation Vs. Artificial Intelligence: What's The Difference And Why It Matters
Article
Lalitha Sundaramurthy
How Cognitive Automation Gives CIOs Time To Innovate
Article
Taylor Wills
5 Ways Cognitive Automation Can Future-Proof Your Enterprise
SUBMIT AN ARTICLE
SUBSCRIBE TO WEEKLY EMAIL
-->