Predictive Analytics & Optimisation
Predictive analysis is an analytical process to develop propensity scores and metrics that assist with customer targeting. These can then be optimised to identify the bet combination of messages for an individual customer while complying with business constraints.
When I started my career at the beginning of the 21st Century I worked as a data analyst in a marketing company. I used to build propensity models on our database and offer the models and functionality to our clients.
In those days we were pointing out profits of optimised targeting mostly with financial arguments. Because if our client wanted to sell, let’s say dishwashers – which in those days was a luxury good – they were looking for customers with the propensity to buy them. And they would send them a letter with an offer and several colourful leaflets. That cost a lot. So they benefited from our propensity scores and sent the valuable envelopes only to customers, whose chances for a positive reaction were higher. That was then - one product, many customers, costly channels and no GDPR…
Later, retention models, and customer behaviours models were in place. We were optimising and predicting customers reactions, possible churn, or fraud. Along with the development of digital marketing and data capacity – we could predict more and more.
Optimisation is always required when there are limited resources. And the Queen of science – Mathematics - comes into the game with huge support. A good data scientist can support a marketer with predictive models whenever it’s required to guess what customers like or how they will potentially behave.
The marketing world is changing and the only thing that doesn’t change is the change. This sentence is a must have in every marketing related article – I couldn’t resist. The world is changing, if you haven’t noticed.
Recently communication has become the cheapest ever – we don’t call our customers, we don’t send them letters, we just send emails, push notifications, banners, pop ups, social media audiences etc. The most limited resource we have now is the customer’s attention. We can’t bother them too often, we can’t be too boring, our messages have to be relevant, attractive and delivered at the right time – otherwise the customer will activate the GDPR shield, telling us – I’m not interested in this relationship.
Marketing automation tools and data mining can help, but we are not talking about one model that drives it all. There isn't a one size fits all optimisation rule. The marketing ideal is to know what to say, when to say it and how to say it. But also – who to say it to.
As a marketer it's your responsibility to own the overall customer contact strategy and decide what the next best offer is for a customer. Often this requires internal organisational changes. Product owners may need to stop pushing you to send two emails per week highlighting only their product, to the whole database, for example.
The first step is a change of perspective. Look at your communications from the customers’ perspective. What is being communicated? How often? Are they reacting to your messages? Have they reported any issues with the product or service?
The next step – look closer at your customers. I bet you are using segmentation, but you can make new one if you want, just for the purpose of this thinking process. Or enhance the one you already have. Query the database to see which products your customers have? How many? Were they offered other products and refused or are they eligible to purchase other products? This is a huge task, but will lead you to a better, clearer view of your customers’ activities.
Now, look deeper and closer at the data you have and identify groups of customers. Which groups are several steps further that others? Try to understand how they got there. What makes them different from others? Why are some customers are more loyal, purchase more, and do it more frequently. Who are the recent purchasers? Was it a successful campaign or some other factor that lead to their purchase? What was special about a particularly successful campaign?
All the above is just an introduction to your discussion with a data analyst or data scientists. The rest is in the modelling. Looking for answers in your data, finding similarities, patterns in customers behaviours and using the patterns to predict future actions.
But the final product of the calculation will be a score – propensity of predicted reaction. As a marketer you will use the score as one of many customer dimensions. Best practices suggest to combine different models and communicate with customers whose chances of positive reaction are highest. This means sending the message they are most likely to open, in the channel they like the most, with the product they are most likely to buy.
As a final note, with optimisation and model driven communications, do not forget about the low propensity customers. For whatever reason they may never reach the high propensity segments and it may be that you don’t have enough data for them... or they just are simply not interested. Whatever the case, don’t just push them to the side - they should be still communicated to, in their own way.