Widespread integration of machine learning (ML) into supply chain management by major pharma suppliers is just a few years away, says cold chain expert Gianpiero Lorusso. Moreover, he adds that biopharma companies willing to adopt AI quickly will have an early mover advantage.
Lorusso explained the ways in which ML tools can enhance biopharma supply to industry representatives during a roundtable session on 26 February at the 2025 Clinical Trial Supply Europe conference in Barcelona, Spain. With sufficient data training, he demonstrated how ML tool Genlots, an AI-powered enterprise resource planner (ERP) based in Morges, Switzerland, can optimise trial and commercial supply chains for several metrics, including cost, emissions and time.
Go deeper with GlobalData
Lorusso likened the use of ML in supply chain management to the large-scale adoption of AI chatbots. He gave the example of Microsoft’s Copilot, a generative AI chatbot. As with Copilot, he stressed that tools could act as plugins, integrating seamlessly with existing supply management software and being able to be removed if desired.
Accordingly, Lorusso claimed that early ML adopters can expect a first mover advantage. By closely monitoring material needs and forecasting future demand, he said currently available AI can bundle purchase orders for “quantity discounts” and maintain minimal reserve stocks without the risk of running out, stating, “You also have a target inventory that you can set in the parameters.”
Maintaining steady stock reserves has become increasingly important in the industry as geopolitical tensions cast uncertainty on supply chains, according to GlobalData analyst Carolina Pinto, speaking at the conference on 25 February.
ML could also prove instrumental in enabling companies to meet emissions targets, as Lorusso pointed out that through optimised purchasing, suppliers would be able to minimise deliveries and the concomitant CO₂ emissions of frequent, small transports.
“When you do it manually [without the help of ML] – you cannot take into account all these parameters before issuing an order,” he said. According to Lorusso, with six months of training, an ML algorithm can not only optimise the ordering process, but offer order simulations with comparative data on cost, time, emissions, and other metrics versus ML-free purchasing strategies.
Lorusso acknowledged attendee concerns around the quality of the “master data” on which AI is trained, addressing the balance between ensuring such data is accurate and gaining the advantages of early adoption. He cautioned that due to the continuous effort needed to update data to align with market developments, “the master data – will never be perfect because business is changing, your products are changing, the market is changing”, imploring industry representatives to implement ML in their purchasing while quality-checking data used in parallel.
