10 Point Plan: Using Predictive Analytics To Improve Customer Lifetime Value

by Matthew Kelleher RedEye

Matthew Kelleher, CCO of RedEye, offers advice on how to implement predictive analytics to improve customer lifetime value.

1. Have A Sound Strategy And Identified Business Goals

I’ve always found strategic planning to be easiest when you define where you want to get to and work back from there – planning your journey is much easier if you know where you want to be. You need to ask yourself, what does success look like? This sounds very simplistic, but is effective.

RedEye started its journey to embed predictive analytics at the heart of its marketing automation platform knowing exactly what we wanted to do. We wanted to use predictive to target the key stages of the customer lifecycle. What did success look like to us? Making significant improvement to the Customer Lifetime Value. This had strategic value for us because increasing the customer value is the end goal for many marketers

2. Get Committed Ownership At A Senior Level

Nothing is going to fly for very long without senior ownership to smooth out the hurdles you face. The longer term the project, the more backing and focus you may need. The easiest way to gather senior backing is to be able to show real strategic value in what you want to do. Anyone trying to garner support of a Predictive Analytics project currently is in luck, on two fronts.

Firstly, it is an incredibly topical subject, for a lot of businesses there is greater danger in not investing in analytics, predictive especially. Secondly, and more importantly for many businesses, Predicative Analytics can have a fundamental impact – improving metrics and delivering tangible results. Getting backing, based on clear business benefits, is key.

3. Test Your Hypothesis

Critical to gaining senior support, is being able to show working examples of both process and outcome. Tests provide better gauges for the amount of resource required and likely outcomes as well as providing pointers for what, and what not, to do. Testing reduces risk as well as being far more likely to convince senior sponsors to get behind the project.

At RedEye, we have worked on using engagement patterns in customer data to predict outcomes. For example, does customer engagement with a brand’s marketing indicate future outcomes like purchase? We worked with Hotel Chocolat and were able to show improvements in customer retention as well as, critically, an increase in revenue of 25 per cent. These types of test results guarantee support for your project throughout your business!

4. Get Your Data Ready

It’s all about the data. You see this title everywhere and when writing a 10-point plan I feel guilty about putting this fourth, not first, even though I’m trying to write in chronological order and not as an indicator of importance! For a lot of organisations, creating the data set on which predictive analytics can be built and developed is the number one hurdle. Lots of issues come up when we are trying to build client data sets, such as, and in no particular order: understanding what data types are required; internal data silos; ability to create a single persistent record if combining customer or individual level data; and business priorities.

This last one is often a key point of failure and dispute and where senior level support really comes in useful, helping prioritise requirements and avoiding your project sinking in the bog of business (and more often) IT priority lists.

5. Data Doesn’t Sleep

And it certainly doesn’t stand still. It changes, updates, grows. Does your plan account for this? Not simply in where you hold the data and hopefully update and archive, but critically, does your model learn? Before Machine Learning, models would be built and immediately start to degrade, requiring on-going and expensive rebuilds. Today the issues are different but still require constant attention to hygiene and archiving.

6. Building Your Predictive Analytics

These days a standard database has thousands of tables with millions of variables. Combining that with the numerous predictive algorithms readily available, or those that can be developed, can be a perfect recipe for chaos. This is where foundations play a huge role. Your descriptive and diagnostic analytics can point you towards the most predictive variables.

I’m reliably informed by our Head of Insight that standard predictive algorithms have moved on from the straight forward linear regressions to more complex tree-based models. Whatever you choose, the outputs, the balance between right and wrong predictions should play a huge role in the model that gets deployed into production.

7. Applying Machine Learning

Here’s where you stand on the shoulders of giants, leveraging the data storage capabilities that have been built over the past decade. Descriptive and diagnostic analytics will only ever give you a starting point. Machine learning algorithms can plough through the massive volumes of data that you have stored and pick out the really juicy key variables which can ensure you have the most accurate predictions in the shortest space of time.

Pivotally, machine learning also makes sure your models are learning from changing customer behaviours and forming the crucial link between your model building and deployment tools.

8. Real Time Processing Capability

Where most analytics implementations fall through the cracks is the lack of real time processing. If you have a prospect in a prospect conversion model who’s made a purchase, they should automatically fall out of the model. To avoid sending them a communication that undoes all the great work your careful prediction, segmentation, and personalisation has achieved, every interaction of an individual in your models should be considered at every re-score or re-build.

9. Integration Is Key To Reduce Complexity

Or to put it another way, actionability – which isn’t really a word but gets the point across. How are the outcomes of your predicative analytics going to be applied? The key is to avoid manual processing of data into application platforms, which increases complexity or introduces data latency issues. Tools such as AI and Automation allow you to act. RedEye is a ‘marketing automation’ business, therefore the analytics is baked into the platform. The outcome, marketing communications, are automated, ready to respond when an individual shows the behaviour allied to one of the identified outcomes. But for other organisations building powerful models is only half the story – ensuring ease of application for long term success is just as important.

10. Analytics On Its Own Does Not Deliver Results

Marketers deliver results, not models, data, or systems. Predictive Analytics delivers data and outcomes, but it takes great marketers to turn that data into insights and campaigns that succeed. This should never be overlooked. The role of Predictive Analytics is to let the data take the guess work out of audience selection. It allows the marketer to do what they are good at, crafting brilliant treatments that influence people to buy more. Ultimately, this is what we all want for our businesses and clients!