Setting the right objective
“Modern supply chains are complex” is a truism. Operators are all too aware of the global forces and local details that drive weird and wonderful supply chain complexity. This complexity isn’t going anywhere, but industry innovators are pioneering approaches to manage it more effectively.
Download ‘Supply chain planning under uncertainty’ slide deck
Planning processes are mature
In a complex environment, diligent planning is required to ensure that supply chains are cost-competitive. The drive to manage cost has spawned an ecosystem of multi-step processes (some more effective than others) and supporting enterprise software. The processes and systems are human driven, and often key knowledge sits with individuals despite the presence of large systems. Which leads to an obvious question…
Can we automate planning?
Fortunately, the answer is often yes. Like other processes that take many datasets and priorities into account, supply chain planning can often be automated or semi-automated. Increasingly, operators are turning to algorithms and artificial intelligence to drive lower costs across multiple segments of the supply chain.
Although many different algorithms can be applied to support decision making, managers can apply a general framework to frame these problems before applying algorithms.
Firstly, define all relevant Decisions. Decisions like:
- “When should I import raw material, and how much should I import?”
- “How much inventory should be stored at each point in my supply chain?”
- “When should I book different transports?”
Operators and planners will easily identify the big decisions they make day to day or week to week, but when applying algorithms we need to consider the little decisions as well.
But decisions aren’t made in isolation – they’re subject to the physical, contractual and practical rules that apply to a business. These can be referred to as Constraints. They might look like capacities associated with road or rail transport legs, restrictions on site storage or throughput capacity or throughput capacity, or even specific timing rules for quarantine and or fumigation for primary products.
Decisions are made in a constrained environment, but this framework relies on an additional element to frame algorithmic approaches. This is the Objective: what result do we want from the algorithmic plan? In a world where we make decisions subject to constraints, we need to know what makes a good decision. Typical objectives often focus on planning to minimise cost, maximise profit or maximise throughput.
We can frame planning problems, then, by defining decisions, constraints and objectives.
Uncertainty is unavoidable
But how do these algorithms behave in a highly uncertain environment? How should they be applied to balance cost reduction with overall supply chain resilience? Uncertainty is unavoidable – this is true for our supply chains, regardless of scale. Uncertainty drives unexpected events, and these events can appear in many different ways. Maybe a key piece of plant breaks down for four hours, or maybe a major customer doubles their order for the next four weeks.
Most importantly, there is a big difference between a cost-optimised plan, and a plan optimised for cost-of-execution. A cost-optimised plan assumes certainty, and perfect, accurate information. Perhaps counterintuitively, highly tuned cost-optimised plans can perform poorly when reacting to change – typically these plans have sacrificed resilience to achieve cost-optimality. Algorithmic plans like these are blind to the costs associated with responding to unexpected events.
What would perfect look like?
In a perfect world, firms would have high-quality, comprehensive data, capturing probabilistic possible outcomes. Mathematical models would scale effectively when solving the largest stochastic problems. In this alternate reality, we could minimise the operating cost subject to hundreds of thousands of possible outcomes. We could use these models to make sure our worst-case outcomes rarely occurred.
This reality isn’t out of the question in decades to come, but for large and complex operations this typically won’t be possible. Even when a planning problem is small enough to approach in this way, often there is poor data (or no data) available to describe probabilities of future events. This data simply isn’t prioritised right now.
This paints a fairly bleak picture of planning to manage uncertainty. Thankfully, there are a number of highly effective algorithmic approaches to manage supply chain uncertainty when planning.
What is the best approach, given the circumstances?
More and more, algorithms using “Resilience Metrics” are proving to be an effective way to handle uncertainty while avoiding the challenges described above.
By identifying or constructing metrics that indicate a plan’s resilience to change, firms can optimise a non-stochastic model while also creating a plan that provides a greater ability to respond well to surprises.
While this approach relies on approximations, it also removes the barriers associated with genuine stochastic optimisation – it doesn’t require huge probabilistic datasets, and doesn’t become too complex to solve quickly and meaningfully.
Resilience metrics can be simple, and may even be intuitively understood and used by planning experts. Some examples of resilience metrics include:
- “Safety stock plus” style measures for products with high demand variability.
- Delivery vehicle route plans with characteristics that allow a second delivery attempt.
- Slack holding capacity or processing capacity across multiple planning horizons.
Poker, not chess
Shipping is a great example of an industry with high levels of uncertainty. Vessel breakdowns and shifting demand for cargos can rapidly shift the profit-optimal plan. Previous approaches to tonnage allocation in the shipping industry have leveraged similar algorithms to those used in vehicle routing, and attempted to create “highly-optimised” plans. These planning algorithms have predictably seen low adoption and fostered a broader cynicism in the industry towards optimisation that ignores uncertainty. To paraphrase one executive at a major shipping line: “We’ve been trying to play chess. We need to play poker”.
What’s next?
As multiple industries that operate large supply chains search for improved resilience, more nuanced algorithmic planning should be leveraged to achieve genuinely cost-optimal outcomes. At Biarri, we hope to continue to play our part in moving this discussion forward.
