Before You Buy AI for the BSA Alert Queue, Write Down What 240 Alerts a Month Costs You Today

Can You Say, Out Loud, What an Alert Costs You Today?

Ask your BSA officer how many alerts came out of the monitoring system last month and you will get a number. Ask how many minutes the average one takes to clear and you will usually get a pause, then an estimate, then a caveat about how it depends. That pause is the reason most community banks cannot tell their board whether anything they bought in 2025 worked.

Deloitte's State of AI in the Enterprise 2026 found that 84% of organizations investing in AI report positive returns. The interesting part is not the percentage — it is the pattern underneath it, which is boringly consistent: pick one messy process, keep a human reviewing the output, and prove time saved or errors reduced before you widen the scope. Banks that skipped the measurement step are in the other 16%, and they generally cannot explain why.

For a community bank between $250 million and $1 billion in assets, the single best process to measure first is the BSA alert queue. Not because it is glamorous, but because it is high volume, mostly repetitive, fully logged already, and staffed by the person on your org chart with the least slack in their week.

Start With the Queue Nobody Wants: BSA Alerts

Whatever you run — Verafin, Abrigo BAM+, Fiserv's financial crime module, or rules built into the core — the monthly rhythm is the same. Structuring alerts on a cash-intensive customer. Rapid movement of funds through a new business account. Wire activity out of pattern for a customer who has banked with you for eleven years. Cash deposits at $9,400 three days running from the car wash on Route 6.

The BSA analyst opens each alert, pulls ninety days of account history, looks at the related accounts, checks whether the customer's stated business explains the activity, reads the last alert on the same customer, and writes a disposition. Most of the time the disposition is "no further action, activity consistent with known business." Sometimes it becomes a case, and sometimes a case becomes a SAR, which has to be filed within 30 days of initial detection and needs a narrative that will hold up when an examiner reads it two years from now.

At a $600 million bank, the BSA officer is frequently also the compliance officer, and sometimes also the person who handles Regulation E disputes. The queue is where the month goes.

Capture the Baseline Before You Switch Anything On

This is the step everyone skips and the only one that makes the ROI argument real. Spend one month recording these, by hand if necessary, before any assistant touches the queue.

• Alerts generated per month, split by rule that fired.
• Minutes per alert from open to disposition. Have the analyst note start and stop times for three weeks. It is annoying; do it anyway.
• Share closed with no further action. At most community banks this is 80% or higher.
• Alerts escalated to a case, and cases converted to a SAR. This ratio is the quality signal — if it moves after you change the process, you need to know why.
• Average days from alert generation to disposition, and backlog age at month end.
• Count of alerts dispositioned after the 30-day filing window would have started. This is the number that keeps a BSA officer awake.

The measurement that settles the argument is minutes per cleared alert, held against a SAR conversion rate that did not move — because faster only counts if you are still catching the same things.

flowchart TD
  A["Monitoring system fires 240 alerts"] --> B["Assistant pulls 90-day history, related accounts, prior dispositions"]
  B --> C["Draft disposition with cited transactions attached"]
  C --> D["BSA analyst reviews, agrees or overrides"]
  D --> E{"Escalate to case?"}
  E -->|No| F["Close as no further action, logged with reviewer name"]
  E -->|Yes| G["BSA officer builds case, decides on SAR, signs and files"]
  F --> H["Monthly scorecard: minutes per alert, SAR conversion, backlog age"]
  G --> H
  H --> B

Add the Assistant to the Front of the Queue, Not the End

The right place to put it is before the analyst, not after. For each alert, the assistant assembles what the analyst would have assembled anyway — the ninety-day history, the related deposit and loan accounts, the customer's stated business from the account opening record, the prior alerts and how they were dispositioned — and drafts a recommended disposition with the specific transactions cited underneath it.

The analyst then reads instead of digs. On a car wash with four years of clean history and the same deposit pattern every Monday, that is a ninety-second read and a click. On a new money services business three months old with wires to two states it has no stated connection to, the analyst throws away the draft and does the work properly, which is exactly the right allocation of a scarce person's attention.

Note what did not change: every alert is still reviewed by a person, every disposition carries a reviewer's name, and no SAR gets filed without the BSA officer's judgment and signature. That is not caution for its own sake — it is what the FFIEC examination manual expects when an examiner asks how your monitoring program actually works.

The Number You Take to the Board

Assumptions below are illustrative for a $600 million bank. Replace each one with what your baseline month actually showed.

That is 23.3 hours a month back, or 280 hours a year — most of a part-time BSA assistant you were about to post a job for. At a fully loaded $41 an hour that is $11,500, and the honest way to present it is not as a cost saving but as a hire deferred, plus a backlog that no longer runs past the filing clock. Hold the SAR conversion line up next to it. Six filed before, six filed after: the program did not get looser, it got faster.

What Stays With the BSA Officer, Permanently

The decision to file is a judgment about suspicion, and suspicion is not a calculation. The narrative on FinCEN Form 111 has to say who, what, when, where and why in the officer's own words, and an examiner reading a folder of narratives that all sound identical will ask questions you do not want. Write them yourself.

Three other things stay human. The decision to exit a customer relationship, which is a business and reputational call in a town where the customer's cousin sits on your board. Elder financial exploitation, where the tell is often a phone call or a lobby visit rather than a transaction pattern. And model validation — if the assistant starts recommending "no further action" on a rule that used to produce your cases, you want a quarterly look-back that catches it, sampled and documented, before an examiner catches it for you.

Also be plain about what this does not fix: it does not tune your monitoring rules. If 82% of your alerts are noise, the root cause is thresholds set in 2019 for a bank half your current size. Faster clearing of bad alerts is still clearing bad alerts.

Frequently asked questions

Will an examiner object to drafted dispositions?

Examiners care about whether your program is reasonably designed, whether qualified people make the decisions, and whether you can show your work. Keep the reviewer's name on every disposition, keep the source transactions attached to it, and keep the SAR decision with the BSA officer. Bring the change to your board's audit committee and put it in the minutes before you start, not after.

What if we do not have a clean baseline and never have?

Then your first month is the baseline. One analyst, a timer, and a spreadsheet with alert number, minutes, and outcome. Thirty days of that is worth more than any vendor's projected savings figure.

How small is too small for this to be worth it?

If you are under about 60 alerts a month the time savings will not justify the setup and the oversight, and you would do better tuning the rules. The break-even for most community banks sits somewhere around 100 alerts a month with one dedicated reviewer.

Should we start with BSA or with something less risky?

If the idea of starting in a regulated queue makes you uncomfortable, run the identical measurement discipline on Regulation E dispute intake or on new account document exceptions first. The method matters more than the process you pick: one process, human review, baseline first, one number reported monthly.

If your first measured process ends up being something on the customer-facing side instead, CallSphere builds AI voice and chat agents that answer business phone lines and web chat, book appointments, and capture leads — a queue with the same virtue as the alert queue, in that your phone system already logs the before-and-after numbers you need to prove whether it worked. Voice agents belong on routine inbound calls, not on suspicious activity decisions.