Function 6 · deposits and customers
Where are the deposits, how much of a rate cut reaches customers, and whom should the bank call before they leave?
Deposits stand at $930.6M, +13.2% over three years, and the largest one percent of depositors hold 12.6% of them. When the policy rate fell 2.58 points, savings passed on 0.38 of the cut at once, while time deposits passed on 0.32 in the first year because a certificate reprices only when it matures. The retention list puts $9,442,588 of expected balance at risk in front of 500 calls.
Deposits by product
Balance
Source: generated, Harborline deposit accounts, September 2023 to August 2026, seed 20260831, as of 2026-08-31.
The policy rate, and what each product pays
Rate, percent
Source: generated, Harborline deposit accounts, September 2023 to August 2026, seed 20260831, as of 2026-08-31.
Checking accounts opened and closed each month
Checking accounts
Source: generated, Harborline deposit accounts, September 2023 to August 2026, seed 20260831, as of 2026-08-31.
Deposit betas: the generator's, recovered, and paid
| Product | Generator's beta | Recovered from the rate sheet | Within 0.02 | Paid rate beta, first year of cuts | Paid rate beta, whole cycle |
|---|---|---|---|---|---|
| Checking | 0.02 | 0.02 | yes | 0.01 | 0.01 |
| Savings | 0.38 | 0.38 | yes | 0.38 | 0.38 |
| Time deposits | 0.72 | 0.72 | yes | 0.32 | 0.66 |
The offered rate betas are recovered from the rate sheet over the months no floor binds and checked against the generator's known parameters. The paid rate betas come from interest actually credited, and they are what net interest income feels.
The retention list: expected balance at risk
The 500 customers at the top of the list hold $114,622,517 and carry $9,442,588 of expected balance at risk, out of $61,088,783 across all 39,200 checking customers. The list is ordered by probability times balance, so a call goes to the relationship worth keeping.
| Customer | Probability of leaving within a year | Balances | Expected balance at risk | Why |
|---|---|---|---|---|
| C037083 | 6.7% | $1,463,425 | $98,069 | checking balance; savings and time balances |
| C029680 | 19.5% | $362,451 | $70,606 | has complained; does not use digital banking |
| C041637 | 5.5% | $1,271,418 | $70,198 | checking balance; savings and time balances |
| C044343 | 7.5% | $777,371 | $58,211 | does not use digital banking; savings and time balances |
| C047468 | 7.9% | $709,516 | $55,974 | does not use digital banking; savings and time balances |
| C040435 | 6.9% | $788,264 | $54,449 | checking balance; savings and time balances |
| C034336 | 9.0% | $574,618 | $51,818 | does not use digital banking; savings and time balances |
| C002484 | 8.0% | $644,511 | $51,623 | does not use digital banking; checking balance |
| C009648 | 18.3% | $279,663 | $51,287 | customer for under a year; does not use digital banking |
| C035962 | 5.8% | $868,795 | $49,987 | checking balance; savings and time balances |
| C019831 | 9.2% | $534,437 | $48,949 | does not use digital banking; time as a customer |
| C019661 | 7.6% | $634,302 | $47,978 | does not use digital banking; savings and time balances |
| C040197 | 8.7% | $545,563 | $47,601 | does not use digital banking; savings and time balances |
| C002401 | 20.5% | $231,661 | $47,550 | has complained; does not use digital banking |
| C011480 | 37.3% | $121,097 | $45,227 | has complained; customer for under a year |
Source: generated, Harborline deposit accounts, September 2023 to August 2026, seed 20260831, as of 2026-08-31. The model and its gates are in the attrition validation report.
Concentration
| Largest depositors | Customers | Share of deposits |
|---|---|---|
| Top 1% | 392 | 12.6% |
| Top 5% | 1,960 | 32.5% |
| Top 10% | 3,920 | 46.7% |
Product holding
| Holding | Share of customers |
|---|---|
| Checking | 95.2% |
| Savings | 52.2% |
| A time deposit | 12.0% |
| Any credit product | 56.1% |
| Checking and savings | 48.0% |
| Checking and any credit product | 53.4% |
2.29 product types per customer on average.
Method and limitations
- The customers, balances and rate sheet are generated. The betas are recovered from a generator that was written with them, which proves the pipeline, not the betas of any real bank.
- The attrition model's AUROC of 0.583 is close to the ceiling of 0.588 that the generator's own hazards reach. A real bank has no ceiling to measure against and should expect less.
- A customer who keeps the checking account open and moves the salary elsewhere has left in every way that matters, and the closure label does not see it.