A long history of use A long-standing familiarity, more accurate than ever nowadays
Artificial intelligence is often considered as a brand-new revolution. In Supply Chain, that is not the case: we have been working with AI for decades. It simply did not carry that name, we spoke of ‘expert systems’, ‘heuristic engines’, ‘forecasting algorithms’. But the idea was already there: to delegate calculations that humans cannot handle across thousands of references.
Supply Chain has this advantage: artificial intelligence tools have been available to us for a long time, battle-tested, well-documented and embedded in our systems.
This is a particularly valuable asset in pre-sales, where clients want to see what they heard about during their Supply Chain certification. My role is of course to show them, but then to bring some nuance during the deployment.
AI in forecasting does not present itself as an opaque system that strips the forecaster of control, but as a set of tools at their disposal for large-scale tasks.
This is where AI reveals its full potential. It gives the forecaster the right tools to independently manage a complete sequence that combines AI calculations and human input:
- AI processing of abnormal data points (Cleansing),
- human cleansing of past exceptional events (stockouts, promotions)
- and build a reliable baseline to start from.
- AI runs all extrapolation algorithms from its library
- and selects the best fit,
- the human steps back in for future events enrichments
- and for validation processes.
Other forecasting processes, such as collaborative planning or new product forecasting through aggregation and disaggregation, genuinely require an intimate understanding of the underlying assumptions and should, in my view, remain in human hands.
The point is not for the machine to decide on your behalf, but for it to enable the forecaster to support and defend their own assumptions.
Planning AI operates as a black box, currently beyond the reach of most planners.
Things get considerably more complex when it comes to planning.
Heuristic modules have for long offered powerful mechanisms:
- backward then forward scheduling,
- demand sorted by priority,
- dichotomous order splitting to smooth the load
- exploration of alternative routings when an operation cannot be scheduled.
“The planner retrieves a plan that meets all constraints, produced in a single click.
In an industrial context, this falls short when faced with the full complexity of real-world operations.
Mastering production plans would require the planner to have a thorough understanding of all AI parameters, as well as perfectly accurate technical data (routings, bills of materials) and dynamic data (work-in-progress, production declarations, inventory, procurement).
Let whoever has no incorrect data in their ERP cast the first stone.
On top of that, the finance team locks routings and bills of materials for cost calculation purposes, leaving the Supply Chain teams with no control over making that data operationally accurate.
AI produces a plan the planner cannot explain.
And projects fail to deliver.
From this observation grew a conviction: what is needed is a planning tool that puts the planner in control, built around three simple principles:
The planner works in Excel first to be free, and must be given full control over the modelling.
Then, the planner spends their time verifying data: inventory, work-in-progress, order confirmations. All of this data can be interfaced and the planner left free to correct what needs correcting.
The planner builds their production plan order by order, in an iterative way: that is exactly what the tool must offer them.
Finally, several planners work together, by site and by production level: let us make use of real-time technology.
4 MVPs to build, allowing the planner to:
- Model their industrial network: decoupling points, sites, resources and routing and bill of materials templates.
- Correct their operational data: by bringing together in one place all the information needed to verify and fix it.
- Plan production need by need: process each demand one by one, check whether the load fits the resources, and whether components can be obtained.
- Work in real time with multiple users on the same plan, the MMORPG of production planning.
AI only intervenes to plan the demands the planner selects, in the defined order, and the planner must be able to go back and rework the plan.
To be continued…