My belief is that predictive modelling can yield almost unmatchable results, if thought through properly at the campaign planning stages, at data extraction and analytical stages and at creative execution and direct communication stage (if the latter is needed).
Predictive modelling is a methodology applied to a campaign data selection, which predicts how consumers will behave in response to the communication, offer or incentive appropriately targeted for them.
So often when I discuss predictive modelling with clients, they are convinced that they are using predictive modelling already. However, what they are really doing is second guessing (hopefully through some data analytics) who they think will respond to a campaign. Inevitably, the majority of campaign planning time is given to the look and feel of the creative execution, instead of the strategy and delivery of almost pinpoint accurate data selection.
Let's take a product like lingerie from a retailer selling clothing, beauty, luxury luggage and homeware products, like a House of Fraser* in London. Typically, the data insights analytics team would say that they do predictive modelling on a direct marketing campaign to drive optimum results for lingerie sales. By this, they may mean that they look at who is already touching the lingerie category (maybe frequently & deeply or only marginally) and cross sell further into this identified customer base. Yes, this is a basic form of predictive modelling, but it can and should be so much more effective.
However, imagine truly analysing the data base from every single possible indicator as follows: Lets say that the customers who currently buy lingerie from a store like this are also displaying the following shopping characteristics (which can be extracted from the existing database):
Note that none of these shopping behavioural characteristics are linked directly to lingerie purchases - this is deliberate.
If you apply this data selection criteria (points 1 - 6 above) to the original set of lingerie shoppers, you can challenge the accuracy of this model. After many campaigns tested in this manner, industry experts predict that you should be able to obtain at least an 80% positive correlation (aka accuracy) in your customer selection criteria.
So what now? You have your six indicators which tie the lingerie shopper profile together. Let's say that there were 400 000 lingerie shoppers on the retailers' database, out of a total database of 3 million customers. Therefore the current opportunity of non-lingerie shoppers sits around 2.6m customers.
The lingerie category team and the marketing team are keen to produce a powerful direct marketing campaign, with the objective to get more shoppers buying lingerie from their store (ie new lingerie shoppers).
It's quite simple what to do from here. Apply the six identified indicators above to your 2.6m non-lingerie shoppers to extract a target audience for your lingerie direct marketing campaign and you will have an accurate prediction of who may respond to your campaign.
I referred to pinpoint accuracy earlier (maybe slightly optimistically) but it is certainly more accurate than what many marketing organisations currently do in their general database searches as pre-campaign analytics.
If you typically achieve a 5% response rate from campaigns (which obviously varies depending on who you are targeting and how exciting an offer is being communicated), you will certainly achieve a higher response rate to this campaign, if you follow a predictive modelling exercise as described and illustrated above.
Also remember that the basic and advanced communication elements of effective direct marketing must also be applied whether you are using email, SMS or glossy mail packs.
*Note that the 'House of Fraser' example is made up for illustration purposes only and is in no way based on real customer data facts.
