No Black Boxes

Last week we published our second chapter in our blog post series on an Enterprise Customer Data Platform. The second chapter was on Freedom & Flexibility. In this chapter we talked about ensuring marketing gets access to the data needed to make decisions and change course on the fly.

The first chapter was on managing 1% of your data and why it’s challenging to access the other 99% of your data.

The third chapter is on AI & Behavior, read on to learn more.

Chapter 3: AI & Behavior

Thou shalt ingest & execute any model – no black boxes

AI is a popular topic these days. Everyone likes to talk about it; Marketers want to find out how AI can help them; and most Marketing tech vendors claim some form of AI. I’ve been thinking about AI since my days at the Stanford PhD program and I had the opportunity to interact with the world’s top experts and pioneers at the space. There were two takeaways I got from these interactions:

First, algorithms are useless without data. In fact, Google’s head of AI famously said that more data beats better algorithms. What this means for a Marketer is that unless you have all your customer data connected, in one place, even the best AI algorithms will fail to give you any results.

Second, there’s no such thing as a model that works across companies. All these vendors that sell “out of the box models and AI” are essentially selling models & AI that works for someone else’s business – but certainly not yours! What will happen in those cases is that “out of the box” will turn into a long and expensive consulting project, as the vendor will desperately try to fit into your business and your customers business a model that was developed for different customers and products. This is not an Enterprise approach.

So what do Enterprise CDPs do here? They recognize that models can come from many places – you may have data scientists in place; and there is a large number of consulting & product vendors that specialize in just AI & models. They key is

• giving easy access to all customer data so that these models can work
• once the model works, putting it in the hands of the marketers (with a powerful automation framework) so that they can be deployed in multiple use cases.

ActionIQ does exactly that. It can take a model that predicts, for example, whether a customer is “full price” or “discount”; apply it to billions of behavioral events so that it performs super well; and then help deploy that model in any campaign where that intelligence everything is automated, powerful and transparent!

A last point on “out of the box” models. We call these models “black box” because you never know how they work, nor can you influence them. That’s a massive problem in an enterprise environment. Consider for instance a product recommendation model. Perhaps the algorithm recommends a product that’s low on inventory. Can you suppress those recommendations easily? If the model doesn’t work, will you be able to tell why? Or will you just lose trust because you’re never sure when the model works or doesn’t? Enterprise CDPs make sure that you can both tune a model to your business needs and, when it doesn’t work, understand why it didn’t work.

AI in the E-CDP model does not replace the marketer. But it puts them in the driver’s seat, like sitting on the cockpit of a powerful fighter jet. The Marketer sets the strategy and evaluates the results; the ECDP takes care of the rest.

Thou shalt trigger & personalize based on journeys, not events or products

When Marketers ask me what are low hanging fruits for AI, two promising areas are triggers and personalization. There are many products in the Market that promise trigger events (emails, mobile messages) and product personalization (people that bought this bought that, Amazon-style). These approaches are not bad but they’re outdated. The key challenge with them is that they rely on an explicit action (a customer abandons their cart or a web session) or on a relationship between products (people that buy beer also buy diapers). What’s the problem with that? That neither are influenced by the customer journey. Triggers & Personalization based on journeys has immense power that Marketers are starting to realize.

Here’s an example: I recently bought a Travel guide about Spain to prepare for an upcoming vacation there. A month later, Amazon kept recommending more travel books about Spain! I’m sure some people buy two or three guides together, so it’s easy to see why the recommendation algorithm made that choice; but if the recommendation considered my customer journey, it would have moved to recommend me products that may be useful for my trip besides a book. It would have recommended new suitcases; travel-size bath products like a folding toothbrush; a new phone with a great camera; a book about learning how to speak Spanish; etc.

Not only would the recommendations be more useful, but it would broaden the relationship I have with Amazon beyond books. In other words, recommendations would include ‘gateway products’ that increase not only sales but the Lifetime Value of your customers.

Now you can see why triggers tend to be limited. It’s easy to send a reminder email when someone abandons their cart. But what are the behavioral patterns that are special and unique in your customer base that imply an interest to buy a product? That’s a much broader and more powerful approach that again increases not only sales but also engagement, loyalty and lifetime value.

An Enterprise CDP is the core technology that can help activate and automate such models. In addition, with a modern ECDP it’s easy to start because all the customer journey data are in your fingertips; so low hanging fruits abound. ActionIQ’s automation and powerful UI means you can keep testing new approaches and increasing the sophistication of your Marketing AI even with the team that you have today.