In the fast-paced world of eCommerce, identifying those customers most deserving of your time can often mean the difference between success and failure. On top of that, the absence of contracts or established start and end dates can make predicting churn and estimating customer lifetime value a daunting task.
The typically lengthy intervals between purchases tends to add another layer of complexity.
For eCommerce managers, it’s crucial to understand that the success rate of selling to an existing customer hovers at around 60-70%. When selling to a new customer, this figure drops to a mere 5-20%. Despite these odds, U.S. companies are losing a staggering $136.8 billion annually due to avoidable consumer attrition.
Coming to grips with these challenges is likely a big part of your daily routine. In today’s blog, we'll demystify the leading approach to predicting churn and estimating eCommerce lifetime value. By the end, you'll have all the valuable insights you need to safeguard your most important customers and confidently navigate the uncertainties of the eCommerce world.
What Are the Opportunities?
Predictive analytics for churn and lifetime value is pertinent to all areas of marketing. Besides the vital insights provided by the analysis itself, it also unlocks a range of new marketing activities that improve your bottom line. You can expect to:
- Proactively Mitigate eCommerce Churn Risks: identify and address churn risks before they emerge, empowering you to take preventative measures to discourage users from defecting. Remember, acquiring a new customer can cost five times more than retaining an existing one, and increasing customer retention by just 5% can boost profits by an impressive 25-95%.
- Maximize Purchase Frequency: Identify those users most likely to buy again and those who have room to improve their frequency. This includes customers who have recently reduced their buying habits and those who frequently browse without purchasing. You can tailor your marketing initiatives to re-engage with these valuable consumers.
- Generate Incremental Revenue with Targeted Promotions: Leveraging CRM campaigns with bespoke promotions is a strategic intervention. You can incentivize high-value or price-sensitive users to return and make purchases. Achieve a better understanding of how to offer discounts to maximize ROI.
- Retarget High-Value Users: By focusing your retargeting campaigns on high-value users, you can concentrate your marketing resources on the customers with the greatest potential for long-term profitability.
- Optimize Acquisition Campaigns for Efficiency: By creating lookalike audiences of users with the highest projected lifetime value (pLTV) and optimizing campaigns towards pLTV, you can make your acquisition efforts more cost-effective. This approach also makes bidding more efficient, with value-based bidding leading to a 14% improvement in marketing performance.
Use “Buy ‘Til You Die Models” to Predict High-Value Customer Churn
For eCommerce managers seeking to more accurately forecast customer churn and lifetime value, so-called “Buy 'Til You Die” (BTYD) models can be highly effective.
These models consist of two main steps:
Step 1: Predicting Churn and Forecasting Transactions
The Beta-Geometric Negative Binomial Distribution (BG-NBD) model is employed to forecast churn and the number of transactions for each customer. This model scrutinizes the frequency of a customer’s of purchases and their recent activity.
The first step involves creating a Recency and Frequency profile for each customer. After some analysis and visualization, you can pinpoint customers most at risk of churning and proactively cater to their needs - significantly reducing the chance of losing valuable clients.
- The probability that a user will purchase again
- Expected number of future purchases
- Predicted number of purchases for new customers
Step 2: Predicting Customer Lifetime Value (CLV)
While the BG-NBD model is useful for predicting churn and transactional volume, it falls short of providing insights into the monetary aspect of a transaction. This is where the Gamma-Gamma model comes into play.
It works by estimating the average transactional value of a customer. By combining the output of Step 1 with the Gamma-Gamma model, you can determine the projected CLV for each customer.
- Projected CLV for each user
- Expected average spend transaction for each user
- Expected average spend for new customers
All the underlying models are open-source, which means that with the right underlying data and data science resources, you can easily implement them yourself.
By applying these predictive models to your eCommerce operation, you can make significantly better-informed decisions - driving more powerful growth, retention, and profitability for the business.