The Marketer’s Guide to Predictive Analytics

We break down the basics of predictive analytics, its best practices in the world of retail, and explain how to apply the technology behind this “buzzword” to the real world.

In the business world, it seems like the more people talk about something, the harder it is to cut through the jargon and understand just what they’re getting at. Predictive analytics is an incredibly powerful and useful technology that unfortunately has become yet another business buzzword, a term used so often and across so many contexts that it becomes hard to understand what it really means.

So let’s break it down: what is predictive analytics, and how do we use it in retail?

What Predictive Analytics Is

Predictive analytics takes everything you know about your customers to date and applies statistical models to reliably forecast what will happen next. It sounds a little bit like magic, but it’s actually just math — though admittedly, it’s math that’s so complex and multi-layered that it can seem like magic to the untrained eye.

For retailers, predictive analytics allows you to infer what individual customers, customer segments, and your entire customer population will do in the future based on the information you have about what they’ve done in the past. Machine learning and probabilistic modeling are used to look at dozens of different variables and factor them into these predictions, searching for the variables that most reliably and consistently predict future outcomes.

Compare that with historical analytics, which uses trends in your recorded data over time to predict outcomes. A business can use historical analytics to understand which methods have brought the greatest success over time and use that understanding to inform future decisions.

Do these definitions sound a little similar? While both predictive and historical analytics use past data to predict future outcomes, the key difference is the level of complexity behind the predictions that each approach generates.

Let’s say that you wanted to use historical analytics to generate a weather forecast, for instance. You might see that the temperature today is 75°, and that the temperature yesterday was 74°. The day before yesterday it was 73°, and the day before that was 72°. Historical analytics would visualize that trend and lead you to the reasonable prediction that tomorrow it will be 76°.

But of course, there are many, many more factors that are much better predictors of weather than just last week’s temperature. This analysis wouldn’t even account for what time of the year it is, let alone barometric pressure, warm and cold fronts, or cloud patterns. Predictive analytics would account for all these factors and run hundreds of simulations to find the weather conditions that are most likely to play out tomorrow. Similarly, it can use a huge wealth of different information about your customers and their buying behaviors to identify just how likely they are to make purchases, which products they’ll most likely purchase, and when they’re most likely to purchase them.

What Predictive Analytics Isn’t

As you can imagine, predictive analytics is a pretty amazing and useful tool that can help retail marketers focus all their resources on the right customers at the right times with the right messaging. But since people in the business world tend to talk about it in an imprecise way, there is a misconception that predictive analytics can turn all your raw customer data into relevant insights with the push of a button. The reality is that predictive analytics is a tool like any other, and requires some effort and skill to be put to use effectively.

Predictive analytics is only useful if the insights it’s surfacing about customer behavior are relevant to your organization, but different organizations will have different ideas about what’s relevant. That definition of “relevant” will differ even further according to what position in the organization the user fills, what campaign they’re currently working on, what channels they’re using, and so on. One thing predictive analytics can’t predict is what you’ll need it for — that’s something that only the end-user can decide.

Furthermore, predictive analytics can only work with what you give it. Your analytics platform has to have access to all your raw data in order to generate insights from it, which means that members of your marketing team have to pull information from your CRM, ESP, and any other relevant sources to get the results they want. Luckily, if you’ve got a platform like Custora, all that stuff can be seamlessly integrated into custom dashboards without the need for much manual copying, pasting, and uploading from your marketing team.

Now that you have a solid understanding of what predictive analytics can do for you — and what it needs from you to do it — we can look into some simple best practices to guide how you apply this technology to your marketing campaigns for real results.

1. Ensure the predictive capabilities are purpose-built to answer mission-critical questions for retailers.

The information you pull from predictive analytics should be relevant for retailers. As we’ve said, what qualifies as relevant depends on any number of factors, so be sure to take the time and care to outline exactly what kind of insights you hope to generate from your analytics engine.

Regardless of your organizational or personal needs, one thing that’s true across the board for retail is that your analytics should be focused on solving customer-centric problems. For example, you’re probably going to be interested in knowing exactly when a particular shopper will make their next purchase and what the value of that purchase will be. Predicting that can be nearly impossible without the help of predictive analytics, because these relationships are so complex that only machine learning techniques will find them.

2. Use collected data to identify predicted "high-value customers."

For a typical retailer, 50% of revenue is driven by just the top 10% of their customer base. You need to focus on and personalize this “high-value” segment in order to increase repeat purchase rate and cement loyalty.

Personalizing this segment has to go beyond what you already know about your average high-value customer. A predictive model should not only collect a large amount of data, but also dig deep to surface additional, unexpected insights into customer behavior and attributes. For example, your data might reveal a distinguishing behavior displayed by these MVP customers. By using this behavior to find lookalikes on Twitter and Facebook, you can produce insights on these shoppers that you never could have guessed from your CRM data alone.

3. Apply AI and machine learning to the real world.

You’re not using your top-of-the-line analytics engine to surface a list of quirky “fun facts” about your customers— you’re using it to look for new ways of improving campaign performance, keeping customers engaged, and increasing revenue. Keep your analytics focused on the real world.

Imagine you’re a retailer who’s worried about churn. You want to move the needle almost immediately after your high-value, loyal shoppers show the very first signs of slipping away. But how do you identify these customers, and how do you react quickly enough to keep their business?

Predictive analytics sounds great, and that’s because it is! Unfortunately, some marketers never get to see just how great it is because the data hasn’t been made accessible to them.

It’s a difficult question, because while there are a few obvious warning signs that a customer is slipping away, everybody has different ways of signalling they’re losing interest. But with predictive analytics, you don’t have to use a one-size-fits-all rule to identify customers at risk of churn. Your analytics can identify a baseline set of unique purchase tendencies for each high-value customer, then surface a segment of them every week who are veering from those patterns. Your marketers learn the early warning signs of churn, proactively send win-back messages, and take concrete steps to reduce attrition.

4. Data should be democraticmake sure your predictive analytics is accessible to the teams that use it.

Predictive analytics sounds great, and that’s because it is! Unfortunately, some marketers never get to see just how great it is because the data hasn’t been made accessible to them. Inaccessible data is useless data, and it represents a huge wasted opportunity for your brand.

That’s why any predictive analytics your marketing organization invests in needs to be configured in such a way that your marketers can easily access its data. That means not just that the insights it produces are understandable to individual marketers, but that they’re relevant, and can be turned into actionable strategies that are easy to read, execute, and measure. A great platform ensures that marketers within your organization have the ability to intuitively pull all the information they need from their analytics.

In Summary

The trouble with the buzzword-ification of predictive analytics isn’t that everybody’s excited about what it can do — we think that’s great, because it can do a lot of incredibly cool things! The problem is that the less clear people are on what predictive analytics really does and how it works, the more likely they’ll be flummoxed by it when it’s actually in their hands. We want marketers to leverage the full power of this technology, and to do that, they need to know what’s required of them to use it effectively.

Implementing predictive analytics into your everyday marketing activity isn’t a one-day process. It takes patience on the part of your team, some trial and error, and a willingness to learn new approaches to the age-old challenges of marketing. But the more you know about what to expect, the more quickly you’ll be able to apply today’s most advanced predictive technology to the work of keeping your customers happy and boosting your revenue.


Like this? You might also enjoy these.

predictive analytics

How I Predicted Pebble Backers with 98% Accuracy

The following is a guest post by R.M. Bryce, winner of the Pebble competition....
predictive analytics

How Bayesian Probability Models Can Make CLV ...

This is part three of a three-part series exploring ways to calculate CLV in a...
predictive analytics

How We Would Do It: Predicting Customer Pregnancy ...

There has been a lot of buzz recently about how Target was able to predict...

Big News: Custora's Next Chapter

  I am excited to share that today we announced the acquisition of Custora by ...

Swatting Buzzwords: 360-Degree View of the Customer

There are some terms in the retail marketing tech world that get tossed around...

Meet the Marketer: Kelsey Foy of Eloquii

In this ongoing series, we chatted with some of our customers to learn more...