Future Retail View: AI turbocharges supply chain demand forecasting

Retail Technology caught up with Raphael Bertholet, Supply Chain CTO for the retail and CPG division of SymphonyAI, on the impact of AI innovation in demand forecasting and the advent of Forecasting-as-a-Service

Supply chains face unprecedented challenges, where volatility and uncertainty are now the norm, putting the ability to match supply to demand into sharp focus.

Hot on the heels of engaging with industry practitioners and analysts at a key industry trade show, Raphael Bertholet, Supply Chain Chief Technology Officer (CTO) for the retail and consumer packaged goods (CPG) division of retail software provider SymphonyAI, confirmed high interest in developments to improve demand forecasting accuracy.

“At the beginning of this year, NRF confirmed how central demand forecasting is to the overall success of the supply chain,” Bertholet told RetailTechnology.co.uk. “We heard that from customers and analysts, who wanted to know how AI and generative AI can significantly improve demand forecasts.”

He said interest in demand forecasting tech development was high in terms of accuracy, scalability, and the amount of resources that retailers and consumer packaged goods (CPG) manufacturers have to direct at it from a corporate level.

Looking back to stay ahead

Sophisticated machine learning (ML) algorithms have been central to demand forecasting systems for many years. However, recent global disruptions have spurred new AI innovation.

Bertholet explained: “Using an AI-driven forecasting approach, we can now use more contextual data to model various scenarios in future. We’re gaining that not so much from time series data but by virtue of the contextual data we’ve gathered.”

He added that only AI-driven forecasting software can fully utilise contextual data. At the same time, it can support the complexity of simulating various scenarios to meet the need for increased supply chain resilience and adaptability.

“So, rather than wait for something to happen and then figure out how to react, the idea now is to enable the modelling of a variety of scenarios so that we can have plans in the can, so to speak,” he said.

Managing costs and margins

A good demand forecast should also ensure retailers and CPG manufacturers maximise potential revenue and minimise waste. In combination, they contribute to overall profitability.

“So, in that sense, the demand forecast is, has been, and always will be the critical first starting point,” Bertholet agreed. “But the demand forecast is not as useful in aggregate. We need to be able to forecast accurately by store and by channel too to manage the ecommerce upsurge.”

Accurately understanding demand by channel ensures that inventory and workforces can be positioned appropriately to satisfy ecommerce orders - picked out of live or dark stores or distribution centres, etc. - and fulfilled optimally for the highest margin, efficiency, and customer satisfaction.

He added: “What we are very interested in at SymphonyAI is to link the forecast with the customer behaviour insights we have through our customer analytics solutions. This approach ensures we bring a retail understanding to a problem that could otherwise be processed with a theoretical-only approach.

“This requires understanding and modelling the contribution of external factors, as well as prices, promotion, or transferable demand, due to assortment decisions at the customer segmentation level. We do this before looking at it at the ‘aggregated item-store level’ to build an item-level store forecast of the demand. Reconnecting the understanding of the customer to the demand forecast is very powerful."

Innovating with artificial intelligence

Bertholet also highlighted how the day a traditional statistical forecasting model is implemented will be the best it performs. Then, without careful management, it will degrade over time.

“The opposite is true of AI,” he stated. “The day [the AI-driven model] is implemented is probably going to be its worst performing day, and then it will, programmatically, by training and learning, improve over time.” However, the ability to make changes is also critical.

“Users can either add particular events or say, ‘no. I believe there’s something I know the machine doesn’t, and I think the forecast is actually going to be 10% more,’ or whatever. But we do it within the context of a concept called ‘forecast value add,” he explained.

“We now persist in our database the original AI-generated value and the new forecast value generated after the user change. We measure accuracy against both values and display through the UI [user interface] and visualisations to determine whether the intervention has improved, degraded or had no impact on the forecast, in aggregate and individually over time. Based on that feedback loop, we can adjust the AI for more accuracy going forward.”

Democratising demand forecasting

The supply chain software provider is also introducing a generative AI demand forecasting copilot, which can guide users through onboarding a new item with no historical sales data and create a demand forecast by linking it to similar items already sold, for example.

“It takes what was before maybe a 30-minute process down to five seconds, which is powerful,” Bertholet said. “We’ve also found gen AI is very powerful in identifying gaps between forecasted and actual demand due to events such as supply chain disruptions, which can then, in turn, be used to improve the model.”

It has also launched 'Forecasting as a Service' (FaaS) demand planning delivery using a software-as-a-service (SaaS) subscription model. The service creates and manages the forecast with ingested, cleansed, harmonised and enriched retailer data.

“The end users, the retailer and CPG customers themselves, have very little to do in terms of the day-to-day production and maintenance of a forecast, which means a couple of things,” Bertholet said. “One thing is that their scalability goes through the roof.”

The other is that they can maintain a consistent, specialised team that can successfully handle highly accelerated growth. The team can also spend less time on manual tasks and more interacting with downstream teams on replenishment and tactical planning.

Mission-critical resource delivery

The advent of FaaS also puts supply chain software delivery under the spotlight. Here, Bertholet said that the scalability of the Oracle Cloud Infrastructure (OCI) allows SymphonyAI to scale this new service-based offering efficiently.

“That translates into direct benefits for customers, not only in terms of performance but also cost. We can optimise the performance of our solutions with Oracle technology, including the Oracle databases we use, in addition to the OCI technology,” he concluded.

This interview is part of a sponsored campaign.