How to harness predictive marketing and machine learning to engage customers

By Steven Ledgerwood, Managing Director, UK at Emarsys.

This article is brought to you by Retail Technology Review: How to harness predictive marketing and machine learning to engage customers.

Predictions are notoriously difficult. When the CIA created a report fifteen years ago on what to expect in 2015, most of its estimations were wrong. Whilst the organisation foresaw the growth of the Internet, the 2008 financial crisis was not on its radar.

So if we can't rely on political estimates, social predictions or even the weather, is there anything we can predict with confidence?

Consider human behaviour

It might seem to be a stretch that we as humans are more predictable than the natural world, but aside from irrational exceptions we are generally creatures of habit. Over a sustained period of time, we tend to eat the same food, wear the same clothes, watch the same movies and shop with the same brands.

Now we have access to analytical, algorithmic and diagnostic technologies, which are filtering through to business and down to retailing to capture such behaviour. As a result, many marketers are implementing tools that predict human behaviour commercially.

If you are a retailer with a vast customer database, gathering information from human behaviour offers significant opportunities for growth. Even the most individual and random shopping habits can be logged, stored and analysed. Creating real-time, personalised content is now possible in ways that would have only been dreamt of 50 years ago.

Automatically predict for profitability

Having advanced capabilities such as the above is becoming increasingly essential to attract, engage and retain customers. Retailers know from the rapid evolution of ecommerce that if they are to succeed, they have to keep customers engaged across every channel and device. Data has become a crucial element in this battle to engage and retain.

Using data efficiently is imperative. For example, creating automated messages with that data, which are triggered in response to customer activity, including: shopping basket reminders to address cart abandonment, and product recommendations based on a customer's known likes and dislikes.

Behind these triggers is a complex statistical analysis and machine learning software that processes diverse behavioural data including page views, check-outs, add-to-basket events and search queries on a website. Customer interactions with thousands of products are constantly processed, giving real-time, individual recommendations with every page refresh.

Solutions are continuously testing and updating, establishing a database of known customer behaviour so that it can be predicted in the future.

Underlying trends

Expectedly, this involves a lot of data, and currently marketing platforms still require an element of human interaction. Segments must be built, offers created and timings set.

It's unlikely that the human element will be abandoned any time soon, but retailers can begin to look beyond the limits of human knowledge by adopting a more progressive perspective on customer intelligence.

Meeting customer demands, competing online and in the high street, and incorporating new channels are massive challenges to retailers, therefore the benefits of improved customer intelligence data are many. They include:

• Having the ability to build a unified customer profile that spans across channels and devices
• Enabling the retailer to gain a better understanding of the customer
• Supporting the promotion of products and offers that will trigger a positive response in customers, leading to higher engagement
• Lifting a retailer's strategies, expanding conceptual thinking and boosting the success of marketing programmes

One point to bear in mind is that each retail operation is different as is each customer, so recommendation models should be designed specifically for each respective stage of the buying process (from research and discovery, basket purchases, to post-purchase). These models also need to be tailored to take into account channel-related behaviour.

Only genuine behaviour is captured by the most advanced machine learning solutions which use complex learning algorithms to filter out irregular online activity (i.e. with automated bots - only genuine human behaviour is captured). And for those retailers with large or regularly updated catalogues, these algorithms can also deal with long-tail items involving little or no behavioural traffic.

Limitations aside, while it's still not possible to predict future events, predicting customer actions based on their behaviour and habits is rapidly becoming the reality of retail marketing today.