From Quarterly Forecast
to Hourly Truth.
A Tier-1 industrial supplier replaced a three-week demand cycle with a streaming forecast wired into SAP — and shipped it in eleven weeks, freeing $24M in working capital and cutting stockouts by 38%.
At a glance
A ForecastThat Actually Forecasts.
- Client
- Acme Manufacturing
- Industry
- Industrial · Tier-1 supplier
- Region
- North America
- Layer
- Intelligence · Data · Systems
- Duration
- 11 weeks to live
- Operating model
- Co-managed retainer
The situation
Acme's planners ran a three-week demand cycle on quarterly cadence. By the time the spreadsheet model landed in S/4HANA, the assumptions were stale, the line was already short on three SKU families, and three-week stockouts were the price of running the business.
The team didn't need a better spreadsheet. They needed a forecast that moved at the speed of the operation — and an ERP that knew what to do with it.
What we built
- A streaming forecast layer
Trained on five years of order history, MRO signal, channel sell-through, and macro indicators — refreshed hourly with confidence intervals attached to every SKU-region pair.
- A decision policy inside SAP
The forecast doesn't sit in a dashboard. It writes proposed replenishment orders into S/4HANA, with planner override gates above an exposure threshold the CFO signed off on.
- An ops cadence the planners own
Daily standup against the override queue, weekly retro on misses, monthly model performance review. The planners moved from data entry to senior judgement.
The handover
By week sixteen, Acme's data engineering team owned the pipelines, their planners owned the override queue, and our role compressed to a bi-weekly architecture review. The layer is theirs.
Outcomes
Eleven Weeks.Production. Not Pilot.
Working capital freed in the first six months.
Stockout incidents across the top 200 SKUs.
Service-level on the three previously red product families.
From kickoff to live ERP writes.
Median decision latency on replenishment proposals.
Per SKU-region forecast — accountability before automation.
The forecast wasn't the win. The win was that our planners stopped fighting the system and started running it.
A System That Sticks.Long After Go-Live.
The industry built a generation of AI tools that give you better information and then leave you to do something with it.
Explore Vertical AgentsTheir data engineers run the pipelines, their planners own the override queue, and their SRE team owns the on-call. We sit in the architecture seat for the bi-weekly review.
Every replenishment write above the exposure threshold goes through planner sign-off. Drift surfaces in the weekly retro before it touches a PO — and the model is re-trained on the cadence that beats the drift.
Almost certainly. The playbook transfers across discrete and process manufacturing — it's the data contract, integrations, and decision policy that get retuned per operation.
Want the same engagement, shaped to your operation?
Your Operation Has a Next Layer.Let's Build It.
The gap between where your business is and where AI can take it isn't a technology problem. It's a deployment problem. We solve it.
