Is machine learning the future of loyalty programmes?

Many brands swear by loyalty schemes. Collecting data on the customers and making them special offers and rewards to stimulate frequent purchases - for many years it has been seen as the most effective method to boost customer loyalty.

However, it seems many companies do not actually gain value from the loyalty schemes, or they even lose money, overspending on promotions that do not bring the customer back. According to research by McKinsey, those with the least emphasis on customer loyalty saw a greater year on year revenue return than those with a high emphasis - by as much as 13%. The gap in EBITDA for these companies is also widening with every passing year – companies currently not focusing on loyalty have a 16% advantage over those that do. 

But despite these findings, the loyalty scheme is not going anywhere soon – and rightly so. The ambition of encouraging loyalty is prudent and potentially profitable. The idea is right, it’s the implementation that is typically at fault.

Ideal loyalty schemes are based on the idea of personalisation – provision of relevant offers and targeted promotions based on the data the companies have about the customer. That is the main challenge to solve, and it is often done wrong.

The ambition of encouraging loyalty is prudent and potentially profitable. The idea is right, it’s the implementation that is typically at fault.

Current loyalty schemes rely on data being collected and analysed by the marketing or customer experience department, who decide on the rules for promotions and implement those. Of course, their decisions are based on interactions history, customer profiles and the knowledge they have about their behaviour.

But there is a reason that the data collected in these processes has been dubbed “Big Data”. The problem is, it’s too big for humans to analyse. Humans are quite simply limited in the complexity they can handle, in the hypotheses they can produce and the responses they can process. Not to mention the speed with which humans can make these calculations - compared to computers, humans are painfully slow.

And that is where the current approach is flawed. Many companies running loyalty schemes are undermining their own data’s potential by using it only as a support to human decision-making. In effect, their personalisation efforts are not implemented – and many customers continue getting the offers they don’t need, or the company misuses its marketing money without getting an increase in loyalty. Simply because humans inevitably generalise and simplify when making the marketing decision for the whole customer base – not being able to process all the data available.

Instead, brands need to move past traditional business intelligence and embrace machine learning. These new technologies give the possibility to actually implement the personalisation on a level of each individual customer, targeting each offer up to the “group of one”. Algorithms are able to calculate the most suitable action from hundreds of thousands of hypotheses – and then also learn from every action and reaction fast.

In contrast, human-run loyalty schemes re-examine effectiveness and edit the programme rules on a periodic basis – perhaps quarterly or even annually, and base their decisions on outdated information, unable to adjust smoothly to the changes in preferences and behavior.  

The underwhelming reality

There is one more point to take into account – of course, loyalty schemes should provide the customer with a targeted offer that will make them feel valued and stimulate further purchases. On the other hand, the budget spent on promotion should be considered. A good example is the room upgrade in the hotel for a loyal customer – it makes the customer happy but generates little additional costs for the company, if the better room was anyway free.

For solving the loyalty task, we cannot simply use the algorithm that would give the customer the personalised “next best offer” he will be likely to accept. The cost of promotional offer should be included in the model – and that is what machine learning can also do. Using the new technologies, we can deliver the customer the best offer at the minimal cost, balancing between the desired effect and the budget optimum.

Such variables are for the marketing team to decide, basing on the overall strategy – in order to guide the actions of the machine. The new technologies bring the possibility to predict the future actions and prescribe the action according to the desired KPI – be it increasing the average check, stimulating the customer to return more often, improving the margins or delivering the cost-optimised loyalty offers.

Many companies running loyalty schemes are undermining their own data’s potential by using it only as a support to human decision-making.

In addition, not only the type of promotion can be decided, but also the best timing, and channel to use – all based on the analysis of innumerable factors – many of which a human would not even take into account.

Retailers of all sizes need to start investigating the potential of machine learning. This topic that came from science and mathematics, is quickly moving to become an essential tool for companies that possess vast amounts of data. And loyalty management is one of the obvious use-cases to be tackled first.

With all the competition on the market, the benefits of the new approach are clear and undeniably outweigh the current underwhelming reality. Indeed, as counter-intuitive as it may be, the route to loyalty is in relying on the machines to serve customers as if each of them was personally served by a human. On a real scale.