If you operate in the supply chain or the planning world, it is hard to not come across terminology like supply chain planning 4.0, ‘light touch’ planning or ‘lights out’ planning. They all rate high in the hype cycle and for good reasons. A lot of technology progress has been made in recent years, and progress continuous at dizzying speed. Autonomous vehicles already exist in the supply chain, so why should we not strive to achieve autonomous supply chain planning?
Key points in this article:
- describe why we need a new wave of planning technology to support autonomous planning
- address some of the key requirements for this new wave of planning technology
- highlight several elements that still hold us back from fully automated planning
The three waves of supply chain planning
If we look back to the history of supply chain planning, it can be argued that we are in the third wave of integrated supply chain planning software.
Wave 1 – Enterprise Resource Planning
Building on developments that date from the 1960s, the first wave really started in the 1980s with Enterprise Resource Planning (ERP) software. Initially ERP was facilitating and automating transactional businesses processes like inventory control and planning functionality like Materials Requirements Planning (MRP). On a transactional level, ERP has been a great advance for business and there is hardly a large business that can properly function without it. However, with functionality including financial accounting, human resources, sales & distribution, quality management and asset management, to name just a few, planning was never the sole focus of ERP. This changed with the next wave of integrated supply chain planning software.
Wave 2 – Advanced Planning Systems
Advanced Planning Systems (APS) gained momentum about 20 years ago with the goals to facilitate a forward view of the business, integrate plans with other functions, and automate and optimise supply chain measures like forecast accuracy, inventory holdings and customer service. After 20 years, APS are now in the maturity stage of their product lifecycle. This makes them a commodity in a crowded and competitive marketplace, just like the databases and ERP systems that support them.
The common principles of demand, supply, inventory and replenishment planning that APS must address hardly changed over the years. Neither have the master data and planning parameter requirements that drive them. And although APS systems have become visually more appealing, and have added functionality and optimisers, they are still dependent on accurate data input. This remains a challenge, as most global businesses run more than four ERP instances and four to five supply chain planning technologies. They still operate without a one source of truth to support the best possible decision making.
Wave 2 shortcomings
APS often misses advanced end-to-end decision intelligence and many APS vendors don’t cover the full end-to-end supply chain with their technology. Most APS don’t provide the option to design a full digital copy of the underlying supply chain planning processes. Furthermore, APS systems are not able to extract value out of the large amount of data that today’s supply chain creates. The reality is that enterprise reporting and decision making is still tied together by spreadsheets, with over 90% of companies being dependent on them. Planners still spend significant more time on gathering and generating information than on decision making for their schedule. In short, APS are a far cry from supporting ‘lights out’ planning.
These shortcomings won’t be solved by including marginal improvements to existing APS products. It is more likely that a third wave of supply chain planning software is required to solve these issues. This third wave of supply chain planning software will support further digitisation, automation and will use ever increasing intelligence to make, communicate and action business decisions. It will relieve planners from the cumbersome limited value-added tasks like gathering, cleansing, formatting and segmenting data from multiple source systems, in order to focus on actual planning decisions, collaboration and the bigger picture. In doing so, third wave planning technology might indeed turn the lights out in the planning department, and maybe beyond.
Wave 3 – Autonomous Supply Chain Planning Systems
What follows is what I believe are some of the key requirements of third wave supply chain planning software, including some of the challenges still to overcome.
A digital twin for supply chain planning
To fully automate a process it needs to be digitised, so the planning technology knows the planning steps, decision moments and communication channels to share decisions and outcomes. This requires the ability in wave three systems to create a ‘digital twin’ for supply chain planning.
A ‘lights out’ demand plan has to understand to clean history, make a baseline, add some NPD, include promotional volume (based on an automatically optimised price point), check a bit of cannibalisation here and a bit of competitor activity there and do some demand sensing in the short term. You get the idea of what the digital twin has to do!
Similar for a ‘lights out’ supply plan. Some IF…THEN…ELSE must be programmed into a digital supply chain planning process, to understand what to do when in over supply or undersupply, when a machine breaks down or when there are multiple sourcing options, to name just a few possibilities.
These simple examples show that digitising a planning process becomes a big task, with lots of cross functional or cross geographical impacts. A task that can’t be underestimated and a task that will get only bigger once more elements of the E2E value chain will be included, like plants, warehouses, tier 1 and 2 suppliers and customers.
Common planning & analytic data layer
This data layer must solve the information shortcomings of second wave planning systems, especially for large global enterprises, that run on many instances of ERP, APS, TPM, CRM and other systems, and don’t have a common source of truth. The common data layer only uses relevant planning data from legacy & external systems and IoT sources, with minimum latency, maximum detail, security and speed. It needs to have the ability to tap into dozens of data sources and dynamically read from and write to sources if required. The common data layer will support the digitised supply chain planning process.
The current trend of creating data lakes are a good start. However, if the unstructured data in the data lake does not support decisions as required in the digitised supply chain planning process, it is hard to see how it can support full automation. Currently data lakes do not commonly write back information to the source systems, whilst full automation requires read and write capability.
In wave three planning technology, descriptive, predictive and prescriptive analytics must support every decision in the digital supply chain planning process. Descriptive (what happened?) is useful to send automated reports to stakeholders and for technology to learn about the past. Predictive (what will happen?) automates scenario planning across every digital process step, relevant planning parameter and decision point. Prescriptive analysis (How can I make it happen?) will select the best course of action out of the predictive analysis based on a defined business goal and some clever probability analysis.
Automated & dynamic problem solving and decision making
A predictive analysis can easily blow out of proportion in terms of options available. Reasonably straightforward production problems can be NP-complete (the hardest type of problem in computational complexity theory) and not be easily optimised. If we try to automate and optimise the value chain, there are quickly millions, billions or trillions of choices available to decide the best course of action.
Many APS optimisers use linear programming to solve hard coded and therefore static supply planning problems. Wave three planning technology has to support additional smarts like probabilistic Bayes network modelling, decision trees, graph search, game theory and others to decide best courses of action and the most likely result. It will use the digital planning process to apply these algorithms more dynamically, depending on process step, input variables and decision point.
This dynamic algorithmic intelligence will mimic and automate human reasoning, judgement and decision making and lead to RDA (robotic decision automation) or cognitive automation, replacing the planner for these activities. Soon enough in this environment, the reason for any given decision may be far too complex for planners to even understand.
Flexible goal setting
As smart as ‘lights out’ algorithms can be, they still need a goal. What do we optimise in a decision, function, business or value chain? Do I maximise customer service or profit, minimise costs or something else? Goals can change during economic and product lifecycles and don’t always have to be logical. A business logically wants to optimise EBIT, but might decide to incur losses to gain market share.
Wave three technology must provide flexible ways to update business goals into the digital supply chain planning process, in order to guide advanced analytics and automated decision making. Human interaction is likely needed to provide goals to the wave three AI for a long time. Artificial Intelligence without a goal is like an autonomous vehicle without a destination.
Artificial Intelligence without a goal is like an autonomous vehicle without a destination.
Planning decisions need to be automatically executed. Therefore, planned orders need to become production orders, stock transfer orders and many others and must be written back to execution systems. The ‘lights out’ planning software needs to be able to write back these transactions across the E2E value chain. In real time, and in any of the underlying systems it got its information from. Finally, the technology needs to automatically communicate key decisions with expected impacts to stakeholders in any department where the lights are still on.
The automatic conversion of the master production schedule to production orders sounds plausible and straightforward. However, current research suggests that iterative order release mechanisms between scheduling and and production planning often fail to converge. Automated execution will also be a journey of trust between human and machine
Wave three systems will have to document its decisions, the expected outcomes, the actual outcomes and create a feedback loop to learn about what works best. This is where machine learning kicks in. With supervised machine learning, humans are still required to train the algorithm until it has a desired level of accuracy and can run by itself. To be fully ‘lights out’ and continuous self-learning, wave 3 technology needs to be able to train itself and apply reinforced machine learning.
We’ve seen significant progress in reinforced machine learning with famous examples from Google’s AlpgaZeroGo in the games of chess and Go. These are games with a clear goal, boundaries, rules and perfect information. Recent progress includes Pluribus, a poker-bot that learned to defeat world-class poker players in a matter of days. The game of poker has imperfect information, which was no issue for Pluribus to bet and bluff and win against five of the world best players in a six-player, no-limit Texas Hold ’em game.
Playing poker might come closer to supply chain planning, which has to deal with incomplete and imperfect information as input to decision making. However, we still have a while to go before we can let a self-learning machine fully takeover the complexity of a global supply chain.
Once the automated planning decision has been taken and executed, some self-maintenance might be required. Changes to master data, planning parameters, or algorithm settings need to be updated based on the latest available status, prediction or learning. This requires wave 3 technology to automatically update settings; in itself, in wave 1 (ERP) or wave 2 (APS) systems, or any system it uses as a source to run itself.
‘Lights Out’ Planning is coming…over time
To cover all of the functionality described in an integrated platform, on a global scale, the ‘lights out’ planning technology needs to have a flexible architecture, be scalable, be able to two-way interface with hundreds of entities and IoT sources and have to ability to absorb and dissect significant amounts of data. The average supply chain professional will understand that the wave one and wave two systems they are working with, can’t provide the requirements for ‘lights out’ planning.
The good news is that we seem to be on the cusp of the breakthrough of wave three supply chain planning software. Many of the requirements discussed are already available or becoming available soon. In coming years we’ll see acceleration in the availability of these types of functionalities. We’ll see wave three vendors appear on the scene, we’ll see them grow and converge in to E2E offerings that will support a road map towards autonomous supply chain planning.
We are starting to hear the first examples of ‘lights out’ or autonomous planning from early adapters. These are specific cases, with a clear scope, in specific segments of the supply chain. We can again see the analogy with the autonomous car, which is available in specific business and social segments and situations, however where full, hands-off automation (level 5) on mass scale in a busy city center is still some time away. From the early adapters phase, ‘lights out’ planning will have to go through a similar journey to reach level 5 automation at mass scale.
Nevertheless, these are exciting times in the supply chain planning world. Autonomous supply chain planning is coming, but some patience is still required before we can turn all the lights out!
Enjoyed this read? You might like my article on self-driving Integrated Business Planning