In this brief project, a colleague and I analyzed and predicted the customers behaviour of an asian healthcare retailer interested in assessing the effectiveness of its promotions and discounts. The notebook is split into 3 sections:
- Data cleaning and Prepration (wrangling).
- Promos and Discounts insights.
- Time-series Baseline sales forecasting.
The objectives and the tasks are explained below. The analysis is maninly focused on the performance of some products:
CONTEXT
• Oliver Wyman have been engaged by an Asian health and beauty retailer to help design a set of promotions in line with their existing promotional offering and long-term sustainability goals
• The retailer has an online and offline (in store) presence in multiple countries across Asia but we have been engaged to focus on the Malaysian business unit and their offline promotions
• They have also asked to the team to focus on a subset of the products within a specific product category
• To support the Oliver Wyman team, the client has shared multiple datasets for us to understand the current performance of their products and the promotions they offer as they want to move towards a more data-driven approach. Previously individual category managers would rely on industry expertise to set up promotions and campaigns but were unable to review the effectiveness of the promotions coherently across the business. The project was divided into different tasks, listed below:
TASK
You are working with the new project team and as the data and analytics consultant on the project, you have been asked by the senior members of the team to assess the data shared by the client, conduct exploratory data analysis and build a simple model to predict product baseline sales.
More defined tasks
-
• The local business teams shared the transaction data for their entire transaction data history. Unfortunately, some of the business teams use manual steps to create the transaction data, which impacts the overall quality of the data • Additional details on the datasets provided are available in Appendix A • Look at the data provided and prepare it so that it can be used for further analysis/modeling – What data quality issues exist in each dataset? Are values in line with expectations? Were there challenges in joining the datasets? • The leadership team is also interested in the overall quality of the data and would like an assessment of what things can be improved to make the data more useful as the organization looks to become a more data-driven organization. What recommendations for improving their existing data can you think of?
-
• The local business teams shared the transaction data for their entire transaction data history. Unfortunately, some of the business teams use manual steps to create the transaction data, which impacts the overall quality of the data • Additional details on the datasets provided are available in Appendix A • Look at the data provided and prepare it so that it can be used for further analysis/modeling – What data quality issues exist in each dataset? Are values in line with expectations? Were there challenges in joining the datasets? • The leadership team is also interested in the overall quality of the data and would like an assessment of what things can be improved to make the data more useful as the organization looks to become a more data- driven organization. What recommendations for improving their existing data can you think of?
-
• The client wants to be able to understand the performance of products under promotion with a more quantitative measure • In order to do this, we need to first understand the performance of products without promotion e.g., what is the baseline of sales without promotions, and then measure how effective promotions are on top of this baseline • We will need to train a model to predict a baseline of sales without promotional activity. During this process, you may need to consider – If or how to use days where a promotion is active in the training data – What other factors may influence consumer behavior beyond promotion (e.g., weekends, seasons, holidays, natural disasters) and how they can be appropriately captured as factors in the model (binary, numerical, categorical etc.) – How to capture long-term trends connected to the product • A common metric to measure promotion effectiveness for a given product is elasticity. This relates the additional uplift in sales gained to the discount given away and the client is interested in using it as a consistent promotion performance metric
It is not possible to share the data due to confidential agreements. Hope you enjoy!