Three things every CMO should know about self-learning AI

By now, most marketers have heard of self-learning AI.But what is it, really? And why is it so important?

1. It automates experimentation.

Marketers do a lot of tinkering. There are many cases when the only way to know what works is to try a bunch of different options (A/B tests) and learn from the results.

For example:

• What’s the right time to send emails?

• Which channel is best for communicating with different customers?

• What landing page will give a better conversion rate?

• What bidding strategy will give the best results with social media ads?

Anyone who’s done this knows it’s very tedious. It’s even harder because of the different overlapping variables: Is the click rate on that email low because of the copy? Or did we send it at a bad time? Or did we target it to the wrong people?

Self-learning AI exists to address this problem. A self-learning AI ‘agent’ selects actions and then learns from the results. Over time, it gets better and better. It automates the whole process of experimenting to discover what works, and then smoothly transitioning from experimenting to optimizing. Put another way, the agent navigates the ‘exploration / exploitation tradeoff.’

This helps marketers in three ways:

1. Saves time (no need to be constantly designing and analyzing A/B tests)

2. Improves marketing performance (since the self-learning AI will be able to discover more patterns than an ‘unaided’ human)

3. Gets there faster (since the ‘agent’ is designed to experiment as efficiently as possible — i.e., to learn as much as possible from the smallest number of data points)'

2. It’s different from traditional Machine Learning.

When most people hear that self-learning AI actively experiments to discover patterns, they think: ‘Isn’t that the same as any Machine Learning?’

It’s not. The most common type of Machine Learning is supervised learning. It works like this:

1. Start with a dataset with a ‘label’ for each item (e.g., photos of animals, each labeled with ‘cat’ or ‘dog’)

2. Train a model on this labeled dataset to predict the labels from the data

3. Use the model on new (unlabeled) data — e.g., given a brand-new photo of an animal, have the model say whether it’s a cat or a dog

As you can see, there’s no experimentation here — the model doesn’t actively build its own dataset; it just does pattern recognition on existing data.

So does that mean self-learning AI is better than supervised learning? Not at all — it’s just useful in different situations. Self-learning AI is best when you need to learn patterns through active experimentation and this experimentation is too costly to do through a huge number of random actions.

A great example of a marketing application that does not benefit from self-learning AI is predicting customer churn. Here, we aren’t trying to pick a marketing action, so there’s no experimentation involved — we just need to recognize based on the data which customers are likely to churn. Supervised learning is the tool of choice in this situation.

On the other hand, marketing is full of examples where experimentation is needed and we don’t have the luxury of taking a vast number of totally random actions to build a historical dataset. In cases like this, self-learning AI is the best solution:

• Selecting the best landing pages (including personalizing them to individual customers)

• Optimizing ad bidding strategy

• Determining the profit-maximizing promotional offer to send to each customer (that’s what OfferFit does!)

3. It requires specialist data scientists.

While self-learning AI can be an incredibly powerful tool, not everyone is equipped to do it right. Data scientists who use self-learning AI need to know the tools of supervised learning, but they also need a set of advanced techniques specific to self-learning.

This knowledge is still pretty rare: while self-learning AI has made big waves in academia, practical applications have only started appearing in the last couple of years.

For companies using self-learning AI, that means they need to carefully vet products and partners to make sure they really know their stuff. Few do.

What does this all mean for marketing leaders?

First, self-learning AI is not an empty buzzword. It’s a new, powerful technique that will save marketers huge amounts of time and unlock higher levels of performance.

However, it’s not trivial to get it right. Savvy marketers will will take great care to identify the tools built by genuine experts. They’ll also move quickly to get a head start on the competition — after all, self-learning AI is on its way to becoming a must-have in the marketing toolkit.

About us. OfferFit’s software uses self-learning AI to personalize promotional offers.