AI use-case triage — Machinegeist

Seven surfaces, five tests

Seven surfaces where AI plausibly earns its keep at Vandemoortele, scored the way we would triage them. Every cell is a hypothesis built from public sources; the point of this hour is that you correct us. Sources and reasoning in our desk research, available on request.

Surface	Business value	Business anchoring	Data readiness and access	Ownership	Timing and risk
Demand planning and forecasting	High. Waste down, service level up, planner hours back, on shelf-life-sensitive volume	Strong: central S&OP team owns it and the pain is recent	APO and BW history rich, Anaplan live; access medium via BW extraction paths	Central supply chain with the CIO backbone	Proceed. Acquisition perimeters phase in later
Procurement	High. The 2025 margin story was commodity prices	Strong: an AI agent is already in development, momentum exists	Ariba and D&B foundations; access medium	Procurement with the CDO/CIO seam to settle	Proceed. Transparency duty (EU AI Act Art. 50) applies to the agent
Master data and acquisition integration	Medium to high: the integration-year multiplier, KPI scope gaps close	Diffuse: recurring theme, dedicated governance roles, no sharp problem owner yet	Data exists across 5+ ERPs; access hard by definition	Unassigned, spans CIO and business lines	Careful. UK perimeter frozen until CMA (~Oct); rest can proceed
Commercial and customer-facing	Medium. Portal share growth, next-best-offer, churn signals	CDO home turf: portals and CRM live and funded	Portal, CRM and HubSpot data fresh; access easiest of the seven	CDO, cleanly	Proceed. Art. 50 applies to anything customer-facing
Sustainability and CSRD data	Medium. Compliance relief plus freed analyst time	Publicly named as resource-heavy and still being refined	Smart meters, traceability feeds; Databricks landing exactly here	Sustainability with the CIO platform	Proceed. CSRD calendar drives it
Production, quality, maintenance	Medium to high per plant (OEE, waste)	Mixed: investments made, engineers still to be convinced	Plant data partially on Databricks; varies by site	CIO and plant managers	Park this year unless a plant pulls. Food-safety law binds QC
Transport and cold chain	Medium. Routing, load consolidation, OTIF on 2,000+ orders/day	Own transport company exists; no public AI initiative yet	TMS and telematics presumed; unverified	Vandemoortele Transport Solutions with group ops	Proceed as exploration: small, acquisition-proof perimeter

How we would sequence a shortlist: pick one use case that pays within two quarters (demand planning or procurement, both build on motions you already run) and one structural bet (master data or sustainability data). Ownership assigned per use case: who runs it, who sponsors it at ExCo level. EU AI Act classification is one triage criterion among five, not a project.

If demand planning is the center: three concrete plays

Three plays behind the demand-planning surface. Same status: hypotheses, for correction. In each one your systems keep doing what they are good at; what gets added is the layer of judgment and knowledge around them, encoded so it compounds.

The play	Your tools keep doing, the brain adds	First slice (weeks, not months)
1 · Demand-signal bridge. My.Vandemoortele orders, baskets and CRM promo calendars feeding the short-horizon forecast	APO and Anaplan keep computing the forecast. The brain interprets the signal: what a promo intends, which customers move first, when basket drift means demand and when it means noise	Backtest, one product family, one market: would portal and promo signals have improved the 4-week forecast against actuals? Data out, no integration
2 · Planner copilot. The monthly exception review drafted in advance; the planner decides	The statistical engine stays untouched. The brain encodes what today lives in planners' heads: override logic, customer patterns, promo memory. Every cycle it captures more	Shadow one cycle, capture the override logic, draft alongside the planner the next cycle; measure time saved and hit rate
3 · Day-1 planning kit for acquisitions. New SKUs with no history planned from behavioral twins in the existing portfolio; master data resolved with AI assistance	The ERP integration runs its own course. The brain does what integration timelines never allow: resolving entities across catalogues and matching new products to their lookalikes, ready when Bunge closes	Dry run on Délifrance SKUs already in-house; the kit is proven before the date you do not control

All three build the same underlying layer: the planning brain, the judgment and knowledge around your systems, encoded so it survives rotation and compounds across cycles. The systems of record stay the systems of record; the brain is the memory between them. First extension once any play is live: an at-risk-stock playbook on the food-loss tracker you already run.

Terms: S&OP = sales and operations planning · APO = SAP Advanced Planner and Optimizer (current planning system) · BW = SAP Business Warehouse · OTIF = on-time in-full delivery · TMS = transport management system · OEE = overall equipment effectiveness · CSRD = EU Corporate Sustainability Reporting Directive · CMA = UK Competition and Markets Authority.

Machinegeist · machinegeist.ai · prepared 2026-07-09 for the AI in Supply Chain working session