1 min read

Machine learning improves sales forecast by 30% for orthopaedics supplier

Adam Hadley 2 January, 2023

CASE STUDY PRIVATE EQUITY SUPPLY CHAIN

Profile

PE-backed orthopaedics and healthcare product supplier serving hospitals in a range of international markets.

 

Situation

A PE-backed prosthetics company wanted more accurate forecasts to improve the performance of their supply chain. They business wanted to leverage historic performance trends to improve forecasting accuracy of future sales on automated basis.

 

Action

Our approach to generating the forecasts was founded on developing a deep understanding of sales patterns historically through exploratory analysis of the business’s data. We were able to establish strong underlying seasonal patterns that were unique to each market and product that the business had not previously been aware of. Utilising these insights, we then progressed to utilise a seasonality-based forecasting approach. A machine learning algorithm was utilised to decompose historic sales into seasonal patterns, growth trends, and a ‘randomness’ value. These could then be extrapolated to predict the likely future sales up to 18 months in advance. We also deployed a simulation methodology to generate confidence intervals for the forecasts.

Once the model was developed, we consolidated the logic into a user-friendly, robust, and lightweight tool that could easily be handed over the client for on-going use.

 

Tools and techniques used in this work

  • Python

  • Machine Learning Forecast Algorithms

 

Impact

The model displayed significantly improved performance compared to the business’s existing forecasts, reducing the absolute error by 31% and providing a much longer-term forecasting view with minimal effort required to generate.

 

So what?

Our focus on understanding the business dynamics as a priority allowed us to unearth new insights that could be utilised to drive value for the business. Improved forecast accuracy ensured that stock levels were optimised with minimal wastage, and that team members did not have to dedicate high levels of resource to forecast generation.