FMCG
Forecast MAPE achieved, down from 22%
OTIF delivery rate achieved
About
The distributor needed a planning engine that closed the loop from demand signal to the final kilometre. Spreadsheet forecasts ran 22% off target, vans left half-empty while others overflowed, and Rs. 18 crore of working capital idled in surplus stock. We built a stacked forecasting model feeding a 3-D truck-loading optimiser, with a Monte-Carlo slider planners use to stress-test shocks before committing a load.
Industry
FMCG
Company size
5,000 – 10,000 employees
Founded
1946
The Company
A national FMCG distributor planning on instinct
The distributor manufactures and distributes a wide packaged-foods portfolio through depots feeding a national dealer network. Forecasting sat in spreadsheets maintained by a small planning team, disconnected from the truck-loading decisions made every night at the depot.
The disconnect meant a forecast could be reasonably close and the loading could still be badly wrong — vans dispatched half-empty on one route while another van turned away product that wouldn't fit, with no visibility linking the two decisions.
The challenge
Forecasts and loading plans that never talked to each other
Spreadsheet-based forecasts ran roughly 22% off target at the SKU-depot level, and because loading was planned manually against those forecasts, the errors compounded — Rs. 18 crore of working capital sat in surplus stock built against demand that never materialised.
Manual loaders also ignored cube utilisation and axle limits in practice, returning 17% of SKUs for re-shipment and drawing regulatory fines on overloaded routes.
The Solution
A stacked forecast feeding a 3-D loading optimiser
We built a stacked Prophet-LightGBM ensemble generating SKU/depot forecasts, feeding directly into a mixed-integer optimiser that packs vans in three dimensions while respecting driver hours and shelf life. Planners can pull a Monte-Carlo slider to stress-test demand shocks before committing a load.
For live leadership reviews, a Streamlit interface calls Google OR-Tools' 3-D bin-packer with a simulated-annealing polish and posts the final manifest straight to the TMS, with an auto-drawn load diagram for the driver.
The Results
MAPE below 10%, OTIF above 95%
Forecast MAPE fell to 9.8% against a 12% target, and OTIF reached 96%. Rs. 1.1 crore was freed from inventory as the tighter forecast reduced the safety stock the planning team had been carrying to hedge against its own error rate.
On the loading side, 100 vans now optimise in under 30 seconds — averaging 14 seconds — at 88% cube utilisation, saving roughly Rs. 18 lakh a month in freight that used to be spent moving half-empty trucks.
KEEP READING
Explore more studies on AI, analytics, and enterprise intelligence.
