The promise of autonomous agents
The latest supply chain trend is based on AI agents.
Called “Agentic AI”, this technology promises to enable artificial intelligence to perform tasks autonomously, without human supervision.
This technology is often described as the future of productivity in supply chain planning and execution. It’s the hype of the moment, according to the high priests of the cyber hype… https://www.gartner.com/en/webinar/689201/1539147
It’s promising… but totally unaware of the current state of most corporate management systems
An giant leap
The introduction of AI agents into the supply chain management systems of many industrial and distribution companies, including some of the world’s leading firms, will probably have to wait a while – and first involve the structuring of data and business rules followed by human agents. Let’s call it “Agentic HI” for Human Intelligence…
The vast majority of companies struggle with an inflexible ERP system, data of average quality, multiple Excel sheets – and above all they face a crucial lack of operating principles, alignment between functions, business rules. They lack a clearly defined operating model. They lack a feedback loop to improve the system. They are constantly confronted with the variability of internal policies and managerial decisions that are not always rational
Inserting an AI agent into such a system means adding the risk of hallucination to a “garbage in / garbage out” flow…
Start with a discipline that can be automated
To automate, we need to model. Modeling means representing physical reality and codifying business rules in a digital solution that humans can trust. Once we have confidence, we can automate and entrust day-to-day operations, as well as part of the model improvement process, to a machine.
Let’s take a basic example: placing orders to replenish stock.
If you let your buyers judge how much to order based on forecasts they don’t trust, a safety stock value that was established a while ago by someone else, and friendly pressure from the manager who doesn’t want any shortages, there’s a long way to go before you can put in AI…
You can set up a replenishment model – let’s call it DDMRP – ensure that it’s always correctly sized – let’s call it “Autopilot” – and define a management rule: when the flow equation reaches the top of yellow, replenish to the top of green.
Now you have a basic model. You can measure its effectiveness and improve it. And you’ll soon be able to automate it. In the example below, the discipline is followed and the performance is acceptable, so we’re probably ready to entrust it to an autonomous AI agent – or simply, as a first step, to activate automatic approval.
If, on the other hand, you have a situation like the one below, where your buyer ignores the recommendations and places a huge order for no apparent reason, you won’t be able to validate the relevance of your operating model.
In this case, there’s no need to think about automation. You and your teams must first validate the model and the business rules, ensuring that the defined discipline is followed on a daily basis, and that you orchestrate performance reviews to initiate model improvement. How do I know if my replenishment model is effective, if I don’t follow its recommendations?
From HI agents to AI supervisors…
Our recommendation in the quest for performance and productivity is therefore to start by involving your teams in validating the operating model that matches your flows. In this process, your planners will become, on the one hand, supervisors of this operating model, and on the other hand, “HI agents”, who carry out the discipline of the model, without tinkering in Excel… Once this discipline is well established, it will be easy to transfer this activity to an AI agent – and refocus your planners on monitoring the model and adapting it to market realities…
