Forecasting customer demand for packages in SIG Indonesia market for - PowerPoint PPT Presentation

Forecasting customer demand for packages in SIG Indonesia market for better customer sales promotion Team 2 Shang-Chi Tu, Beverly Lin, Ken Wu, Jimmy Wang

Background Founded 1853, SIG is headquartered in Switzerland and offers services from a complete range of packaging (beverage + food), flexible filling machines & solutions for smarter factories to address the ever-changing needs of consumers.

Business goal Through forecasting future monthly sales volume of packages, SIG could offer new sales promotion for different customers in Indonesia. Client & Stakeholders Humanistic implication • SIG Dept. of Sales • Over-forecast: resource-wasted • SIG’s Indonesia customers • Under-forecast: work overloaded Challenge Opportunity • Many other factors should be considered. • Set up suitable marketing strategy among customers and seasons • Salespeople workload rearrangement Forecasting goal • Forecast customer demand for packages in the next 12 months on a customer and product type level

Data description Source: from SIG Measurable: number of sale volume on a customer and product type level Time period: Jan. 2009- Dec. 2018 Aggregation Amount :120 records in each series 45 series in total Frequency of collecting data: Monthly Unit: Thousands of package sales volume External data: 2018 Monthly Forecast provided by SIG

Data pre-processing Type A Type B Type C

Methods 954104 PC010125A Training period: 2009/1~2017/12 Partition Validation period: 2018/1~2018/12 Inactive New customer Ensemble Forecast next Customer Naive (total ordering period using External 12 months = 0 (2018 ordering < 24 months) Data sum = 0) Repeat Exponential Regression Customer Neural Auto Snaive smoothing Naive (with trend (total ordering Network Arima (ets) and seasonality) period > 24 months) We apply these methods to the repeat customer and chose the one with best RMSE(Regression).

Evaluation Ensemble: Forecast by customer segmentation Our forecast+External Data Benchmark: Snaive Inactive New Repeat Inactive New Repeat Total Total customer customer Customer customer customer Customer 0 0 1066.1 2230.2 RMSE 2699.6 2334.8 4030.4 2220.6 1152.7

Actual Results Ensemble Naive New customer New customer Validation Training Validation Training

Results Actual Ensemble Regression Repeat customer Repeat customer Training Validation Training Validation Highest RMSE = 9071.003 Lowest RMSE = 674.076

Recommendation For Customers : New Customers ： more data, more accurate Repeat Customers ： Improve the dramatically changes in demand can make the forecast model better For 2019 Sales Promotion : pay attention on those customers with potential to buy more in the future Try extra external data : such as customer satisfaction scale Demo