Results of US Banks’ Branch Network Analytics; Telling Tales of Missing Logic

In this age of digitization and mobility, banks are facing a challenge in deciding the level to which they maintain a physical footprint (a.k.a. bank branches). In the US, as more people are depositing checks online, paying bills electronically, accessing cash through ATMs, and generally bypassing traditional teller-serviced experiences, there is a decreasing need for physical branches. But there are certain nuanced yet necessary considerations when it comes to deciding which branches to close and the mathematics behind it.

The LTP Data Analytics team analyzed the bank branch network footprint against several parameters. Bank branches in the U.S. were studied against two major factors viz. population and GDP, to understand the correlation. LTP is thankful to Factual, a location data company, for providing us bank branch data from their Global Places database in order to complete this analysis.. Although Factual data is typically available for commercial consideration, they graciously shared it with us (under NDA) for performing this analysis. It’s really great to collaborate with such companies, who believe in the ‘open innovation’ model just as we do, and conduct analyses that help our readers to stay informed and innovate.

This analysis was done from two different perspectives: firstly at an overall state level (using GDP and population data from US Census), and secondly at a Zip Code level (using Income Level data from What was most interesting was that we observed a major change in the patterns between the state level and the zip code level maps. The following illustrations/charts reveal the insights that we unearthed:

Chart: Banks vs. Population (State level)

We mapped all the states based on the number of bank branches they have against their population. From a statistical standpoint, the following chart indicates that there is a high level of correlation (R-square value of 0.955). There are larger numbers of banks in states with larger populations which is the ideal scenario. We then went ahead and tested the other factor.

Chart: Bank vs GDP (State level)

This time we plotted the states based on number of bank branches against state GDPs. From a statistical standpoint, again, the following chart indicates that there is a high level of correlation (R-square value of 0.9102). There is a larger number of bank branches in states with higher GDP values which is what you would generally expect. Also since the population and GDP both have been considered in the bank network planning, it is all good and logical so far. However - and you knew this was coming! – there’s more to the story, so please read ahead…

Chart: Population vs Banks (Zip Code level)

At LTP, our DNA is research and analytics. So we love doing this stuff at a granular level, asking more questions and writing more queries to the database. So, after the state level analysis, we went down to the Zip Code level. And this is where things got really interesting. From a statistical standpoint, the following chart indicates that there is a low level of correlation (R-square value being only 0.3879) between the number of bank branches and the population at the Zip Code level. There is no correlation as such and that troubled us. Large numbers of Zip Codes with lower populations have a relatively larger number of bank branches and the scenario is totally opposite for certain Zip Codes with higher populations where there are fewer bank branches.

It seems that the logic and the rigor in place to set up the network of bank branches at the Zip Code level is arguable. We looked at other reasons why this could happen but nothing would explain the apparent anomaly.

We looked at 10,000+ random Zip Codes for this analysis, not all zip codes, but that doesn't affect the analysis.

Chart: Banks vs. Income Level (Zip Code level)

From a statistical standpoint, the following chart indicates that there is a very low level of correlation (R-square value being only 0.0851) between the number of bank branches and the median income levels at a Zip Code level. The median income is being considered as a factor here to represent the GDP or wealth in the Zip Code (this is understood to be an acceptable assumption). There is quite a bit of irregularity considering the overall scenario of distribution of bank branches using this lens. Certain areas with lower median income levels have a larger number of bank branches and the scenario is opposite for certain areas with higher median income levels.

We looked at 10,000+ random Zip Codes for this analysis, not all Zip Codes, but that doesn't affect the analysis.

Can the mistakes be corrected? Can we bring more logic to the branch network?

Considering the current market trend, banks have been shutting down branches in response to the changing consumer behaviors. JPMorgan Chase is planning to cut 5% of its bank branches by 2016 while Bank of America has cut its branch network by nearly 15% since 2011. According to SNL Financial, 2599 branches were shut down while 1137 were opened in 2014 itself, bringing the net loss of branches to 1462 as compared to 1487 in 2013.

Thong Nguyen, co-head of consumer banking for Bank of America, recently told The Wall Street Journal that the bank wants "to follow where the customers really want to go," noting that mobile-banking users have been growing by about 20% over the past few years while traffic at branches is down 10% or more.

Given our Zip Code level granular analysis, it appears that bank branch reduction should be implemented more surgically, with a lot more thought. Giving the digitization and transformation in banking needs, there will be clearly be fewer branches required. Also, the economics of a low-interest-rate, narrow-yield-curve environment makes it less profitable to be spread out too much in terms of number of branches. But at the same time, branches would be required for certain customer segments. So as an example, considering the population chart we showed you above, the bank branch reduction should ideally be performed in areas having lower population levels, where the current number of branches is already high. There are many such considerations that banks can look into while closing branches. This will help them to create a better network for serving their customers’ banking needs more effectively with the remaining branches.

General economics also plays a role such as in the on-site branches in third party stores. JPMorgan Chase closed 45 branches in the Chicago area when Dominick's Finer Foods announced it was closing or selling 72 stores, many with bank satellites, while the Bank of Oklahoma said this week it will shutter 24 grocery store branches, primarily because of the preference for off-site banking.

SunTrust Bank Case Study:

Chart: Population Range vs No. of Branch Closures (ZIP code level)

The above chart represents the number of branch closures in areas by defined population range. A noteworthy aspect here is that areas having populations of less than 10,000 witnessed only 14 branch closures which is quite less than those areas which are highly populated. Ideally, more branch closures should occur in areas having lower populations.

So…should SunTrust have closed relatively more branches in areas with populations of less than 10K?

Chart: Income Level vs No. of Branch Closures

The above chart represents the number of branch closures that have taken place, categorized by the income levels at the Zip Code level. Look at how there is less correlation between the two aspects. Most number of branches were closed in middle income groups? Not in the upper and not in the moderate? I will leave it at this for now. You have some food for thought. I promise to come back with more analysis on this topic.

Special Note - Thanks again to our Data Partner, Factual, for providing access to their data for this article. Factual is a data company that helps make sense of what’s happening in the physical world, enabling developers, publishers, and advertisers to build more relevant and personalized mobile experiences using the context of location.

Note- This article was updated at 9:30 PM IST on 13th March 2015. Couple of changes to the last two paras.