Businesses today face unprecedented levels of uncertainty. The pandemic is front of mind, but even when that ends, geopolitics and climate change will continue as sources of volatility that will impact businesses and disrupt their supply chains.
Organisations view uncertainty and disruption with apprehension, worrying they will have to stockpile inventory to avoid stock-outs. But this disruption is also an opportunity. Organisations that account for volatility will be able to react to it more quickly, be more resilient in the future and run leaner operations than competitors.
There’s already a body of best practice for building more resilient supply chains by explicitly accounting for uncertainty. In this blog we’re going to explore how AI can help supply chains deal with uncertainty and become more resilient.
A limited approach to accounting for uncertainty
Most operational teams are already familiar with statistical methods that enhance or automate decision making, like demand forecasting. But some of these methods only pay lip service to volatility. So, operations teams account for uncertainty by manually adding in safety buffers. For instance, the finance department might subtract 10% for one product family because they know demand forecasts tend to be on the high side. Or a supply planner could keep greater amounts of stock ‘just in case’ because demand is so volatile.
These manual corrections (also known as heuristics) work if volatility remains consistent over time. But this approach isn’t built for a post-pandemic world.
To improve the way their supply chains handle uncertainty, organisations need to move beyond manual corrections. The diagram below shows three main ways that operational teams can do this. We’ve explored each of these approaches in more detail in the rest of the blog.
‘Bolted On’ confidence bounds
Most predictive tools will give you a single value. For instance, next month you’ll sell 724 of SKU X, or that assembly line produces 7412 widgets a day.
But as organisations mature, they replace these estimates with technology that tells you a range of possible outcomes. For example, it could say you’re very likely to sell between 690 and 758 boxes of SKU X, or 724 +/- 4% of SKU X. This significantly improves decision making: a procurement team might decide to purchase raw materials based on demand for 758 boxes to ensure there is sufficient availability of raw material, but a production planner might decide to only make 724 boxes, to avoid high inventory carry costs.
But how can organisations start surfacing volatility explicitly in the statistical tools they use to support decision making? Better supply chain management starts with better forecasts, and accounting for uncertainty is no exception.
This is where probabilistic forecasts come in. They provide a natural framework for estimating uncertainty. Instead of emitting a single number, like most traditional forecasting or machine learning methods, they return a range of likely values. Probabilistic forecasts can be used for a range of use cases across the supply chain, such as forecasting demand, likely machine downtime, staff sickness or the need for spare parts.
The advantage of a probabilistic approach to forecasting is that different decision makers can choose their risk appetite while still using the same underlying forecast. Their decisions are now based on a predicted range of outcomes that account for volatility. This means they can ditch the manual corrections and action data-driven insights with confidence.
And accompanying predictions with an accurate quantification of the uncertainty in that prediction builds trust in the prediction itself. Supply chains are plagued by manual overrides, sometimes rational, sometimes less so. Quantifying how uncertain a prediction is helps direct attention towards those predictions most in need of manual intervention.
Get a connected view of uncertainty
To really plan for an uncertain world, bolting confidence bounds onto individual statistical measures isn’t enough. That’s because most of the decisions made by operational teams depend on upstream decisions; they’re all connected. The solution? Connect the different statistical tools that support these decisions.
For example, by taking into account demand volatility for all SKUs that require a specific raw material, you can mitigate the risk of not having enough of it if one SKU has a huge demand spike, all without holding excessive amounts of inventory. And by feeding uncertainty in demand patterns and likely machine downtime into workforce planning, you can be more confident your organisation has enough workers to hit a particular service level.
But to achieve this connected view of uncertainty, the statistical tools used to support decision-making need to account for the uncertainty in their inputs.
In an ecosystem as complex as an organisation’s supply chain, having a transparent, connected view of volatility gives operational teams oversight of all possible outcomes when making a decision. This gives teams the ability to plan for the worst, making the supply chain more resilient.
This requires gradually re-designing the suite of statistical tools available to operational teams to account for volatility. It also requires changing operational processes and educating users to think about uncertainty across the entire supply chain, rather than in silos.
Surface uncertainty in prescriptive tools
Organisations that manage all of the above will be able to understand what’s likely to happen to their supply chain. They will have a clear view of demand and the implication for downstream decisions in a way that explicitly builds in uncertainty at every step: in demand, in machine downtime, in raw materials or workforce availability.
But knowing what is likely to happen isn’t enough. Supply chain leaders also need to know what they can do to change it.
Businesses at the highest level of the uncertainty maturity model have the tools to run simulations over the connected decisions made in the supply chain. They can ask questions like: “if I added an assembly line, what would this do to my product availability?” or “if I reduced raw material lead times from three months to one, how would this affect my inventory carry costs?”.
Being able to run simulations over the entire supply chain mitigates coordination problems that plague operational teams: rather than optimising locally in their silos, teams can make decisions knowing the effect their decision will have on the rest of the organisation.
Prescriptive tools should explicitly show all the possible outcomes of a particular decision, rather than only showing the most likely outcome. This means you can build in volatility before making a decision, rather than having to account for it afterwards. And instead of optimising in silos, teams can make decisions knowing the full impact they will have on the rest of the organisation.
Faculty employed these statistical methods to build the COVID-19 Early Warning System, the operational tool the NHS uses to forecast hospital admissions – or, demand on life-saving equipment. Here, “stock-out” equates to running out of beds or ventilators. Knowing all of the possible outcomes allows planners in NHS trusts to make time-critical decisions backed by data. They don’t have to rely on manual corrections that could put lives at risk.