Winning artificial intelligence

artificial intelligence

What will be the most successful application of artificial intelligence (AI) in the supply chain? Since IT vendor OpenAI launched ChatGPT in November 2022, there has been a real hype around generative AI. According to recent research by analyst firm Gartner, half of supply chain executives expect to adopt generative AI (GenAI) this year. However, they probably do not yet have a concrete idea of how to meaningfully deploy this text- and code-generating AI application, trained by a large language model, in their supply chains and at what scale.

Coffee producer Jacobs Douwe Egberts integrated a user-friendly chatbot into scale up Garvis’ demand planning software in a successful pilot project in mid-2023. With it, random employees get accurate and accessible answers to, for example, the financial and logistical implications of a promotional campaign.

Biggest obstacles

The biggest obstacle to enterprise-wide adoption of generative artificial intelligence is the necessary harmonization of the underlying supply chain software and the data generated. In addition, it is necessary to have encryption technology to prevent corporate data from leaking into OpenAI’s public application.

Also, the licensing costs of generative AI can be huge. A ChatGPT enterprise licence costs upwards of $9,000 for 60 users, with additional costs for each query (‘prompt’) on top of that. If half of all companies really do adopt generative AI, I foresee this bubble bursting next year due to exorbitantly high costs and disappointing revenues.

Personally, I see the greatest potential in the supply chain in so-called ‘narrow’ AI. Swiss start-up Afflux, born out of the EPFL technical university in Lausanne, has already completed a whole series of successful AI projects in the supply chain using the combination of a simulation model and optimization algorithms. By accurately modelling an existing production line in a simulation model and then using various algorithms to optimize production scheduling, productivity improvements of between as much as 10 to 30% have been realized.

Tailor-made modelling

Standard software for production scheduling often has only five or ten technical constraints, while a simulation model can approximate reality with many more bottlenecks with up to 99% accuracy. These digital twins of production lines and distribution networks will be the most successful AI applications in supply chains. However, they require customized modelling.

Martijn Lofvers, Chief Trendwatcher Supply Chain Media