Smarter Demand Forecasting That Improved Sales Margins
Daily sales forecasts for every product and store, for a US commerce and logistics company
About the customer
Our client is a US commerce and logistics company that helps B2C and B2B businesses sell and fulfill orders across multiple channels. We built an inventory and demand forecasting module that predicts sales for each stock keeping unit (SKU), the unique code for each product, using historical data stored in the client’s data warehouse.
What needed solving
Unsatisfactory forecasts
The existing sales forecasting pipeline wasn’t delivering satisfactory results, impacting sales margins and operational efficiency
Every SKU, every store
Sales had to be predicted for many SKUs across different stores
Large data volumes
The solution had to process large volumes of data and still provide accurate predictions
Existing infrastructure
Predictions had to integrate smoothly into the client’s existing infrastructure
How we built it
Data connection
Connected our Python environment to the data warehouse and linked each SKU to its store.
Data analysis
Explored the data to understand how the different features relate to each other.
Model selection
Tested several models and chose Prophet, an open-source forecasting model from Meta (originally released as Facebook Prophet), after multiple iterations.
Daily forecasting pipeline
Built on Kubeflow, it runs every day. Each day it forecasts sales for each SKU for the next 7, 14, 30 and 90 days, for all paying customers.
Pipeline monitoring
A monitoring service keeps the pipeline running every day without downtime, so the results stay reliable.
Better forecasts and higher sales margins, at scale
Technologies
Domains
Have a Query? Let's Have a Conversation
Do you have a project in mind? We’d love to know what it is. Email us or send a message from our contact page.
Have a similar problem?
Tell us what you are trying to solve and our team will get back to you.