manual forecast overrides
50% → <1%
Pilulka · Demand forecasting
Demand Forecasting: Pilulka Cut Manual Forecast Overrides from 50% to Under 1% Across 18,000 SKUs
See how Duvo built a full demand-forecast tool for Pilulka in 2 weeks that cut manual overrides from 50% to under 1% and recovered €35K/yr.
planner time returned
~7h/day
forecast accuracy
99%+
The challenge
Manual forecast overrides cut 50% → <1%
Pilulka is a Czech online pharmacy and health & beauty retailer. Demand forecasting across its ~18,000-SKU assortment ran on a static, rule-based forecast — a fixed-logic engine that could not see what was actually happening in demand. The rule-based forecast was inaccurate. It did not account for recent demand peaks or emerging trends, so it consistently produced the wrong numbers. The result: planners had to manually override 50% of its output every cycle — half the forecast was effectively rebuilt by hand.
How Duvo improved the workflow
From signal to verified outcome.
Each step runs autonomously. Every action is logged, traceable, and reviewable.
01
Reads demand across three windows
Pull sales over the last 3, 7 and 30 days. The short windows catch products that are newly trending — strong over 3 and 7 days even if the 30-day history looked weak — so emerging demand is forecast, not missed.
02
Forecasts across the full assortment
Generate a per-SKU forecast over all 18,000 SKUs in ~1h20m, select the SKUs that need action, and produce the recommended quantities. Compare against the existing system forecast; flag the predicted stock-out day per SKU from current sales velocity.
03
Applies every constraint
Hold minimum stock levels and keep warehouses full enough to serve availability (category A 99%, A+B 96%); hold DIO under 30 days; round to carton quantities where the supplier ships by carton; apply MOV/MOQ and allowed order days; reflect live promo activity; recognise new products and exclude clearance / outlet items that will not be reordered.
04
Solves the minimum-order gap
When a forecast line reaches at least 90% of the supplier's minimum order value, the forecast proposes how to fill the remaining 10% to clear the threshold — rather than leaving the line short.
05
Explains
Write the reason for every recommendation so the planner can judge whether the forecast logic is valid, then approve.
The difference
Before and after DUVO.
Before
- Static, rule-based forecast blind to recent demand peaks
- 50% of output rebuilt by hand every cycle
- 10/30-day history only
- Manual, inconsistent supplier-rule compliance
- Inventory discipline (DIO) not actively held
After Duvo
- <1% of forecast lines manually overridden
- 99%+ forecast accuracy — planners approve, not rebuild
- 18,000 SKUs forecast in ~1h20m per run
- 3 / 7 / 30-day windows catch trending products early
- 100% of forecast lines compliant with supplier rules
Systems involved
Static rule-based system forecastWarehouse and stock dataSupplier order rulesPromo / campaign data
The difference Duvo makes
- Manual overrides 50% → <1% · ~7 hours/day of one planner's time returned
- €35K/yr commercial target plus recovered revenue from captured stock-outs
- Category A availability 92% → 99%, A+B to 96% · DIO target <30 days
- Supplier-rule compliance 100% of forecast lines · every recommendation auditable
- Built in 2 weeks
Quote
“Duvo helps us react to shifts in demand far earlier. Instead of deciding mainly on ten-day or thirty-day views, we see the signals from the last three days and can act on them before the problem ever shows up in availability.”