What Machine Learning Can do for Your Retention Marketing ROI

Marketers are in need of a scientific approach to determine which customers to engage and how best to engage them.

Even with all the metrics and sophistication of today's marketing tools, marketing leaders often have a difficult time quantifying the ROI of a campaign. In many cases marketers still have to make an educated guess as to whether their retention marketing campaigns, in particular, could have produced more lift, or been less costly to execute, or both.

And while every marketer knows that retaining existing customers is cheaper than acquiring new ones, there are still hefty costs associated with many of the marketing tactics used to retain valuable customers.

Retention marketing that achieves high, measurable return on investment relies on two things: making sure that retention tactics are focused on the right customers and delivering to those "right" customers the right offers.

Consider the following scenarios:

  • A casino offers you a coupon for free slot play redeemable at your next visit.
  • A mobile operator offers you a new phone to sign on for another year.
  • A hotel chain offers you a room upgrade to book this month.

These types of offers are common for trying to increase engagement and boost retention. But when the focus is on return on investment, a number of questions come to mind:

Are retention offers being targeted to the right customers? And for each target customer are we determining the optimal offer?

When it comes to retention marketing, marketers often do too much. They cast their nets too wide targeting offers to customers who aren't at real risk of leaving. And they make offers that they don't need to make, giving deals, discounts, and bonuses to customers who eventually leave anyway, or who would have otherwise stayed.

Two issues: predicting churn and making the right offer

So what are the major issues preventing marketers from maximizing the return on their retention marketing investments?

1)They are unable to accurately predict who is going to leave and when. The reality is that accurately predicting which customers are at risk of churning is difficult even despite state-of-the-art churn modeling techniques. This is because traditional churn models rate customers with a "churn score" that depends on a highly manual and time-intensive process during which marketers create aggregate attributes in a data warehouse. Because of the way customers are represented with a traditional data warehouse approach, granular signals that indicate customers may leave are frequently lost. Furthermore, since marketers typically update customer churn scores only once or twice per month, customer signals that emerge faster than that aren't taken into account. For marketers this means having to execute retention offers based on inaccurate and untimely customer predictions.

2)They can't run enough tests to determine the optimal offer. What every marketer wants is the ability to determine for each customer in any given context the offer that achieves the perfect balance between effectiveness and efficiency. Yet testing and analyzing offers is a manual process, one that can take several months and is not capable of scaling to achieve true one-to-one personalization across a large customer base. As a result of the process, there are only so many variants that can be tested and many options are left unexplored.

At the same time, marketers tend to measure the success of campaigns according to response rate. The more people who respond to the offer, the more successful is the campaign. But even if the response rate is high, marketers don't know whether they could have offered something less, or provided a different type of message instead of an offer, or if certain offers are better for some customers and not others, or which offers were cannibalistic, or whether other offers could have resulted in more lift, etc.

A scientific approach to marketing using machine learning

What marketers need in order to realize more effective and more efficient retention marketing is a more scientific approach to determining which customers to engage and how best to engage them.

First, marketers need to look at better, more accurate ways to predict churn. New types of sequential churn models, for example, allow marketers to more accurately predict churn because they are based on a different, more granular behavioral representation of the customer, and can be continuously updated to reveal meaningful signals that prompt timely action. Increased performance lift results from better accuracy and the fact that marketing actions can be initiated with enough time to be effective.

Second, marketers need to explore more sophisticated personalization technologies. Consider the benefits of personalization that is powered by robust experimentation capabilities combined with machine learning. Thousands of different offers can be tested across many different contexts via controlled experiments. Statistical measurement can then determine the impact of each one on 60 or 90 day retention, for example, rather than just response rate. The benefit of this approach is that the machine learning recursively tests and discovers the best methods of keeping customers by detecting nuanced individual behavior patterns around purchases, consumption, and other engagement signals, both obvious and non-obvious, at massive scale.

For marketers this eliminates the process of manually developing targeting rules based on guesswork, and replaces it with machine automated personalization that's based off of science-based decisioning. Essentially the machine automatically discovers and executes optimal targeting that drives both marketing efficiency and effectiveness simultaneously.

Efficiency + loyalty increases ROI

The key to keeping customers engaged and building customer loyalty rests largely on being able to deliver a personalized customer experience. Providing relevant offers and messages that are tailored to the needs of individuals translates into goodwill toward your brand.

But success is not performance lift at any cost.

To achieve exceptional return on retention marketing investment, marketers must have the ability to continually test and fine-tune at massive scale, and measure the impact of what is being executed and spent to save customers.

In past years, personalization for the purpose of saving customers was impossible at large scale, but today companies have the technology to get to true one-to-one personalization for millions of customers. Machine automated technologies exist for not only monitoring and analyzing dynamic behavioral data to reveal important predictive signals but also determining among thousands of options how best to engage a customer to provide a better customer experience.

Customer retention is an investment, and to maximize their return on that investment, marketers need to champion the need for data-driven technologies that make use of machine learning to drive better results.

Lara Albert is VP of Global Marketing at Amplero, a self-learning personalization platform built by Globys.