Increasingly, marketers are embracing the opportunities that artificial intelligence affords by processing data at scale to make better decisions.

Econsultancy’s latest AI, Machine Learning and Predictive Analytics Best Practice Guide delves deep into the key areas involved in the adoption of these technologies, including the actions marketers need to take before applying them, how to apply them effectively, and how to avoid any unintended side effects.

What is predictive analytics?

Predictive analytics does not describe a specific approach. As the Best Practice Guide states, “At its simplest level, drawing a trend line (or regression) through some historical sales data to forecast next year’s sales is an example of predictive analytics using a mathematical model.”

There is lots of room for confusion or nuance here. “There has always been a relationship or overlap between predictive analytics and aspects of AI and/or ML. For example, neural networks (which have a clear element of self-learning) have been part of the predictive analytics toolbox for decades, but not necessarily promoted as “AI” in their usage by customer analytics teams.”

Predictive analytics uses machine learning or statistics to predict the future of anything from sales trends to patterns in consumer engagement.

In marketing, predictive analytics can be applied across a range of different touchpoints, from initial brand awareness to post-purchase activity. It can help more efficiently and accurately predict behaviour along the customer journey by drawing on historical browsing data and the profiles of other users.

As a result, the implementation of this technology can have a positive effect on metrics like conversion rate as it allows marketers to target users with more personalised and relevant content based on their likelihood to take certain actions.

Below, I’ve outlined some ways predictive analytics are used in a marketing setting and how it can help improve the performance of key metrics, alongside some real-life examples.

Recommendation engines

The use of predictive analytics is most often recognised by general consumers in ecommerce settings, typically by suggesting products to those browsing an online store.

Recommendation engines use algorithms to determine purchase patterns among likeminded customers to help them find products according to their browsing behaviour and other contextual factors. A recent Salesforce report suggests predictive recommendations influence on average 26.34% of all orders placed.

Obvious examples of brands that use recommendation engines effectively include online marketplaces like Amazon and popular fashion brands like ASOS and Boohoo. In fact, many well-designed ecommerce websites have now embraced this feature, although some of these engines are clearly informed by better quality data than others.

Amazon recommendationsASOS recommendations

Amazon and ASOS suggest products based on my browsing and purchase history.

There are three primary ways in which recommendation engines can be programmed to serve these personalised suggestions, as outlined in Econsultancy’s AI, Machine Learning and Predictive Analytics Best Practice Guide:

  • Collaborative filtering algorithms work by looking at the purchase and/or ratings history of the user and generating recommendations based on a few customers who are most similar to the user, disregarding any items in the set that they may have already purchased.
  • Cluster models work by dividing the customer base into a number of segments and then classifying users into segments that contain the most similar customers. The purchase and ratings history of the users in the segment are then used to generate recommendations.
  • Search-based models build keyword, category, and author indexes but do not work well in recommendation at scale if the business has customers with large numbers of purchases and/or ratings.

Similar methods can be used for content-based services like streaming platforms. Netflix uses a complex set of data about its users to serve them suggestions for what to watch next, rather than having them browse thousands of titles manually. This is then personalised down to the smallest detail such as the type of static imagery used that algorithm predicts will appeal the most to a user’s specific tastes.

Recommendation engines can be an invaluable tool for brands looking to improve customer engagement, conversion rate and average basket size by way of upselling or offering alternatives based on items that have already been viewed or purchased. It can also keep customers consuming content for longer in the case of streaming platforms – behaviour that is then fed back into the algorithm.

However, with this technology comes the risk of over-personalisation, in which customers begin to make very specific purchases based on suggestions but the algorithm fails to help them discover new products they may not have considered. In these cases, overall conversion rate can rise but basket size can stagnate or decrease over time, making it important for marketers to identify this as soon as possible if/when it happens.

AI, Machine Learning and Predictive Analytics Best Practice Guide

Targeting offers

Predictive analytics has many uses in online advertising, ensuring ad budgets are well spent by predicting those who are most likely to engage and targeting them accordingly.

Many brands target offers to consumers that are most likely to make use of them. In this case, the data used to do so can often be found in loyalty accounts, where basic information about a customers’ age, gender and address are usually stored.

Browsing and purchasing while logged in, or using a loyalty card in store, helps brands to determine more granular information about a specific customer profile – for example, how much they spend at each checkout, and the kind of items they buy. The data is analysed against a set of rules, triggering personalised marketing communications such as notifications and emails containing the most relevant offers for that person.

As a result, brands can encourage more frequent shopping in customers that show the most potential to convert, or those that show particular loyalty, thereby delivering incremental revenue over time. Of course, the more they buy, the more data can be collected and the more relevant these offers can become – a win win.

Targeting offers is a type of predictive analytics which can be used to push interested customers down the funnel to a first purchase or repeat purchase by offering actions that are most likely to serve their needs.

Lead scoring

Predictive analytics can be useful for sales teams too, by determining which business leads should be prioritised above others. It helps to identify which leads are of the best quality, leading to a more efficient and effective sales strategy that is more likely to yield new, high value customers. Econsultancy’s AI, Machine Learning and Predictive Analytics Best Practice Guide explains how this works:

“Machine learning can be trained to score leads based on certain criteria and what can be learned from past customer behaviour and actions. If, for example, customers that have interacted with particular content have gone on to purchase services or become more valuable customers to the business, this behaviour can be used to allocate scores to new or different customers when they exhibit similar behaviour.”

Integrating predictive technology with CRM platforms and marketing automation will help strengthen the amount of data accessible to the algorithm. In turn, this will enable more accurate forecasts for sales departments and give marketers insight into the most valuable to types of leads for nurturing.

Estimating CLV

Acquiring a customer can cost between five and twenty-five times that of retaining an existing one, according to Harvard Business Review. Maintaining good relationships with customers through offering content that is relevant and useful to them means that brands are much more likely to improve loyalty and sentiment over time. Predictive analytics can help to elevate customer retention by forecasting things like lifetime value or churn rate, allowing marketers to take action in advance.

Customer Lifetime Value (CLV) is an important metric which determines how much revenue can be generated from each individual customer. It is much easier to predict their inclination to re-purchase, and therefore their total value to a brand over time, by integrating predictive analytics into this process.

An RFM model, which measures recency, frequency and monetary value of repeat orders, can be used to help gauge a basic sense of which customers should remain a highest priority. However, having machines analyse the huge amount of customer data now available (in addition to this model), means that CLV can be based on many more detailed metrics than ever before.

Using machine learning to isolate customers which are most valuable lets marketers focus more of their time, energy and budget on encouraging repeat conversion in these individuals. Learnings could then be applied to those more reluctant to re-purchase, allowing the opportunity for further uplift in CLV across the wider customer base.

Estimating churn propensity

In contrast to forecasting CLV, predictive analytics can also be used to estimate churn propensity (the likelihood that a customer will end a subscription or abandon your brand for a competitor).

It is crucial that marketers are able to accurately identify customers that are most at risk of disengaging with a brand so that additional steps can be taken to retain them before it’s too late. Targeting disillusioned customers with incentives like discounts or loyalty rewards is one typical action marketers can take in the short term. After all, offering a one-off discount is much cheaper than trying to re-acquire the customer once they have made the decision to leave.

However, it is also imperative to analyse any patterns behind disengagement and put preventative actions in place for long term benefit, rather than relying solely on reactive action.

Segmentation

Predictive segmentation acts to recognise and separate individual online visitors into specific customer groups that are most likely to perform certain actions. Some of these segments may be deemed more valuable than others depending on their behaviour – e.g. visitors who return to your website frequently show a higher intent to purchase.

Having this data analysed and divided up allows marketers to see clearly the segments are not performing as well as they could, which can then be rectified as appropriate through a tailored marketing strategy. As data about customers and their behaviours becomes more robust, marketers are able to make more informed decisions about upcoming activations. One such example could be a targeted email campaign that aims to re-engage former customers that have not visited a website for x number of months. A/B testing is another particularly effective way of helping narrow down content to provide an experience best suited to each segment.

Knowing more about the types of people that make up a brand’s customer base can also help with recognising opportunities for bringing new products to market.

Segmentations and Personas Best Practice Guide

Preparing for predictive analytics

There are plenty of practical challenges presented when first implementing predictive analytics.

Before applying it across marketing and sales processes, organisations must prepare for the change. They must take the time to choose specific areas that they believe will most benefit from the technology and identify how success could be clearly measured once it is integrated. They should also choose and stick with a well-informed approach that aligns with the goals and purpose of the business.

Ensure your teams have the skills on hand to interpret any resulting data patterns, and decide whether it is worth hiring specialists in certain fields of machine learning.

Crucially, any adoption of machine learning should be ethical – it is important to factor in human supervision and accountability.

Implementing predictive analytics

The growth of AI and processing power has had a huge impact on the accessibility of predictive technologies. While larger corporations traditionally may opt for solutions from licensed products like SAS and SPSS, there are several free or more inexpensive options for smaller organisations.

Programming languages like Python and R have allowed for free open source predictive solutions for businesses. Here’s one example.

Less technical users can take advantage of the (limited) capabilities Google Cloud AutoML, Microsoft Automated ML and Amazon Personalise can afford them.

For more detailed information about how to approach the integration of predictive analytics, and its more complex uses, please refer to our AI, Machine Learning and Predictive Analytics Best Practice Guide, or book your place on our upcoming webinar.

AI, Machine Learning and Predictive Analytics Best Practice Guide