COVID-19 sent a clear message to companies: supply chain disruption is a real and constant threat. Chip shortages made carmakers pause assembly lines. Scarce ingredients slowed production of sanitizers. And rising demand for key drugs and vaccines put pharma businesses under real pressure.

The good news: artificial intelligence can help. We’ve applied two novel AI methods to address supply chain disruptions like these.

AI in the mix

First, we’ve used what are called counterfactual explanations to help clients quickly update existing products. This identifies specific changes companies can make to reduce or avoid potential disruptions. For example, if a key ingredient for a chocolate bar is suddenly going to be unavailable in large quantities, counterfactual explanations can help find recipes that use less of the problematic ingredient, but still create a chocolate bar that meets the company’s criteria for taste, cost, and so on.

So how does it work? You may remember how we’ve used counterfactual explanations in the past. Simply put, we look at the effect that making small tweaks or changes to various inputs to a model have on its output. In our previous blog, we showed how this could be used to explain why an applicant for a loan was denied, and also show them what changes they would need to make to be approved. Well, counterfactuals are versatile – they work for product development too!

In this case, we look at the effect that small tweaks or changes to ingredients would have on predictions returned by a machine learning model about the recipe’s results. The model assesses a specific key performance indicator defined by the product manufacturer. This could be something like how much many calories are in a chocolate bar, for example.

This approach reveals two things. First, it uncovers what matters most in the model in terms of ingredients. And second, what changes can be made to an existing product, while taking factors such as cost into account. This lets companies “test” a wide range of possible recipes and uncover likely candidates before moving on to physical tests in a lab.

In contrast, traditional ways to find alternatives from a near-limitless range of options can often miss the best possible substitute. Some ingredients may be more expensive in one region than another. Alternatives might have undesirable properties or effects on the resulting product. Or there simply may not be enough of them. And because finding substitutions manually requires people with domain expertise, who are already in high demand, the whole process is going to be much slower.  

More fundamental change

The second approach we’ve applied is called deep generative networks. These come into play when a more fundamental change to a product or formulation is needed. We’ve used deep generative networks for the discovery of new drug molecules, for example.

In the context of supply chain disruption, we’ve used these networks to explore product formulations and suggest component substitutions. The AI uses data from current and historical formulations to find brand new ones. The results? New products at potentially lower cost with higher performance.

And the same technique could be used across countless industries. For example, they could assist in the drive toward more sustainable products by finding alternatives to components that don’t meet sustainability standards.

A solution for now and next

These two approaches give companies a way to use AI to mitigate risk and ensure continuity through global supply chain disruptions. And of course, their use isn’t limited to times of disruption.

We could soon be using generative networks to inspire the design of novel products for completely new markets. Or we could use counterfactuals to improve the performance of the equipment that underpins supply chains. Picture a model that identifies the minimum changes needed in operating conditions to keep a piece of manufacturing equipment at peak performance. Or one that finds the changes needed to keep a delivery truck on the road for more time without maintenance. AI is helping address today’s problems – and uncovering tomorrow’s possibilities.

For more information on Counterfactual Explanations at Accenture Labs, contact Luca Costabello. To learn more about Deep Generative Networks for novel product formulations, reach out to Jer Hayes.

Luca Costabello

Research Scientist


Jeremiah Hayes

Computational Creativity Research Lead – Accenture Labs, Dublin

Recent posts

Subscription Center
Subscribe to Accenture's Technology Innovation Blog Subscribe to Accenture's Technology Innovation Blog