Reducing Member Churn With AI and Machine Learning
Membership churn is a problem.
Did you know it costs five times as much to attract a new customer than it does to keep an existing one?
To put in perspective: returning customers are more likely to spend roughly 67% more on the company’s products or services and with just a 5% increase in customer retention, it could generate a 25% increase in profit. With this increase in profit, it would mean that you would have to spend a lot less on the costs of acquiring new customers and in any organisation, the less you have to spend and the more profit you can obtain, the better!
Member churn is defined by whether a customer or a member unsubscribes or no longer purchases from, your organisation. There’s also the case of people who become ‘dormant’ that contribute to churn – dormancy, in this case, referring to people who remain signed up to your services but do not interact with your organisation or attend any events or open any emails from you. They haven’t succumbed to attrition, but they don’t generate any revenue either.
You want the member churn percentage to be as close to zero as possible.
The churn is not completely unavoidable; some people may have only wanted to make a one-time purchase, or they may have gotten the relevant qualifications they wanted to gain from your organisation and that was that.
There is a way you can predict if a member is at risk of churn… with AI and machine learning.
But Our CRM System Already Collects Data?
If you threw a bunch of Membership organisation CEOs into a room, what do you think they’d discuss?
Well, probably lots of things, but no doubt the topic of member churn will come up and the whole room will lament the fact that they have all these members but they’re just lying dormant, or they have people leaving after a month and they’re generally just struggling to retain those members.
Worst of all, they’re spending lots of money trying to gain new members.
How Does Machine Learning Work, Exactly?
A predictive churn model is a classification tool looks at past activity and uses that to identify the steps or stages when a member is falling away.
By adding this level of predictive analytics you’re better armed with the knowledge of the behaviours of your members and the moment they begin to lose sight of the value of your organisation – and what you can do to win them back around.
What kind of data does the Predictive Churn Model look at?
The data collated for analysis are:
- Customer profile
- Post code
- Average income
- How did they hear about your organisation?
- Do they interact with outreach (ie newsletters, emails, links on the website)?
- Products they interact with
- Purchase history
- Date of last purchase
- Have they asked questions? Were they answered?
- Complaint resolutions
- Where do they complain – email, phone, or Twitter?
This is a small snapshot of the breadth of information you can get to find out which event leads to them unsubscribing. Relevancy ensures accuracy and with machine learning, you can input and sort more data than is humanly possible.
But firstly, that analysis will only be as good as the data it’s given so the data needs to be categorised into three criteria:
Complete – Are all the dimensions (labels) filled out? Are there bits of data with missing values? If that’s the case, you will have to do some data exploration and fill in some of the missing values. For example, if there’re any values of ‘County’ missing you can fill in the value by checking the address.
Clean – Are there multiple values for the same dimension? For example, ‘Derbyshire/Derbys/Derbs’. If that’s the case, then you must cleanse the data to ensure uniformity.
Accurate – And, overall, is the data correct? Are the values are all in the right place, nothing answered ‘N/A’, negative revenue values for some transactions?
There’s one more step in predictive+diagnostic analytics that you need to do to prepare your data, and that’s creating the ‘target variable’. For churn analysis it can be something as binary as ‘Will churn?’ and you can fill in the values for this variable by analysing that all-important historical data. For example, you’d have a value of 1/TRUE for members who cancelled their subscription and 0/FALSE for those who renewed.
Machine Learning And The Future Of Membership Organisations
After you’ve exhausted all the information you can gather, and you’re going bonkers for data and segmentation and classification, you will see those trends.
And if you haven’t spotted at least three highly correlated attributes, then you simply haven’t asked the right questions.
It’s incredible stuff to be able to have all this customer data consolidated into one place and unlock all these trends that have been leading to all that churn!
From this model you can make smarter business decisions that have been wholly formed from behaviour analysis, giving you the confidence to ensure business growth and resilience.
I Need To Know More About Predictive Analysis
Machine learning launches organisations just that bit further. If you’re interested in making better informed, high-quality and risk-assessed decisions for your organisation then it’s time to contact the experts…
That’s us, by the way, so drop us an email!