Can you predict the future?

One of the happy spin-offs of running a loyalty programme is that typically it will generate a lot of very robust and useful data about its members.  Gathering, understanding and recommending action based upon this data is generally the responsibility of specialist Data Analysis teams.  One particularly powerful element of this discipline is predictive modelling, a technique used in analytics to create a statistical model of future behaviour. Predictive analytics is the area of data mining concerned with forecasting future probabilities and trends.

The lure of predicting the future is very powerful.  Ask any business owner the question "Would you like to be able to predict how your customers are likely to behave in the next 12 or 24 months at an individual level" and they will bite your arm off; there are myriad ways that knowledge of the future can be profitably actioned in almost any business you care to mention.

Technical expertise at the core

Given this fact, it is perhaps surprising that more businesses do not have predictive analytics at the core of their operations.  The mathematical and IT skills and resources have existed for ages.  So whilst predictive models have long been a staple in large financial institutions (particularly in the credit rating arena) and other industries concerned with risk, it is surprising that such models are not more widespread.  Given the huge amounts of potentially insightful data produced by most medium and large (and many small) sized companies these days, and given the apparently obvious desirability of prediction in business, why isn't this activity more widespread?

The commissioner, the creator and the translator

It is important look at the people who are commissioning the predictive models, those that create models, and those that convert them into true business benefit.  The skills required to undertake modelling of this nature are highly mathematical combined with knowledge of at least one programming language or modelling package.  Alternatively the commissioners and users of predictive modelling are generally highly commercial with a direct focus on marketing and are able to envisage and translate the insight generated into a commercial advantage.

Although predictive analytics is technical at its core, it must be run as a business activity in order to generate customer predictions that have a real business impact. This requires a wholly collaborative process driven by business needs and marketing expertise. This ensures that customer predictions are actionable within a company's operational framework, and that they have the greatest impact within a company's business model.  The biggest barrier to implementing a sustained and successful predictive modeling program is getting this collaboration off the ground in the first place - and getting two very contrasting sets of people communicating effectively.

Powerful or powerless?

Whilst controversial, the perception is that predictive models don't always work straight away, and sometimes they don't work at all.  To be seen as 'having worked' a model needs to demonstrably move the needles in the commercial space, not in the analytical space.  A model may be able to accurately predict - for example - which members of a health club might be in imminent danger of not renewing their membership, but unless action can be taken to change behaviour within this group then the model is not earning its keep. 

Nuffield Health, a leading health & fitness organisation with 55 gyms throughout the UK, is a prime example. They suffered from a very common problem in that many of their members, having taken out an annual gym membership, lost motivation and ended their contract with only one month's notice.  But, by using data analytics and predictive modeling they were able to identifypotential lapsers and instigate sales & marketing campaigns that would help them retain these members.  The model was found to have an accuracy of a least 70% and helped to identify behaviour that indicated when a member might lapse.

Looking at another industry, airlines regularly analyse the data collected on the members of their frequent flyer data programme.  For example, one airline found that people requesting aisle seats have a statistically higher likelihood of becoming Elite tier status than those requesting window seats.  Through the tracking of this behaviour, commercial and marketing decisions can be made in this group of members given their future potential and value to the airline.

Science meets business

The crucial factor though is what you do with this information. Whilst it may be a scientific truth and interesting in itself, can you make it work for you commercially?  This is where the collaboration between commercial and technical worlds is crucial.

Additionally, in some cases the amount of influence a business can realistically exert (given available marketing budgets) over the event being predicted isn't enough to justify the model.

In a similar vein, a model may not be offering enough predictivity in the first instance to generate enough quantifiable returns, and this results in the early termination of the project.  However, the complexity of predictive modelling means that the full benefits are often not felt until several updates of the model have been released and the results analysed and reported on. 

In some cases if the amount of influence a business can realistically exert over the event being predicted (given available marketing budget) or if a model which is not able to offer enough predictivity to justify sufficient quantifiable returns in its first instance, there is a risk that the business chooses not to invest further in the initiative. 

It is widely recognised that they often need time and commitment to evolve into something robust, useable and sustainable.  Predictive models are highly sensitive and accuracy is dependent on complex interactions between multiple variables.  Much like any business strategy, it is important to address and be responsive to challenges in order to ensure commercial success.  However, if all parties come together with a focus on what they want to achieve, then real insight can emerge and the resultant business benefits can be truly significant.