Four Ways Ecommerce Marketers Can Start Using AI Today

From Facebook’s (slightly spooky) facial recognition functionality “DeepFace” to Amazon’s intelligent personal assistant Alexa, over the past couple of years artificial intelligence (AI) has become ubiquitous—part of what many would consider everyday life.

… Or has it?

Whilst AI has become commonplace among technology giants like Google, Instagram and Microsoft, for most marketers it’s still uncharted territory—something encountered regularly outside of work (e.g. Netflix recommendations), but rarely inside.

This is mainly down to a lack of knowledge and transparency; historically, AI has been something only talked about by experts in the computer science field, and consequently the language surrounding it is (even now) pretty arcane and, at first glance, intimidating.

To get to grips with the lingo, download our ebook: “A no-nonsense guide to ecommerce marketing”, here. 

But this is all changing. AI is becoming accessible to almost all industries—especially ecommerce marketing.  Listed below are four ways your brand could be using AI to augment its marketing right now.

Incentive prediction

Artificial intelligence enables retailers to send the right incentive to the right person at the right time—be that 20% per cent off for a new lead, free shipping for a customer about to churn or no incentive at all for the VIPs who’ll shop anyway.

How? Machine learning.

Machine learning algorithms can be divided into two types:

1. Supervised 
2. Unsupervised

The type you choose will depend on the level of control you wish to have over the incentives being dispensed.

If you have zero preconceived ideas about the number or value of discounts you want dispensed, you can use what we call unsupervised learning: a machine learning algorithm that draws conclusions from “unlabeled” data (in other words, has not been labelled by a human being) to make decisions about what to send to who.

However, if you as a marketer want to set specific boundaries (e.g. the gap of incentives can be from X-Y per cent), a supervised learning algorithm will give you the opportunity to do that.

A third type of machine learning is reinforcement learning, where an algorithm isn’t told which actions to take “but must discover which (ones) yield the most reward”. So in this case, an incentive would be optimised in real-time depending on the data available.

For a more in-depth analysis of machine learning, check out pp.7-12 of our ebook. 

Predictive lifecycle stages

At the moment, the most popular way of calculating where a customer is in the customer lifecycle journey is by using the RFM model: recency, frequency and monetary value.

AI is making this process much easier and much more efficient.

Using engagement data from a wide range of sources—purchase history, browsing behaviour, engagement with marketing messages, etc—a machine learning algorithm (explained above) can predict whether a customer is:

• At risk of churning
• Likely to only be a one-off customer, or a repeat purchaser
• Likely to become a VIP

This makes it easy to trigger real-time marketing actions; whether that’s sending an email or activating a social ad (or indeed both in a cross-channel campaign).

Category and brand affinity

Using indicators such as a customer’s past purchase, recent browsing habits, and inter-category and inter-brand affinity, AI can also create a “taste profile” per customer.

An example of this in action could be a multi-brand retailer such as Net-a-Porter sending out a newsletter, personalising each recipient’s subject line by mentioning a brand they’re most likely to be interested in.

Another example? Using taste profiling to automatically create segments which are likely to have the highest engagement with, for example, a product line, instead of having to use manual rules to create said segments.

Product recommendations

AI powered algorithms can discover the best products to display in front of each individual customer, all at scale.

How? By rapidly processing huge data-sets (such as recent and past behaviour), and looking at successful recommendations for people displaying similar customer traits (a process known as collaborative filtering) to extract a much deeper level of customer insight.

Here’s a snapshot from the “Profile based recommendations” option within the Ometria platform:

Conclusion

All of the above can be actioned by marketers today, as long as they have the following two things:

• Data: A bit like a professor needs books, a machine learning algorithm needs data in order to actually work its magic.
• Technology stack: As amazing as it would be to be able to do all of this very cool stuff all on your own, unfortunately it’s not that simple. In order to get the algorithms flowing and predictive messages sending, the majority of ecommerce marketers will need to invest in a platform (such as Ometria) where everything is laid out for you, ready and waiting.