11436 SSO

When Predictive Analytics Crafts the Wrong Cross-Channel Experience

Mar 23, 2015

There are many pitfalls a company can run into when applying predictive analytics to its business. We frequently speak with customers who use predictive analytics for the purposes of outreach (via email,  push notifications, and voice calls), and recommendations (as in product recommendations), and we consistently see companies stumble because they try to solve the “wrong” business problem.

The primary reason: the company optimizes for a business outcome within a channel rather than optimizing for a business outcome across channels. It’s a mistake that’s easy to make, because most analytics vendors focus on channel-specific optimization. These include product recommendation vendors, email and marketing automation companies, media buying and display retargeting companies, and contact center and support apps (voice, chat, and ticketing, for example).

Activity data versus outcome data

To understand this issue in more detail, it’s important to examine what kind of data feeds a predictive model. You have activity data (mobile traffic and email opens, for example), which predicts an outcome, and outcome data, which includes online purchases, in-store purchases, and calls to customer service.

If the goal is to optimize across channels, the outcome data needs to come from a channel separate from the activity data.

Let’s look at an example of web recommendations for retail. Different retail business have different paths they’d like their customers to follow (we call these “customer journeys”). In some cases, providing recommendations that drive customers to transact online is desired (as is the case with Amazon). In other cases, specifically in luxury retail, it’s more important to get the customer to transact in store because it allows for up-selling the customer.

In the first case, the recommendation model should use the “online store transactions” as the input to the predictive model. In the second case, the recommendation model should use the  “in-store purchase transaction” as the input to the predictive model.

The challenge of cross-channel prediction

More often that not, predictive analytics vendors cannot solve the second case simply because they do not have access to outcome data on different digital channels. As a result, most marketing packages using predictive analytics simply are not effective at crafting seamless cross-channel experiences.

So if this is important, why aren’t marketing software vendors taking up the challenge to predict across channels? The simple answer: it’s really hard to do, technically speaking. Doing anything across channels often involves working with the enterprise customer’s data warehouse team (or, increasingly, it's Hadoop data lake). It requires deep skills in integrating disparate enterprise systems, and it requires patience—something that marketing teams don’t always have a lot of.

Apigee Insights and cross-channel analytics

We’ve worked hard to ensure that Apigee Insights, our big data predictive analytics platform, can integrate data from popular data warehouses, and work seamlessly on top of our customers’ enterprise Hadoop infrastructures. We provide customer journey analytics to help enterprises understand cross-channel analytics behavior.

Here’s an example. One of our customers (a luxury retailer) wanted to improve its email campaigns to drive traffic in store. To help accomplish this, we used a combination of web analytics traffic, email traffic, and in-store transactions to more effectively target customers with email campaigns to encourage them to visit a retail location. We’ve worked with the customer’s Hadoop architecture to seamlessly send cross-channel data (e.g. in-store purchases) to Insights, and integrate users’ emails into their various marketing systems. The result is an email targeting system that spams the customers less, and increases the customer’s propensity to come into the store.

We’d love to discuss these topics further with you—if you’ve got questions or ideas, please visit the Insights forum in the Apigee Community.

Photo: Marcus Hansson/Flickr

Scaling Microservices