A/B Test Your Branch Network

  • 3 July, 2014

“Lies, damned lies and statistics” have become the basis for arguments on the future death of branches. A quote actually attributable to Mark Twain is probably more appropriate –“All generalizations are false, including this one.” But what if I were to tell you that there is actually a way for a financial institution to measure the impact of a branch closure, the answer is A/B testing.

While perusing the great content on The Financial Brand an article by David Gerbino referenced much of the research to justify branch closures.

According to Bankrate.com 30% of customers who have not stepped inside a branch in the last six months. As I get older I visit my parent’s house less often but my digital engagement with them has increased. I will be very unhappy if their house is gone when I next go to visit them because you extrapolated that no recent visits means a desire never to visit! Similar to the approach taken by Chase, I am happy for my parents to downsize but certainly not to disappear.

Running a monthly report for customers does not tell you whether to keep a branch open or not. The fact that I did not transact in a branch in the last 30 days is not analogous to measuring the importance of the availability of the branch channel to me.

A/B testing is the process of starting with two identical groups, making a change with one group and tracking the differences from the other (control) group. This allows for all the external factors to be accounted for. How could this be applied to branch evaluation?

Money talks. Create two alternate checking products, one that allows the use of the branch and another, with slightly cheaper fees that only supports non-branch channels. Group A will have no choice, just your regular checking account. Group B will have the choice of a checking account with or without branch access. Measure the difference in conversion rates. Measure the choice between people who choose branch access and those who chose not to have it.

Branch closures have begun and will continue. Identify a branch scheduled for closure and find a branch in your network with similar demographics, costs, transactions, profitability etc. From the point that you announce the branch closure to 6-12 months post branch closure measure the performance of the customers associated to that branch. Measure the customer acquisition and loss in the relevant geographic area.

Closing branches is about predicting how your customers will behave. Measuring future behavior is not easy and is not precise. Using past transaction data creates statistics, but is it the truth?