In Big Data, the More, the Merrier
By now, everybody knows that there are countless new data sources generating data points by the billions per day worldwide. However, not everyone understands how more data leads to more value, or better yet, to more money.
In order to illustrate the path from data to insight to money, I will leverage a well-known use case that pretty much every company has: Customer Segmentation for a Marketing Campaign. Marketing wants to send offers to as many customers as possible, given the available budget to influence their behavior in a specific way. Tailored offerings are created for specific personas. A persona describes a typical customer of a customer segment. The more similar the customers’ characteristics are to the persona’s characteristics, the more effective the offer is. In well defined customer segments, the customers’ characteristics are similar within each segment, but significantly different between segments.
Let’s go back a few years, when data wasn’t so abundant, and assume that we have just five variables for creating our customer segments. Let’s also assume that we want to create four segments. The upper left diagram in the picture below shows our four segments for five variables. The plane of each diagram captures the best two axes that explain most of the variability of the points. (For that, we used a technique called Principal Component Analysis.) As you can see, the segments aren’t good at all, overlapping with each other. This means that there is no characteristic that uniquely represents each segment. As more data becomes available, such as customer behavior based on transaction analysis, the segments become more clearly defined on the other three diagrams of the same picture. The best condition is with 40 variables, showing four distinct and homogeneous segments. In a homogeneous segment, customers are very similar to each other.
Segmenting our customers in four homogeneous clusters provides specific and clear business insight. But, how do we transform insight into value? First, let’s further simplify the visualization of proximity between customers to a single axis. An offer to a persona will decrease effectivity (return) as the customers are further and further away from the target persona. The diagram below shows the return per customer decreasing by $1 per unity of distance from the target persona.
Let’s assume that Marketing will start by generating a campaign for Persona A only. The total budget available is $600 and we have 800 customers in four segments of 200 each. The cost of preparing the campaign for each segment is $50. The first line of the picture below shows a return of $2,550 when we send an offer to the closest 600 customers to Persona A.
As we create more personas that better represent the customer segments, the total return of the campaigns increases, as shown in the second and third lines of the same picture. The third line is an ideal situation because it has 4 personas, who each perfectly represent 1 of the 4segments. Creating more personas, as shown in line four, increases the total cost of creating eight offers, without being any more effective, making the total return lower than in line 3.
This was an example showing how more data can enable deeper business insight, which in turn can be used to optimize the allocation of budget (resources) in order to provide maximum return.