Who Let The Machines Out?

Who Let The Machines Out?

In last month’s post, “No More Carpet Bombing”, we discussed the impact that data can have on the success of your marketing strategy. We mentioned how by targeting households which had recently moved house, DIRECTV in the United States realised significantly improved conversion rates. This week let’s look at a few of the approaches that data scientists use and how these can translate into actionable insights.

We collect and process more data today than ever before. The sheer volumes of these data bring with them inherent difficulties but they also present many new opportunities for businesses. Walmart recently provisioned a private cloud capable of processing 2.5 petabytes (that’s 2.5 million gigabytes) of data per hour. This substantial undertaking is providing Walmart with streamlined solutions to complex business problems. Not all businesses need this scale of data processing however, and even relatively small volumes of data can yield surprisingly powerful insights.
These days, a data set may include demographic, transactional, financial and behavioural data, web traffic and social media activity, event calendars and even meteorological forecasts. Such data can give us a glimpse into a businesses’s operations and their clients’ behaviours and requirements. We can see for instance who has defaulted on their loans, or as in the case mentioned above, whether people have recently moved home, and use those insights as the basis for informed action. The real potential of this data is generally hidden from view and can only be discovered using suites of machine learning algorithms.

Data scientists work to discover hidden links between customer attributes and can predict whether customers are likely to remain loyal or defect, or what they are likely to buy next. Predictive insights like this can allow businesses to target specific customers with tailored marketing that maximises uplift and positively impacts their bottom line while simultaneously delivering value to their customers. What these approaches amount to is enabling businesses to engage with their customers in more personalized and relevant ways.

Who let the machines out?
Clustering and segmentation: Despite the success of DIRECTV’s campaign, segmenting consumers along a single variable such as having recently moved home, can be likened to using a blunt tool. A single factor  approach may ignore other crucial differences related for example to financial or transactional history. When considering tens or hundreds of such characteristics, segmentation quickly becomes impossible to do manually and data scientists will often employ clustering techniques. These are typically ‘unsupervised’ machine learning algorithms in the sense that the algorithm is let loose on the data as it were, without any prescribed outcome in mind. The algorithms are optimised to hunt down natural divisions between groups of customers based on many distinct characteristics. They have the advantage of exposing otherwise hidden patterns within your customer base, and characteristics of these patterns can in turn inform market segmentation in a highly contextual way.

The importance of experimentation 
Hypothesis testing and predictive modelling: Let’s say a marketing department is asked to reactivate a segment of inactive customers. We might hypothesise that offering these customers a 20% discount will have the desired effect. However cluster analysis reveals that the segment is really quite varied, meaning a ‘one size fits all’ approach is unlikely to be effective.  We need to be cautious about expending resources on targeting people who will respond neutrally, or worse, negatively, to the campaign. 

The smart thing to do is to deploy a test campaign to validate the initial hypothesis. The responses of test and control groups can be monitored and these data can be used to measure true uplift and train predictive models to identify the subset of individuals who will be positively susceptible to the campaign. Of course if the uplift is too low, the campaign can be scrapped (or modified), avoiding unnecessary expense.

Recently, Ixio Analytics tested the impact of an account reactivation campaign for one of our clients, an African financial services provider, and demonstrated 15.9% uplift. If you’d like to learn more about how we help organisations leverage their data towards achieving their business goals, give us a call. Your data will be glad you did. 

Share This Artcle :