By Paresh Patel, consulting services director at Qbase
Securing sustainable revenue is hard for any business, let alone when you’re faced with a pandemic and a high degree of market and economic uncertainty. And while this isn’t an article about how to survive the pandemic, it is here to provide an illustration into how machine learning models can help businesses to create sustainable, ongoing revenue streams.
The best of the pandemic has seen brilliant moves by companies to reinvent what they do and how they do it to maintain revenue and even create new income streams. From small businesses like gyms, who have welcomed members’ family in for free or offering to suspend memberships to countless discounts, value adds and BOGOF offers. But what if you could sustain your revenue level by predicting the move that a consumer will make in response to your marketing efforts or even the market environment? This is where machine learning models come in.
The process uses software tools and techniques to analyse data and build models that can predict how your customers will behave in the future. The analysis looks at lots of data variables – from past consumer behaviour and demographics, through to external data such as the weather or economic indicators – to build an accurate picture of future behaviour.
Machine learning modelling is key to helping businesses understand their customers’ triggers and challenges, and in turn building more sustainable revenue sources. It can show businesses where other sources of potential revenue may lie; and it can help aid customer retention.
Identifying new revenue streams
For companies who saw their industry sectors shut down overnight, identifying new revenue streams was critical. Here modelling can investigate consumer data to not only improve segmentation but also to identify patterns that could indicate new areas of interest or likelihood to buy.
Take a professional catering company for example. When the hospitality market closed down, so did its sales of professional catering equipment. Looking at its customer portfolio, it identified a small section of consumers who were buying and using professional kitchen equipment at home. Using machine learning modelling it was able to predict the type of product that these ‘home chefs’ were likely to buy and build marketing campaigns around these product lines to target this sector.
Companies with subscription, membership or reoccurring revenue business models were hit hard during the pandemic. Those that rely on regular payments from consumers have had to become inventive to stop the cancellations when times get hard. Here machine learning modelling can be used to track behavioural patterns that provide signals to when a customer is likely to churn, enabling organisations to be proactive in developing a retention strategy These predictive models are called anti-lapsed or anti-churn models.
If we look at charities, modelling can also be used to convert customers to a more regular giving product for example. Are you seeing opt outs at certain points in the customer journey? What happens if you have customers that renew every three months? It’s likely that they’re more at risk of opting out than those that renew annually. How do you move them to annual renewals? Identifying behaviour patterns in the annual renewal sector can determine a course of action for converting short term renewals. Do they need a certain amount of communication during that period? What marketing comms collateral works? What doesn’t? At what stage do you need to send them service messages or reinforce what they’re getting from their membership for example?
Charities have been successfully using retention strategies to try and maintain donation levels. Building predictive models to identify which donors are likely to stop giving – and then offering them the opportunity to pause or downgrade their giving to a more manageable level. This ensures that the donor relationship isn’t lost completely and can be retargeted at a later stage.
Anti-lapse strategies can also be built to retain customers. Those organisations with members can, for example, build a model to identify behaviours that link to customers not renewing. Consider the tenure of someone with one year’s membership versus behaviours linked to a customer with ten years for example. Identifying who is likely to lapse means that the organisation can make sure that they are talking to them at critical pinch points of the customer journey or life moments, offering the right information at the right time.
Building machine learning models doesn’t have to be complex. It’s about understanding the results or impact of certain behaviours. Start by looking at the different stages of the customer life cycle. Draw the customer journey and plot the key points. What happened between the first order and the second? Did you target the customer with a recommendation? Did they unsubscribe? This valuable data lets you change the way that you react with other customers at the same stage.
Once triggers are identified, they can be scored to reveal nuances of behaviour. Say a customer has received a renewal letter but didn’t renew or cancel either. Scoring here will be different to those who cancel.
Machine learning modelling shouldn’t stand still though. It needs to be reviewed regularly as triggers, such as customer behaviour and their life stage will change over time. Try and operationalise your modelling. Make sure you report on it, run control tests on it to validate it, and teach and train colleagues how and why to use it.
Consider using a model factory to organise and manage machine learning models. A model factory is simply a collection of models that are organised in way to answer key parts of the customer journey, such as pinch points. They are catalogued by outcomes such as predicting a customer who will churn or converting them to a different product.
There are of course challenges to machine learning modelling. Both the volume and accuracy of your data is key to the success of the models. It’s worth looking closely at your data to see what governance should and could be applied. Is your data connected properly or does it stand in silos?
To gain a clear picture of behaviour you need to ensure that data works together. Talk to the data domain experts. They should be able to quickly identify where and what data should be considered for use and help you to short cut some of the pre-analysis work that needs to be done.
Finally think about how far back your data goes. Does it provide an accurate reflection of what’s happening right now? Have behaviours changed or remained the same? All questions that you need to answer before modelling starts.
It’s clear that machine learning modelling is beneficial for businesses who need to deliver sustainable revenue models. Using data to predict what the customer does next and then using that information to target them more accurately can help to create new sources of income, cross sell and up sell to existing customers and even encourage new behaviours. Ultimately, machine learning models will not only show where you should invest in your customers, you should also see the long term benefits through increases in customer life time value. But remember to achieve this, it all starts with your data, make sure is fit for purpose, relevant, accurate and explainable or you may end up putting the same customers off doing business with you.