Problem Description: FMCG companies like Nestle face trouble in forecasting demand for smaller
regions which comprises nearly 50% of their business and is highly critical. This is due to high volatility in
demand. Due to this problem more often than not the sales force in these regions face a situation
wherein they are either short of inventory and unable to meet demand or have piled up inventory at
warehouses. A model that effectively forecasts sales can be tested on a small region (in this case
Ferozpur sales zone) which if successful can later be deployed to other smaller regions which can be
highly beneficial in management of inventory and thus production.

Brief Description of Data: The data has been sourced from Nestle SDS software (ERP system).The
dataset obtained contains daily bill-wise product-wise sales and retailer-wise product wise sales at a
distributor level for the time period Jan 2012 to Jan 2017. We have considered the top 6 products in
different product categories of Nestle based on Sales volumes, i.e., Maggi, Eclairs, Nescafe Coffee
Classic, Everyday daily whitener, Munch & Cerelac. Since the data is actual company sales data and
Nestle has a very robust reporting mechanism, therefore the data was accurate & complete (except for
2 values which possibly got lost while extracting the data from the system. We manually confirmed
those two values through the system and updated them). Data is consistent, unique and timely, i.e., it is
in ascending order, there are no duplicate time stamps and they match the precision of a calendar.

Final Forecasting Model:
The final model is a tool that is able to forecast the sales of top 6 products by choosing the method most
suited for forecasting sales for that product. The manager will need to input the product name in the
model and the model will use historical data to come up with the forecast. The most suited model is
determined based on the predictive accuracy (the one with the least expected error in forecasting) of the
method on a product time series. The overall performance of the model as compared to the benchmark
is 12-15% better in terms of % error.

Conclusion:
Based on our analysis, we found that:
1. Quarterly forecasting performs much better on predictive accuracy as compared to monthly
forecasting (Refer Exhibit 1).
2. Use of different forecasting models to estimate sales for different products gives maximum
predictive accuracy.
3. Predictive power of some models might be impacted by lack of sales promotion and marketing
campaign information.
4. The forecasts for the month of Feb ’17, Mar ’17 and Apr ’17 have been provided in Exhibit 2

Recommendations:
1. We should go ahead with quarterly forecasting as it provides a better estimate.
2. Forecasting horizon should not be more than one quarter as it is a learning based model.
3. Predictive power of the model should be improved by taking into account the effect of sales and
marketing initiatives along with other external factors (macroeconomic, competitor strategy).