Brief 1 showed where the money moved. Brief 2 showed who the households are and where branch coverage is thin. This brief answers the last question: what is the cheapest way to serve those households, and when does a branch pay? We hold no cost data for shared branching, ATMs, mobile or digital service, so we do not guess at any cost. What follows is ten years of branch data and a break-even tool that runs on your own numbers.
Brief 1 reported the headline: the median new branch still open booked $6.7M in its first year, $37M in year five and $57M in year ten.1 Here is the whole curve. The $500M office cap applies throughout, as it does in Brief 1.
Median deposits of new branches still open, by year of operation, $M
The ramp is slow, and not every branch makes it. 12.4% of the new branches old enough to observe were gone within five years.2 Treat every median here as the experience of branches that survived, which flatters the typical branch.
Year-one deposits rank 0.68 with year-five deposits across 5,344 branches, so the first twelve months carry much of the signal. Among the things we can measure before opening, who opens the branch explains more of the variation in year-five deposits than where it sits.3
Share of variation in year-five deposits explained, out of sample
Where a branch sits still matters at the margin. Brief 1 reported that a branch with a retail cluster inside the Fed radius4 books about 23% more at year five, holding place type, households, income, opening year and state constant.5 This is association, not cause, and it says nothing about branches in places with no cluster at all: 95% of new branches have one in reach, against 15% of banking deserts.
For each new branch we counted the households within its Fed radius and divided its year-five deposits by them.6 The result is a deposits-per-household figure we can scale to other markets, with a warning: it varies a lot, and it does not rise with the number of households.
| Place type | New branches placed | Households in reach, median | Year-five deposits, median | Deposits per household, median (25th to 75th) |
|---|---|---|---|---|
| Urban | 4,050 | 18,424 | $40M | $1,960 ($678 to $4,592) |
| Suburban | 4,044 | 45,680 | $39M | $970 ($389 to $2,320) |
| Rural | 1,137 | 9,658 | $21M | $2,538 ($1,034 to $6,519) |
A suburban branch has 2.5 times the households in reach of an urban one (45,680 against 18,424) and holds about the same deposits ($39M against $40M). Across all branches the rank correlation between households in reach and year-five deposits is only 0.20. Deposits per household therefore fall as the market grows, and a single figure will mislead. Branches with a year-five figure that we could place: 5,172.
A banking desert is a tract with no bank or credit union branch within the Fed radius of its population center. Here is how many households the median desert has in reach, next to the median new branch of the same place type. Counts are the 3,608 deserts we could place (note 6).
Households within the Fed radius, median
Only 1.1% of urban deserts, 0.0% of suburban deserts and 1.6% of rural deserts have as many households in reach as the median new branch of their place type. That is 16 of 3,608 tracts (8 urban, 8 rural), on households alone. Brief 1 applies three tests together, a retail cluster in reach, enough households and predicted deposits, and 1 of 3,625 deserts passes all three. What would that mean for deposits? We have two views, and they disagree widely.
| Place type | Households in reach, median desert | Illustration: households times historical deposits per household, year five | Market model prediction, year five |
|---|---|---|---|
| Urban | 5,148 | $10.1M ($3.5 to $23.6M) | $36.2M ($29.8 to $46.9M) |
| Suburban | 1,982 | $1.9M ($0.8 to $4.6M) | $25.8M ($23.0 to $30.6M) |
| Rural | 1,114 | $2.8M ($1.2 to $7.3M) | $27.7M ($24.7 to $30.8M) |
The illustration multiplies the median desert's households by the deposits per household around new branches (median, with the 25th to 75th range). It is an illustration, not a forecast: new branches sit in much larger markets, so a desert is outside the range of the data. The market model7 predicts far more, partly because it weighs retail, competition and growth as well as households. Its rank correlation with outcomes is about 0.37 to 0.38, so read its band, not its point estimate. Real results for a desert are unknown, which is why the next section lets you test your own numbers.
Enter your own annual cost to run a branch or any other service option, and your net margin on the deposits it gathers. Nothing here is our estimate: the boxes start empty, and every output uses your numbers with the history above.
The break-even tool runs in the online version of this brief.
Margin means the net revenue you keep on each dollar of new deposits, after funding costs. Costs are yours to define: staff, premises, equipment, technology and overhead for the option you are pricing.
The data supports a few facts about each kind of desert, shown below.8 What follows from them is our judgment, and we have not tested it. Use it to decide which options to price first.
| Kind of desert | What the data shows | Our reading, untested |
|---|---|---|
| Urban edge 713 tracts | 5,148 households in reach. A retail cluster is in reach for 39.1%. 9.6% of households lack home broadband, against 13.4% in well-branched urban tracts. Nearest national grocer 2.5 miles. | Incomes are higher and online access is better, so digital service with ATM access usually covers it. Price a branch only if your break-even clears. |
| Suburban 2,394 tracts | 1,982 households in reach. A retail cluster is in reach for 10.2%. 18.6% lack home broadband, against 10.8%. Nearest national grocer 8.1 miles. | Digital alone leaves nearly one household in five out. Shared branching and assisted enrollment fit better, and a branch is the hardest case to justify at this household count. |
| Rural 501 tracts | 1,114 households in reach. A retail cluster is in reach for 5.2%. 27.1% lack home broadband, against 21.1%. Nearest national grocer 17.7 miles, against 5.2. | No retail anchor exists to build around and broadband is thinner. Mobile service, shared locations and local partners fit better than a branch. |
Work these four questions in order.
What we did not test
We did not test field of membership, charter limits, capital and cost, shared branching networks, ATM networks, mobile service or digital enrollment. We hold no cost data for any of them. A branch is the only option whose history we can measure, which is why the tool compares everything to it.
Where the money moved. Since 2021, deposit growth has gone to places with more room, where new households keep arriving (Brief 1).
Who is there. More children and more suburban family households in the markets gaining share, where thin coverage and share gain are different things (Brief 2).
How to serve and pay. A branch has needed a real market, and most deserts are far smaller. Price your options against your own break-even before you build (this brief).
Name two or three. In 30 minutes we will size each one in households, show the households in reach, and run the break-even with your costs.
Request a branch optimization analysis