FMCG and Consumer Goods · AI
Demand forecasting by SKU and region for an FMCG company
Forecasting models that predict weekly demand for each SKU in each region, using secondary sales, schemes, festivals and weather. Planners review the forecast in a dashboard, adjust it with a reason, and the agreed numbers flow into production and depot replenishment plans.
- Client
- A packaged foods and snacks company selling through distributors across western and central India
- Region
- India
- Sponsor
- Head of supply chain and the demand planning team
- Duration
- 18 weeks to first release, then monthly model reviews with the planning team
Engagement model
FDE: Forward Deployed Engineering
Our engineers joined the demand planning team for the monthly S&OP cycle, so every model change was tested against the planners' own judgement before it was trusted.
How FDE worksThe team
- Engagement lead1
- Data scientist2
- Data engineer2
- BI developer1
- Shared QA1
The challenge
Where things stood
- Planners built the monthly forecast in spreadsheets from last year's sales plus a growth figure, at brand level only.
- Some depots ran short of fast moving SKUs before Diwali and other festivals while slow movers piled up and neared expiry.
- Trade schemes and price changes were agreed by sales but reached the planners late, so their effect on demand was guessed.
- Primary sales to distributors were in SAP, but secondary sales to retailers sat in a separate distributor management system with gaps.
The approach
How we ran it
Weeks 1 to 6
Data foundation
Brought SAP primary sales, distributor secondary sales, the scheme calendar, price lists and a festival and weather calendar into Databricks, and cleaned SKU codes that had changed over the years.
Weeks 5 to 12
Model building
Built forecasting models at SKU and region level with scheme, festival and seasonality effects, and back tested them against the planners' own past forecasts to see where each did better.
Weeks 10 to 15
Planner dashboard
Built a Power BI view where planners see the model forecast, last year and their own number side by side, override with a reason, and lock the consensus forecast for the S&OP meeting.
Weeks 15 to 18, then monthly
Live cycles and handover
Ran two full S&OP cycles with the models live, sent the agreed forecast to SAP for production and depot replenishment planning, and trained an in house analyst to retrain and monitor the models.
What it looks like
Representative screens
Drawn to show the kind of system we delivered. Client screens and data stay private.
Depot stock cover against plan, week 42
SKUs at stock out risk
14
SKUs near expiry
9
Overrides this cycle
37
Stock value near expiry
INR 6.4 lakh
- Ahmedabad82
- Indore64
- Nagpur71
- Pune58
- Raipur45
What we delivered
Modules and features
- Unified sales data model combining SAP primary sales and distributor secondary sales
- Weekly forecasting models by SKU and region, with scheme, festival and seasonality effects
- Planner dashboard with model forecast, override and reason capture
- Forecast accuracy tracking by SKU, region and planner override
- Consensus forecast export to SAP for production and replenishment planning
- Early warning list of SKUs at risk of stock out or expiry at each depot
- Scheduled retraining pipeline in Apache Airflow with model monitoring
What changed
Outcomes
- Planners forecast at SKU and region level
- Festival and scheme effects are built into the forecast from the data
- Every planner override is recorded with a reason, so the team can see which adjustments helped
- Depots get early warning of likely stock outs and near expiry stock
Looking back
Secondary sales data was the weak link. Some distributors uploaded late or not at all, and the models had to fall back on primary sales for those regions, which we should have flagged to the client as a limit in week one. Showing the planners' own past forecasts next to the model in back testing earned more trust than any accuracy chart, because they could see where the model was better and where their local knowledge still won. Next time we would get the scheme calendar from sales as a shared system input from the start; chasing it by email each month was the most fragile part of the process.
Start a conversation
Discuss a similar project
Tell us where your team is today and we will walk you through how a project like this would run for you under FDE. The first conversation is free. Delivery work is paid only after you accept it.
A specialist replies within one working day. The first conversation is free.