Tackling massive uncertainty in supply chains with AI

Profile picture for user Gaurav Palta By Gaurav Palta June 10, 2020
Summary:
AI is helping demand planners make sense of volatility during the COVID-19 pandemic. Noodle.ai's Gaurav Palta explains some of the data science being applied to the supply chain

Two workers in warehouse with forklift loader © Marcin Balcerzak - shutterstock
(© Marcin Balcerzak - shutterstock)

Disruption to supply chains as the pandemic swept the globe has led many companies to reevaluate how well-equipped they are to handle system-wide volatility across networks. How is anyone to make sense of demand and supply patterns and manage overall health in the midst of this pandemic — which has introduced a level of uncertainty that current enterprise tools are not designed to process?

Working closely with our customers on a daily basis, we are being asked to help make sense of their demand signals across complex networks and hierarchies. We are also helping them predict and respond to impending supply imbalances within their 0-12 week execution windows, a critical source of value leakage and especially pertinent in current times. Faced with this fast-paced, multi-dimensional chess-game, customers need clear planning recommendations that improve fill rates, reduce inventory, minimize write-offs and control logistics spend.

Noodle.ai was recently identified as one of fifteen startups to emerge stronger out of the current COVID-19 crisis. To understand why, let me share a bit about the issues as we view them and the approach we're taking as a result.

Why are deterministic rules-based systems failing us?

Rules-based systems for allocation and deployment of products have been rendered largely unusable by extreme variability and noise in the current environment. Existing ERP and Supply Chain Planning systems provide a robust repository of basic supply chain data (eg forecasts, orders, inventory, production, shipments and so on) but use simple rules based on standard assumptions to generate the plan. Once reality hits, these plans fail to adjust to true demand and available Inventory. In other words, planning signals become extremely 'noisy' and operators spend the majority of their time deciphering what is real versus what is not. This results in response latency, bad decisions eventually leading to lost sales, high inventory and deployment costs across the network.

What planners and operators need are modern AI-powered systems to create and manage demand and inventory plans in such times. They need the help of sophisticated AI inference engines crunching billions of data points to recommend actions, focusing on those SKUs that are predicted to have the greatest financial impact.

Sensing and predicting demand /supply patterns

In these unprecedented times, it's important for data scientists to build and frequently tune AI inference engines that are powering applications to correctly identify relationships within the data and predict the shape of demand curves — even under the most volatile of conditions. At Noodle.ai, we detected emerging changes in demand patterns in APJ as early as February 2020 and quickly took action to add a greater level of interpretability to our models, sensing the impending crisis before it was widely known. Based on observed behavior from customers across the value chain, we quickly characterized demand behavior along three dimensions:

  • Spikers vs. Decliners
  • Trending vs. Leveling
  • Stable vs. Volatile

Demand patterns detected during COVID-19 by Noodle.ai
(Noodle.ai)

By clustering SKUs on these three dimensions, we better understood how products are behaving, and updated our models continuously to integrate the latest demand profile information. While many supply chain professionals innately understand these types of patterns, human decision-makers will tend to respond with varying degrees of bias, especially when thrust into a crisis. Our data-driven demand profiling is not subject to bias, making it helpful not only through the depths of the crisis, but also through the recovery.

As a result of this crisis, we've also pivoted our AI-enabled prediction models to create more holistic business and consumer indices. This pivot includes the capture and ingestion of significantly more internal and external data and will enable more accurate short-term predictions.

Similarly, managing for supply imbalances also requires a keen eye on the impending impact on production and logistics attainment against the original plan. Combining enterprise data with external data sources — such as IMHE, COVID-19 response measures, mobility and industry specific indices — we were able to establish data science driven scenarios of what shortfalls and constraints our customers could expect as COVID-19 rippled through plants and distribution centers across geographies. Our predictions were clearly different from what rules-based planning systems were programmed to assume. This allowed us to make better recommendations to operators around what issues to target that were the biggest drivers of risk, thus helping keep supply chains healthy during this time of heightened crisis

Finding certainty through human and machine symbiosis

In the age of AI, the symbiosis between AI/machine learning recommendations and human operator knowledge has always been strong. Now, the COVID-19 era requires even greater integration between powerful data science models and the people to interpret them. But this isn't new for us. Noodle.ai was founded in 2016 on the premise that this is the perfect time to leverage the intersection of powerful supercomputers, more plentiful and powerful AI algorithms, cheaper data storage and our deep domain understanding of supply chains and manufacturing.

I'll be presenting Noodle.ai's differentiated approach to AI in supply chains during the upcoming SAP America's User Group's ASUGForward virtual event on June 22nd at 10:35 AM PDT. Until then, learn more about the Noodle.ai Athena Supply Chain AI Suite here. Let's talk!