Predictive Modelling and Analytics
‘Proactive’, not ‘Reactive’, Decision-Making
Predictive Analytics is all about data science. Hidden within the huge mass of data that your organisation collects every day are clues to what your customers and employees think about your products and organisation. The more we know about an individual, and the more history we have of how people with these characteristics behave, the better we can predict what will happen in the future.
It’s just statistics – the same stuff you probably did at Uni, but this time applied to real people in the commercial world. Let’s have a look at how this can be used to analyse your existing products and customers.
A Simple Example – ‘Market Basket’ Product Analysis
Using statistical analysis can quickly show you the potential for product upsell and cross-sell opportunities, by using analytical association techniques to determine the combinations of products that different customers buy, and hence identifying those who do not hold these combinations of products. In addition, these association techniques can be utilised to analyse customers’ transaction history to understand ‘baskets of goods’ that certain profiles purchase.
The diagram above is an example of a basket of goods with the relative strength of the blue lines representing the strongest association between goods. In the example, Wine and Confectionery have a strong association as do Beer, Canned Vegetables and Frozen Meat. These association techniques can be used to, for example, provide offers to customers on their transactions based their current purchases.
As the example also shows, this sort of analysis is already extremely common in the Retail sector – and especially for on-line retailers, who can ensure they capture every relevant piece of customer data at the point of sale.
It also is easy to see how applicable this would be for your business too – assuming you sell more than one product!
Benefits of Market Basket Analysis:
The Next Step – Customer Segmentation or ‘Cluster Analysis’
In the same way that the relationship between products can be mapped, specific customer ‘clusters’ can also be defined using statistical analysis. In the first example below, customers have been clustered based on ‘attractiveness’. In this case, a company with a high focus on quality and service will want to retain and grow those customers that have similar inclinations, and be less concerned about losing customers in the other segments.
In the second example, customers are grouped based upon two characteristics – geographical location and education/affluence. This example shows how clustering can be used to look for relationships between customers that are not immediately apparent.
Benefits of Cluster Analysis:
But what about the ‘Predictive’ part?
Following on from the insights developed through the segmentation analysis, the next stage is to determine if and how those (or other) dimensions within the data available can lead to a prediction of specific customer behaviours. If certain indicators within the data can demonstrate a strong relationship with a particular member behaviour (e.g. buying a new product, cancelling your account), then these leading indicators can potentially be used to predict propensity of a customer to carry out that particular behaviour in the future.
What you really need is plenty of historical examples of the ‘event’ that you’re trying to predict. For example, if you want to know what type of customer is prone to cancel their account – or even better, what sequence of events leads up to it (so you can be proactive and nip it in the bud) – then the more cancellation events you can feed into the analysis the better.
The basic rule of Predictive Analytics is ‘the more data the better’: each new piece of data (whether it correlates with your hypothesis or not) will increase the accuracy of your predictive modelling.
Benefits of Predictive Analytics:
Once these insights have been obtained (e.g. a certain set of customers has a high propensity to cancel their account within the next 6 months), then this should not be seen as the end of the story. There is a secondary programme of work to determine the best way to apply this insight – a level of experimentation is usually necessary to establish what intervention might change the outcome for this set of customers (e.g. make them less likely to leave).
Other Uses of Predictive Analytics
While the examples explored above relate to customers and marketing activity, the same predictive techniques can be extended to:
In fact predictive analytics can, and should, be used for any risk mitigation activity that your business carries out!
Typically, organisations follow a four step process, starting with simple querying and reporting, moving on to multi-dimensional (OLAP) analysis once data has been effectively cleansed and ordered. Once these foundations are in place, the organisation can consider progressing towards predictive techniques enabled by data mining, and ultimately, to a truly personalised customer service based on the organisation’s deep understanding of their needs.
These are the four stages of a Predictive Analytics roadmap, with each stage taking your organisation further towards being a truly ‘proactive’ market player.
As a first step along this path, PMsquare would typically conduct a ‘Data Reconnaissance’ exercise, to identify the additional insights that may be available within your existing data, and how this could be used in each of the three areas of:
As a first step along this path, PMsquare would typically conduct a “Data Reconnaissance” exercise, to identify the additional insights that may be available within your existing data, and how this could be used in each of te three areas of:
“The ability to forecast more accurately and complete the allocations process more efficiently puts us in a strong position with our clients – they were keen for us to develop the new system and we feel it’s a key differentiator for our business.” – David Hogan, General Manager, Gordon and Gotch.
Other Uses of Predictive Analytics
While the examples explored above relate to customers and marketing activity, predictive analytics techniques can be extended to:
and many many more situations across varying markets.
Predictive Analytics Methods
Predictive Analytics can use statistical modelling (mathematical relationships between variables to predict outcome) or Machine Learning (typically classification, regression and clustering techniques based on algorithms that have been written to allow computers to learn from data sets) to predict the likely outcome based on supervised or unsupervised learning.
Deep Learning is a newer subset of Machine Learning that mimics the construct of human neural networks as layers of nodes that learn a specific process area but are networked together into an overall prediction. Examples of deep learning include credit scoring using multiple levels of social and environmental analysis or sorting digital medical images such as MRI scans and X-rays to provide an automated prediction for doctors to use in diagnosing patients.
New to Predictive Analytics? PMsquare would conduct half-day workshops that allow you to explore some of these practical analytical techniques for yourself.
Contact info@PMsquare.asia for more information and for workshop dates scheduled in your state.
For more information