Customer-Centric Analytics Makes Both Mathematical and Business Sense

Experts in database marketing have long advocated the importance of customer-centric marketing. To support a customer-centric business strategy, a company needs to adopt customer-centric analytics, especially if the company is going to leverage analytics as a strategic and competitive asset. By customer-centric, we mean that the orientation of the analytics needs to focus on the customers and whenever possible, we want to leverage customer level data in our analysis. For example, performing a pricing analysis usually involves the understanding of certain product’s elasticity when the price goes up or down. A product-level econometric model to quantify the pricing effects on product demand is still very important. But in addition to that, we should also look at how the price change has affected customer’s behavior. For example, who is buying more? What else do they buy?

Here is a simple example to illustrate the outcome of a product-level optimization vs. a customer level optimization.  Suppose we have a customer base of 10 and we have four product campaigns. We assume that customers are making decisions independently from each other and across product categories. Obviously, this assumption may not be valid. In fact, the violation of this assumption in reality would make a customer-centric approach even more imperative. Here, we show that even with this assumption, the customer centric approach is better. We further assume that no customer will be promoted twice and each campaign needs three customers. We have randomly generated a pay-off matrix. We assume the pay-off matrix takes care of the probability of accepting an offer, the expected value conditional on acceptance and the marketing/service costs.

The problem of selecting customers for campaigns can be conceptually formulated as a constrained integer programming problem.

Let p_ij be the payoff for customer i and product j, and let x_ij be the binary decision variable of offering customer i product j, mathematically, we want to optimize the following system:
maximize SUM(i,j) (p_ij x_ij) , subject to SUM_(i,) x_ij =3 for all j, and SUM_(j,) x_ij <=2 for all i.

If we ignore the constraints, this problem can easily be solved by picking out the largest 12 cells in the pay-off matrix.
The product-oriented selection needs to assume a particular order in which customers are allocated to campaigns. Since we generate the numbers randomly, we can just start from A to D without losing any generality:

Product Oriented Solution

In this case, the total payoff is $10.23. We should point out that this solution is not optimal in the strict mathematical sense, but it is a very good solution. In fact, the way we allocate customers to campaigns is a greedy algorithm heuristic, which in many situations may actually lead to a globally optimal solution.
Let’s look at the customer-oriented approach. In this case, we start with selecting two best products for each customer. In doing so, we would exceed the total mailing budget. We then backtrack to eliminate the lowest payoff cells. The key difference is: instead of making decisions column by column, we make decisions row by row.

Customer Centric Approach

The total payoff has been improved to $10.43. But wait: we have violated one of the business constraints; namely, the total mailing for product A is 4! I am not trying to be intellectually dishonest here. Rather, I am trying to make a point, that if you want to perform customer-centric analytics, you need to have the flexibility of removing or relaxing some of the product-level constraints.

For example, this may mean persuading product managers to allow their budgets to be partially centralized so that an optimal solution for the entire marketing budget can be achieved. This is not an easy task. We need to make sure the solutions are better overall, and no particular product manger will be disproportionally worsened off. One of the common misconceptions we encounter in our pratice is this notion of dichotomy: you can either perform product optimization or customer optimization, not both.

This is not true at all! Both approaches are heuristics to the same underlying optimization problem. In any optimization problem, the issue is always making the right trade-off. We are simply suggesting that we should trade some product-specific goals for a better cross-product solution. If there are product goals that have to be met, we can easily introduce them as part of the constraints in the optimization. In many cases, we can demonstrate clear benefits to eliminating some unnecessary internal competition for customers, which occurs quite often in product-oriented marketing companies.

There is another interesting pattern regarding these two solutions. The product-oriented solution is much more uniform from a product standpoint, while the customer focused solution is more uniform at the customer level. Is this a benefit? Yes, in the long-term. The customer-centric approach mitigates the negative effects of constant cherry-picking customers by your marketing programs. Companies may perform very sophisticated modeling to improve campaign efficiencies which leads to much better ROIs. But after a while, to their dismay, they noticed that their active customer base has shrunk significantly and they have to acquire new customers to make up the difference.

Comments are closed.