How to choose supply chain software when every vendor has an AI story
Source attribution: Supply Chain Movement webinar, “Choosing supply chain software in the age of AI,” hosted/moderated by Martijn Lofvers. This Radar tutorial curates and interprets the original work; it does not imply any partnership, sponsorship or endorsement.
1. Define the decision before the software
Start by identifying the supply chain decisions that need to improve: forecast approval, inventory targets, constrained supply allocation, production scheduling, scenario evaluation or exception management. For each decision, define who makes it, how often it is made, what information it requires and what a better outcome would look like.
2. Map the existing portfolio
Separate capabilities that genuinely need replacement from those that mainly need better integration, cleaner ownership or a better workflow. This avoids treating every process gap as a software gap and reduces the risk of adding another application without simplifying the operating model.
3. Evaluate AI inside the workflow
Do not score AI as a standalone feature. Ask what the capability actually changes: does it improve a forecast, generate scenarios, prioritize exceptions, recommend an action or execute one? Require vendors to demonstrate the capability on representative data and realistic constraints, not only on scripted demos.
4. Test the implementation path
Assess implementation alongside functional fit. Identify required master data, interfaces, process changes, ownership and the role of implementation partners. A slightly less comprehensive system that can be adopted and maintained may create more value than a theoretically perfect solution whose implementation exceeds the organization’s capacity for change.
5. Make the decision architecture explicit
Document where recommendations are generated, where they are reviewed and which system is authoritative for execution. This matters especially when AI layers are introduced across APS, ERP, BI and workflow tools.
Dataleo interpretation: AI has reduced the usefulness of feature-count RFPs because similar claims can now appear across very different architectures. The selection unit should be the decision plus its implementation path. The main failure mode is buying an impressive AI interface without resolving the process, data and ownership changes needed for it to influence actual planning decisions.