Predicting Customer Rebate Rate

By Matt Curtis
Nov 08, 2023

A Heavy Industrial Blew Its Discount Projection by a Cool Billion Dollars, Whoops

Customer Challenge

During the previous year, a heavy industrials company missed the planned customer discount by about a billion dollars.  The biggest cause of this failed forecast was a late in the year deal, supporting a large public-works project.  Furthermore, only about 1/3 of the products were being tracked, globally.  While the remaining products were smaller revenue generators, they did pose a blind spot in the company’s analysis.  

Leadership wanted a better, more accurate way to forecast the customer discount rate across all of their products.  

Solution

The customer wanted an automated method of tracking the discount rate, but they also didn’t want to completely ditch the system that had worked for them, all those years.  They wanted their forecasters to enrich the data in circumstances they felt would have a material effect on the discount rate.

  1. Design a Digital Twin of the Process – The process had three functional stages during a given year: 1) Sale Complete, Discount Data Known; 2) Sale Complete, Discount Data Unknown; 3) Sale and Discount Forecasted.
  1. Build the Query for Step 1: Sale Complete, Discount Data Known
  1. Build a Machine Learning Model to Predict Step 2: Sale Complete, Discount Data Unknown
  1. Build a Forecasting Engine for Step 3: Sale and Discount Data Forecasted
  1. Connect the Three Pieces of the Digital Twin, Add in Human Adjustment Code
  1. Fine Tune Across Products and Geographies
  1. Build a Visualization and Interface  

Major Insight

During the early phases of the project, we attempted to better understand and predict Step 2: Sale Complete, Discount Data Unknown.  I felt we needed a broader perspective in the data, and I wanted to add other product types to strengthen the data around the sales team supporting the operations.  Ultimately, the model we were using couldn’t support the expansion into multiple product sets.  

As we moved from the high-volume product sets to the low-volume product sets, we found a use for the code we had developed earlier.  There were large sets of products that didn’t have the volume to generate accurate predictions, but by combining many different product types, there was enough data to build a model in these smaller product sets.  While this snippet of code/ML model only supported about 5% of machines sold, it supported about 1/3 of the product sets sold.  It was the missing piece to be able forecast all ~500 product sets.  

Results & Value Generated

The model did very well.  In most months, it beat the current process.  In the situations when it didn’t, it was very close.  The real difference between the model and the current process came in the early months of the calendar year.  The forecasters felt pressure to give the sales team the benefit of the doubt, when everyone knew that the discount rate would float back to the historical average.  The model knew better, and it didn’t react to those same pressures.  

The project had an estimated $30MM ROI, but it also spawned several other projects looking into how the company could improve its pricing process and better track the rebates it was handing out to customers.  

Reflection

I’ve felt this way about several projects before, but this project really would have been beneficial to execute a diagnostic analysis before building the model.  The leadership who hired us tried to solve a symptom and not a problem.  They had missed their forecast, thus they needed to fix their forecasting.  During the project, we learned that the inciting event for this project was a red herring, of sorts.  

The actual discount rate prediction was reasonably accurate.  The total amount of money discounted, translated from the discount rate, is where the forecast fell apart.  The KPI the company developed didn’t align with the bottom-line expectations in the P&L.  

The real problem was two-fold. The sales by unit forecast had underestimated the number of products that would ultimately be sold.  The sales team did what sales teams do, fight for more customer discounts to lock-in a sale.  As the company reached its sales goals, it didn’t change the controls on the discounts, leading to more sales, and higher total discount.  

A more programmatic approach would have identified the problem, created a targeted solution, and managed the company’s discount problem, holistically.  

Hungry for more?

Discover more ideas to improve your business

The Business is Changing. Your Leaders are Anxious, Too.

Effective change management for leaders acknowledges their fears, supports their unique roles, and enhances their skills, ensuring strategic alignment and long-term organizational sustainability.

Generative AI Framework: Prioritize & Execute

Embark on a transformative journey with our insightful blog on successful generative AI strategies for businesses. Learn how to effectively prioritize and execute AI initiatives, set measurable goals, and foster a skilled team to drive innovation and growth. Master generative AI and propel your business into a new era of success.

Digital Transformation Pitfalls: You Chose Software Poorly

Uncover the common traps that lead to wasted SaaS spending, emphasizing the importance of a meticulous approach. Learn how to assess current capabilities, conduct architecture reviews, capture requirements, and implement a structured software review process to make informed decisions that boost ROI and efficiency.

Book a Meeting & Find Balance with Reporting & Analytics
Let's make your vision a reality
Bring harmony to your business
Start Now