Home / Case Studies / Frisian Flag SKU Intelligence
The ERP data was there.
Nobody could act on it fast enough.
The challenge
Frisian Flag Indonesia manages hundreds of SKUs across modern trade retail channels. Dairy and dairy-adjacent products are time-sensitive by nature: expiry windows are short, restocking cycles need to be precise, and a slow-moving SKU left on the shelf too long becomes a write-off.
The ERP system held most of the data the business needed: stock levels, movement history, batch dates, replenishment records. But the data sat in a system built for recording transactions, not for surfacing decisions. By the time the information reached someone who could act on it, through manual exports, weekly reports, and field rep updates, the window to do something useful had often already closed.
The consequences showed up in two places. Products approaching expiry were flagged late, leaving insufficient time to run promotions, reroute stock, or pull items before they became waste. And restocking decisions, without a reliable signal about what was actually moving in which stores, were often either too early or too late.
Good data, sitting in the wrong place, reaching the wrong people, too slowly. No early warning on expiry. No confidence in restock timing. No view of which SKUs were genuinely earning their shelf space.
What PGI delivered
PGI designed and built an end-to-end analytics and ML platform that pulled from Frisian Flag's existing ERP, applied machine learning models to the historical stock and sales data, and delivered actionable output through Power BI dashboards accessible to both central planning teams and field teams.
The platform was not built to replace what teams were already doing. It was built to give them information early enough to actually do something about it.
- SKU Performance Analysis — velocity ranking by store and region, slow-mover identification, sell-through rate tracking over time, SKU contribution to category revenue, underperformer flagging by threshold
- Expiry Tracking and Alerts — batch-level expiry visibility per store, rolling countdown to expiry by SKU, alert triggers at configurable day thresholds, at-risk stock volume quantification, dashboard view by store and by product
- Restocking Prediction — XGBoost model trained on historical sales, demand signal by store, SKU, and day-of-week, predicted stockout date per location, recommended restock quantity and timing, model retrained on rolling actuals
How the ML model was built
The restocking prediction model was built in Python using XGBoost, trained on historical transaction and movement data extracted from the ERP. Significant time was spent on feature engineering before any modelling started, identifying which variables actually predicted stockout events versus which were noise in the data.
Key features fed into the model: sales velocity per SKU per store, day-of-week and seasonality patterns, lead time history from previous replenishment cycles, and current stock level relative to average weekly draw. The model outputs a predicted stockout date and a recommended restock trigger point for each SKU-store combination.
Expiry tracking was handled separately through a rules-based system alongside the ML layer. Batch dates were ingested from the ERP, matched to current store stock levels, and surfaced through a prioritised alert view giving planners a clear picture of what was at risk in the next seven, fourteen, and thirty days.
All outputs, the SKU rankings, the expiry alerts, and the restocking recommendations, were delivered through a Power BI dashboard designed in close collaboration with the Frisian Flag planning and field teams. The goal was to make the interface something a field rep could read and act on without needing to understand the model behind it.
Measurable outcomes
vs previous manual process
(from weekly reports)
restocking, in one dashboard
The difference between a seven-day warning and a one-day warning is the difference between a markdown and a write-off. Expiry-related waste reduced. Restocking decisions became faster and more confident. The SKU picture became honest.
Technology & capabilities
Working on a similar challenge?
Talk to our team about SKU analytics, demand forecasting, or ERP-to-intelligence platforms.