The point of a digital twin is to analyze business questions – like improving node-to-node movements across your supply network, or driving policies that improve the customer experience – and solving all the problems that present themselves across your supply chain. In this paper, we explain why your digital twin needs to be a sandbox extension of your supply network’s planning & operations platform – one that applies different statistical, machine learning, and AI algorithms along with various workflow options to solve problems.

If your planning/operations platform and the digital twin are one and the same system, you’ll be able to evaluate choices based on the top algorithms in the market, and then make those choices actionable in real time. They can be strategic in nature, based on network design options that will benefit you 2 years from now, or can fix a problem that is expected to occur in the next few hours. Too good to be true? We’ll explain how.

By Joe Bellini, COO One Network Enterprises

My experience with digital twins goes back to my engineering days at both GE and GM. Whether it was jet engines or my more complex nuclear missile guidance system, I needed to create digital twins to test variables around heat transfer along with validating that the manufactured items actually met the complex design geometries. In this sense my digital twins were digital replicas of manufactured components to be used for analysis and prediction. We also applied prediction within our digital twin representations for the six-axis robots we utilized to manufacture the guidance systems in order to schedule maintenance prior to losing tolerance when manufacturing our gimbals.

In this sense my digital twin was related to a process and the predictive models designed to improve that process.

Digital Twins in Supply Chain Planning, Operations and Execution

At One Network Enterprises we have now applied similar digital twin techniques to supply chain network planning, operations, and execution.

Today, digital twin and digitization in general have become what I call “duffel bag terms”. You can define it many different ways, and it all loosely fits in the duffel bag. A similar situation exists around AI as a duffel bag term.

The reason I created my original digital twins was to generate a useful result that I couldn’t get with the actual units unless I was willing to shoot them into space or sometimes even blow them up under stress.

The benefits of these digital twins were twofold: first finding design issues that would lead to failures so I could correct the designs before any failures occurred; and second, evaluating design changes related to troublesome variables/ measures or for actual failures themselves. Much of my work was done with CAD/CAM/CAE/CIM along with heat transfer differential analysis.

“There is nothing better than having a digital twin capability that is an extension of your operating platform. You can evaluate choices based on the top algorithms in the market and then make those choices actionable in real time.”

Digital Twins Solve Supply Chain Network Problems

I have found that extending digital twin problem-solving techniques to a supply chain network requires that we first must model the entire end-to-end supply chain network to form the foundation for our analysis.

Since the opportunities or problems that will be exposed by our analysis could manifest themselves over strategic, tactical, or operational time-frames, the foundation should be seamless across time horizons. It should offer services, algorithms, and analysis that run across the network representation in real time, whether we are solving problems predicted to happen in six months, or during a delivery scheduled for later this afternoon.

The implication here is that the correct foundation for a digital twin is an execution platform whose item level modeling and representation can scale in detail from execution all the way forward into longer term sales and operations planning. This digital twin is not a simulation.

A Digital Twin is Not a Separate System

Planning software vendors often try to provide digital twin capabilities by using simulation techniques, or even try to import operating data from the ERP system using some kind of rapid response, but this method falls far short in its ability to generate high value results. If a vendor starts talking about the right level of resolution based on differing forecast horizons, that means the foundation/platform cannot scale properly to solve the problem. Some vendors even talk about using approximations for longer term time horizons which is nothing but an average and will return average results.

The real problem is that those architectures are antiquated and not really designed for digital twin type problem solving. Their problem-solving algorithms, if modeled in detail, would run seemingly forever due to their architectural limitations. The in-memory approach we took back in the early 90’s has been surpassed in performance, scalability, and reliability by newer real-time supply network platforms.

An extension of these limitations leads some folks to talk about “concurrent planning”. The fact that there is any demand in the market for concurrent planning means that somehow users have been convinced they need two different software representations or applications for operational and tactical planning and that the two should be interoperable. This is ridiculous.

That’s because modern platforms can certainly scale detail from operational to tactical to strategic. Changes in time, or temporal changes, are nothing but state changes related to something like an order. The order may start life 2 years out as part of Sales & Operations Planning (S&OP), and then change state to a forecast order at 12 months, then a planned order at 6 months, then a committed order at 1 month, then a shipped order at 1 week, then an in-transit order at 4 hours, and then a received order at time zero, then an authorized-to-pay invoice order at time minus 1 day.

On a real-time supply chain network, these are just state changes across a seamless platform where all levels of detail are available in all time horizons. When planning, operating, and executing in this type of platform environment, the entire network becomes more resilient, more responsive, and the likelihood of the plan being executed without major problems is much higher.

Your Supply Chain Network Platform Should Be Your Digital Twin Environment

An end-to-end real-time supply network platform/foundation enables the ability to test out new supply chain policies, network resiliency, the feasibility of strategic or tactical plans, activate alternate parts or suppliers, modify modes of transportation, or even add additional shifts at a plant. In this sense it is the platform itself that also serves to enable the digital twin.

You don’t need to configure and support a separate reference model. Your digital twin is a sandbox extension of your supply network platform that applies different statistical, machine learning, and AI algorithms along with various workflow options to solve different types of problems. The planning & operations platform and the digital twin are one and the same system.

As we have discussed, problem-solving analytics are embedded as part of the network platform, thus obviating the need to add an additional, separate supply chain analytics platform. For example, in the health care sector One Network is providing a resilient platform for the healthcare supply chain, from hospitals upstream through distributors and into suppliers & manufacturers. The platform is designed to predict potential problems related to resiliency to activate alternative plans or actions during times of disruption or to react to inventory disruptions in real time as they occur.

The potential for improvement in healthcare effectiveness and efficiency is tremendous when we consider all the touch points including hospitals, clinics, 3rd party providers, distributors, pharma manufacturers, CMS, HHS and the FDA. The number of shipments, cross docks, forward stocking locations, and distribution centers across multiple modes of transportation originating from hundreds of countries certainly provides an opportunity for tactical and operational improvement.

Predicting, Prescribing and Problem-Solving Supply Chain Issues with Digital Twins

As I mentioned with my GE experience, the point of a digital twin is to analyze business questions – like improving node-to-node movement across the network, or driving policies that improve the customer experience, to implement strategies and tactics that deliver the highest quality product at the lowest possible cost. It should solve all the problems that present themselves across the network, given all the variables in play.

Take note of that last point…all the variables in play. The point of an end-to-end representation of the network is to have real-time visibility and control over all material variables across all trading partners at all tiers and echelons in the network. Hot zones are going to materialize across the network indicating there is an actual or potential problem in meeting targets related to demand, supply, logistics or fulfillment. Given that our digital twin and our operating platform are the same system in a network solution, the real function of the digital twin is to move these hot zones into a sandbox so that the digital twin can run analytics to make recommendations to solve the problem in the most optimal way – and then make those choices actionable as an extension of the execution system.

So now comes the matter of choice. In a network there are many ways to solve problems related to demand, supply, logistics, and fulfillment. The digital twin workbench in the network case has real-time access to every material variable in the network. Theoretically, there really is almost no limit as to how many ways you can solve a problem with this kind of access.

Traditional supply chain management systems typically only give you one way to solve a problem, due to static lead times and stale data – for example, spend more on inventory, capacity, suppliers, or even partial ship and back order.

Your key question when considering digital twin technology will be how many variables are considered when using the digital twin to solve a problem. Most vendors will have very little to say here, or blame it on the difficulty of accessing and cleansing the data.

In advanced AI/ML agent-based networks, your strategies, policies, tactics, customer service levels, revenue objectives, margin objectives and more are all taken into account when determining the best ways to take advantage of network opportunities or solve for problems. Your digital twin sandbox is a prescriptive environment where you will be presented with the top three or four solutions that best meet your targets. You are then free to choose the best one that meets your needs at that time, fully understanding the effect your choice will have across all customers and trading partners in the network.

These AI-based workbenches are available today across demand, supply, logistics, and fulfillment. These are true digital twins with a control tower-based view across the network that can hone in on opportunities and problems, including long term network design strategies, and populate a workbench in a sandbox environment that empowers the user to determine best choice. Then you can make that choice actionable across the network, given that your digital twin is an extension of your operating platform and not a separate system. You no longer need to worry about a baseline shift when running your digital twin analytics, given your analysis is based on single version of the truth in real time.

Within the workbench itself, you can evaluate a number of prescriptive options. For example, in the logistics workbench you could analyze whether a less expensive mode of transportation would also require building inventory in order to maintain targeted service levels. You could look at whether more frequent deliveries would lead to higher transportation costs but be offset by having to maintain less inventory. You could also analyze the network itself in terms of whether adding or subtracting a cross dock node would lower the total landed costs.

These types of nodal shifts really display the power of the network, given they provide leverage and can benefit multiple trading partners across the network.

Activate Your Digital Twin – Automatically

The best part of the approach outlined here is that there is no “big journey” to activating your digital twin and begin driving all the benefits and value. With a real-time network architecture your digital twin, along with all the required algorithms, agents, strategies, tactics, and policies is available as part of a sandbox extension and ready for action. And as part of activating all the trading partners across the network you get what I call a “clean” environment. It’s an environment where all the second guessing and demand manipulation node-to-node upstream in the supply chain (creating the “bullwhip” effect and an unnecessary inventory build across all trading partners) has already been resolved.

But what about the data? The great part of a network architecture is that it exists as part of what we call a “dual platform approach”. Here the network instance sits on top of all the existing legacy systems and deploys a Federated Master Data Management (MDM) capability along with hundreds of pre-built API’s in order to populate the models. Typically, the data isn’t clean, comes from hundreds of deployed ERP instances in larger companies, and also includes data from external sources which are key to populating certain algorithms and creating AI/ML vectors as part of the prescriptive analytics. These issues become non-issues when using One Network’s Value First methodology.

Thus, as part of an overall supply chain control tower approach, the model is populated, for example, with transportation costs at a granular stock keeping unit (SKU)/mode/origin level to destination level for shipments along with visibility to specific SKUs that are in a shipping container based on having order, inventory, and logistics as part of the same real-time instance. It then becomes simple to analyze something like landed costs at a granular level as part of a trade-off analysis. This data would be very expensive to gather and maintain in a separate digital twin environment due to all the silos, yet with the One Network dual platform approach it becomes a simple matter of populating the model.

Vital Decision-Making Capabilities

Considering the impacts of the recent global pandemic, the ability to provide global visibility across the network considering both global available-to-promise (GATP) along with global decision-support management becomes a key factor when reacting to hot spots driven by daily and weekly disruptions in supply. The need to provide life-saving products and equipment has created a huge sense of urgency in the face of unprecedented demand surges, supply disruption related to air capacity for shipments, or even customs shutdown in certain countries.

Your digital twin can be used for strategic and tactical planning in addition to hot spot problem resolution. When AI agents in the network report that a trend could lead to a demand or supply impact weeks or even months out, the time to react is now, in real time. Actions can be taken around alternate supply, substitute parts, inventory pull forward, outbound postponement, or even allocations.

Another key capability is the ability to speed up or slow down a network. Given that One Network is deployed across most industry segments on a global basis, in some sectors demand has more than doubled during the pandemic while in others demand has dropped by 50% or more. In all cases, it has been a simple case of “turning a few dials”, activating choices through the digital twin workbench, running autonomous agents for decision-making where the performance guardrails are clear, and moving forward.

For example, certain customers ran a total landed cost scenario in order to meet potential surges in demand for certain products, where we predicted warehouse capacity issues based on future demand profiles. The workbench provides choices such as overflow warehousing, manufacturing postponements, manufacturing sourcing, shipment strategies around floating warehouses, and more – enabling faster, more accurate, more optimal decision making.

Digital Twins Are Part and Parcel of a Supply Chain Network Operating Platform

In summary, there is nothing better than having a digital twin capability that is an extension of your operating platform. You can evaluate choices based on the top algorithms in the market and then make those choices actionable in real time.

Those choices can be strategic in nature based on network design options that will benefit you two years from now, or can be to fix a problem that is expected to occur in the next few hours. And if you are so inclined, you can gamify the digital twin decision-making workbenches among your teams using the targeted outcomes as the measuring stick for rewarding performance.