E-commerce companies are constantly seeking innovative ways to drive revenue growth. One of the most common ways to do this is by running marketing campaigns. Companies use targeted discounts to increase their customers’ Lifetime Value (LTV) by using a variety of mediums like E-Mail (EM), Social Media, and Direct Mail (DM). However, determining which customers to target with campaigns is often time consuming, manual, and based on messy data.
To this end, one of our recent projects was to develop an Artificial Intelligence (AI) solution that leverages Machine Learning to help a client identify how to optimize their Marketing Mail campaigns to maximize their ROI.
The purpose of marketing campaigns is to maximize the ROI. To accomplish this, strategists must answer three important questions:
- Which customers should the Ads be sent to?
- How many customers should be targeted?
- What discount rate, if any, should be offered?
The answers to these questions are often difficult to compute because they use complicated predictions of marketing spend, revenue, and customer behaviour. Many times, companies will base their decisions on manual calculations that are prone to error. In other cases, companies will simply target their most valued customers.
Choosing the Right Measuring Stick
The biggest challenge in evaluating the performance of a promotional campaign is the ability to measure its ROI. At a first glance it seems simple, we would look at the revenue generated as a direct result of the campaign, and subtract the cost of the product and derive the profit or loss.
In practice, our clients found this approach to have serious flaws. The initial bump in revenue incentivized by a hefty discount is often followed by slump in sales by the same customers in the months to come. Old school segmentation methods like the RFM model target the shop’s best customers, offer them a hefty discount, and rejoice at the immediate returns. When we consider that these loyal customers may have made the same purchase without the discount, we see that the measuring stick should be Uplift instead of Revenue.
Revenue ✘ How much Gross Revenue was generated by customers who used my Discount Code or Promotion?
Uplift ✔ What will be the future Lifetime Value (LTV) of my customers who used my Discount Code, and what would it have been had they not received the discount.
We worked with our client to address their challenge and provided an AI backed solution. The implemented solution included the following features:
- A Machine Learning model to identify which customers to target with marketing Ads to maximize ROI
- A reporting tool to analyze the results of historic campaigns
Machine Learning Approach
We developed a Python Machine Learning algorithm to identify who to target in future marketing campaigns. We first used the model to identify how customer behavioural attributes like Recency, Frequency, Value (and many others) related to the potential for a customer to convert.
We ran hundreds of simulation of different Machine Learning Models and allowed the algorithm to choose the most suitable Model. The model used these simulation, in conjunction with our input coefficients to determine which customers a promotion should be sent to maximize conversions and Uplift ✔
The below chart summarizes the features the Machine Learning model identified as having the most significant importance for predicting a customer’s conversion in a future marketing campaign using a Random Forest Model
The first and most important feature is a customer’s spend in the past 6 months, which considers recency and monetary value. Another important feature is Historical Discounts. Despite letting the Machine make the predictions, the outputs passed the eyeball test! We had expected that customers that had been sensitive to discounts historically are more likely to respond to future marketing Discounts – this turned out to be true.
An AI Solution
The below report is one of the tools we built for the client. By choosing parameters of Month of Year, List Size, and Discount Rate, the report generates a forecast and the optimal list of customers to send marketing ads to in order to maximize Uplift ✔