←
Go back to case studies
February 15, 2024
By the Compass Team
category
Machine Learning + AI
industry
Consumer Packaged Goods (CPG)
technologies
Alteryx Analytic Apps
Python
Using a low-code, no-code solution in Alteryx, we created a set of workflows to normalize all data inputs into our defined data model, performed feature engineering, and applied a clustering model using scikit-learn in Python to identify similar stores. The model is executed using Alteryx Analytical Apps which provides a simple point and click user interface so the client does not require any coding nor technical knowledge. The solution provides an intuitive way for business users to build clusters as well as the flexibility to adjust their desired scope for each market.
To effectively provide the right product mix to their customers, Category Managers at a multinational beverage firm were looking to perform clustering across their stores using machine learning. Our client was looking to tap into our firm’s data science and data engineering expertise to not only build a robust solution but to interpret the results.
The multinational beverage firm experienced inefficiencies in providing the right balance of products to stores due to their growth. In the past they had used their expert conventional wisdom to set arbitrary thresholds to determine the product assortment. Failure to have the right product assortment could lead to an overstock in some stores, and an under-stock in others, which could result in other companies stealing market share and revenue loss.
To improve product assortment nationally, the beverage company wanted to use a modelling approach that was powerful enough to provide result significance while balancing explainability to non-technical stakeholders. The idea behind the model is to be able to understand what is the right product mix to allocate at the store level given a myriad of store inputs.
The clusters identified act as a guide for Category Managers who make decisions on product assortment and allocation.
Our client wanted to identify stores which have the same relative distribution of products sold. Previously, arbitrary thresholds were set in different markets to segment stores and provide specified levels of products. The process was not data backed and relied on the intuition of experts. This solution was unsustainable if key people left the organization.
The data for different market segments had different underlying sources and therefore different number of columns. The models should be based on the same structure of data to be able to make consistent decisions on product placement. A way to standardize data was necessary to implement the solution
The client had limited technical expertise to maintain the solution and re-run when necessary. Therefore, a code-based solution was not going to be maintainable for the client so a solution with a user interface would be necessary to clean the data, run the machine learning model and visualize the results.
To solve the problem, the Compass Analytics team utilized Alteryx Analytical Apps to create a set of chained apps to first clean the data into a standard format and another set of Analytical Apps to run a clustering model using Sci-kit Learn in Python.For the first set of analytical apps the team developed a user interface for business users to select the source of data and specific columns to use to standardize all datasets into a standard format. Validation steps were built in to make sure that a final dataset can be used to perform the clustering model. This interface provides an intuitive way for any user to standardize the data.For the second set of analytical apps, the team developed a user interface to select the parameters to run the clustering model. The Alteryx workflow leverages the Python Tool to implement the clustering model in Sci-Kit Learn and generates diagnostic reports which give insights into the quality of clusters and their product characteristics. This interface provides an easy way for any user to run the clustering model and change key parameters.The output of the model is exported to Tableau where the key metrics are highlighted and geographic distribution of stores and their clusters are visualized.
To solve the problem, the Compass Analytics team utilized Alteryx Analytical Apps to create a set of chained apps to first clean the data into a standard format and another set of Analytical Apps to run a clustering model using Sci-kit Learn in Python.For the first set of analytical apps the team developed a user interface for business users to select the source of data and specific columns to use to standardize all datasets into a standard format. Validation steps were built in to make sure that a final dataset can be used to perform the clustering model. This interface provides an intuitive way for any user to standardize the data.For the second set of analytical apps, the team developed a user interface to select the parameters to run the clustering model. The Alteryx workflow leverages the Python Tool to implement the clustering model in Sci-Kit Learn and generates diagnostic reports which give insights into the quality of clusters and their product characteristics. This interface provides an easy way for any user to run the clustering model and change key parameters.The output of the model is exported to Tableau where the key metrics are highlighted and geographic distribution of stores and their clusters are visualized.
The solution built primarily in Alteryx provides a user interface for non-technical users to perform the data cleaning and set clustering model parameters to run the model. Therefore, the company can continue to gain value from machine learning if their team does not have any expertise in coding or Alteryx
The solution allows the company to rely on what the data says is most optimal instead of relying on individual expertise which would be lost if team members leave the organization. This allows for improved product segmentation and more optimal placement of products leading to greater revenue over time.
The solution provides a comprehensive report of the significance of modelling results and easy-to-understand visuals which help the business understand their product segments in a quick and concise format.