In these uncertain times, consumer packaged goods companies should take advantage of their data, and apply AI and data science, to support agile supply chains.
As the pandemic rages around the world, the only certainty is uncertainty. Supply and demand of consumer packaged goods (CPG) may change at short notice, then change again. It may ultimately settle into a new normal, or revert to the way things were.
CPG companies need to adapt to these changed and changing circumstances by becoming more agile in their supply chain management. To understand changing demand trends, and ensure their supply chains are ready to meet them, they need new processes to allow rapid data-driven decisions. These must be designed around the current highly volatile situation and be resilient enough to evolve as situations change.
Matching supply and demand
When considering supply chain changes, we need to look at supply and demand. Much has been written about the possible drivers of each, and we won’t repeat those here. But it is worth considering what makes these different from normal times.
- Demand: Models need to look at how consumer and business habits are different. E.g demand for certain food staples have risen to a point it is hard to meet; whilst demand from the hospitality industry has dried up. In between these, many products will be in constant demand flux, depending on changing rules, earning capacity, and consumer buying trends. In the absence of normal social structures, people will change behaviours more rapidly and unpredictably.
- Supply: Certain products or commodities will become more or less available as the severity of lockdowns change. This is likely to be seen in ripples that move back and forth across the world as different countries ease and tighten lockdowns, and outbreaks pop up and subside.
Using data intelligently for agile, resilient supply chain management
It is impossible to know what supply and demand will look like in a year, so large investments are risky unless well planned. It is likely that companies will need to move away from just-in-time, and build in much greater resilience, allowing them to absorb inevitable shocks and bottlenecks.
Organisations should make better use of data to build in this resilience and become more agile in their supply chain management. How can they do that?
Start by using data to simulate different demand scenarios. The first thing to know is that existing demand models, of who buys what and when, no longer hold. Models are not a perfect reflection of every possible consumer decision, that would be impossibly complex. They rely on a set of assumptions based on consistent patterns of observed real-world behaviour – such as when people get colds, what people do on sunny days, or what demographics buy luxury goods. Even in normal times, they are regularly tweaked to reflect changes in these assumptions.
Right now, all these assumptions have gone out the window in one fell swoop. Feeding new data in will produce poor results, since the model is no longer predicting the environment before us.
Organisations need to utilise the new data as it is generated. There is much less of this data to work with, and not enough to completely retrain models. Dealing with this situation requires innovative thinking from experts who understand the gamut of analytics and data science techniques available, as well as around how to curate the data itself.
Covid related challenges are new, but looking at examples from different areas can shed insight into how to approach problems with very limited data.
One of our clients used a mechanical centrifuge to separate compounds from large mixtures. This took several days to complete, so any batch failure created considerable wasted time, labour and materials. They wanted to understand how to avoid bad batches, but only had very limited data: 19 examples of good batches and two of bad batches.
The strategy selected was to try many different techniques and approaches and identify those that appear to separate bad batches from good. These gave short term insight and could be further validated later when more data was available. The best approach used dimensionality reduction on the time series and clustering algorithms. Having identified the key drivers of batch failure events, Tessella developed a generalised predictive model that also considers any reconfigurations of the centrifuge during reassembly. It also accurately models how the manufacturing process evolves over time due to equipment maintenance, making what is otherwise a point solution, the starting point for a full, end-to-end predictive maintenance regime.
This may seem a long way from consumer demand models, but the approach and statistical techniques have broad application for gathering valuable insights form very little data, whilst setting up systems that can improve as more is known.
Supply chain management
Once we have a good set of demand scenarios, we need to match supply. This is likely to involve changing product lines, adding flexible capacity able to switch between multiple lines, or even building new facilities. Rapid repurposing of production lines is a big investment involving transfer of equipment, technology, and data as well as new governance, SOPs, and QA/QC.
To optimise this process, we can model new approaches in-silico. This allows us to look across multiple facilities and understand where the opportunities are to make changes, and what the costs would be of doing so. We can also run interconnected simulations and see how one change impacts supply chain resilience as a whole. This supports informed decisions and de-risks investment.
Tessella helped the UK government do exactly this type of in-silico modelling to enable critical decisions for infrastructure investment, through the development of a digital twin of UK infrastructure.
Such models need to incorporate information on where weak links are, such as overreliance on single sources of supply, which could be damaging if that region suddenly goes into a strict lockdown. They must address questions such as: what would be the impact of temporarily losing a supplier (tier 1 vs tier 2 vs tier 3), and what is the time frame for recovery? What is the cost of increasing overall capacity vs stockpiling goods? Which facilities can switch product lines, how long would this take, and what would the knock-on effects be for the rest of the supply chain?
An agile approach to data for volatile times
In both demand and supply modelling, there is a need to move quickly whilst taking informed decisions. It is important to take rapid, considered steps forward. Being truly data driven means the next phase of work is always informed by the previous results.
To make the best use of data and develop the most responsive models in volatile times, we recommend three principles.
- Work with people who understand volatility and uncertainty: The temptation to make snap judgements in responding to the sudden shocks will be huge. But leaders need to be sure they only act upon causal relationships in the data, not the accidental correlations.
Business-as-usual experts are not necessarily the best people for this highly unusual situation. The deeper your expertise in a specific model and dataset, the harder it can be to step out of it. Bringing in data experts who are familiar with modelling novel situations, and responding to sudden shocks, can help apply the right thinking to your challenges. This may include people with experience outside your industry, such as those who have delivered crisis planning models in the financial crisis or for infectious diseases.
- Move fast: There is no time for a 12 month study or to deploy decision-automation software. You need people who can look at new data and come up with reliable and trustworthy answers quickly, whilst still being thorough. You need human experts who know what they are doing and can quickly find useful insights and report them directly to decision makers. Sometimes this may be in timescales of less than a day.
- Harness a broad range of data skills with proven real-world experience: Quick insights need a range of skills, which go outside normal market or sector thinking. Presenting a problem and a dataset to a team with a mix of statistical analysis, modelling, AI and ML, as well as domain understanding and real-world experience of applying those skills, will dramatically improve the quality of the answers and the time it takes to get them.
Tessella supports supply chain decision makers through a service delivery model that allows fast on demand access to a broad range of data experts. We have proven frameworks for dealing with uncertainty and volatility that have been developed over 40 years of real world experience. Contact us to learn more.