Banking AI agent · SAR filing
Suspicious-Activity Reporting AI Agent
Assemble the case behind an escalation, draft the Part V narrative and run the filing clock from a date a BSA officer entered — who alone determines that suspicion exists.
- Typical delivery
- 4–6 weeks
- Deployment
- Your stack
- Officer decides
- Pre-filing
- After launch
- Agent Care
What this agent does
Builds the file the determination rests on
In
- 01
An alert is escalated, and the transactions, the customer file and any prior filing are pulled in behind it.
- 02
A case is normalised across the monitoring, payment and customer systems, each fact carried with its source.
Reason
- 03
A review concludes, and the officer's determination and initial-detection date are entered, never inferred.
- 04
A case is tiered under 12 CFR 21.11, whose amounts and the 30- and 60-day clocks did not move in 2025-26.
- 05
A narrative is drafted to stand alone — who, what, when, where and why, with method of operation noted.
Decide
- 06
A clock runs forward from the entered date toward the 60-day ceiling, which no investigation may extend.
- 07
A case reaches the named BSA officer, who determines whether the bank suspects and whether it files.
Out
- 08
A decision is retained with the case for five years — including a no-file record where your own rule wants one.
- 09
Writes run only inside the approval boundaries agreed at implementation, and never outside the perimeter.
Product statement
The agent assembles and drafts; the named BSA officer determines, signs the filing, and personally carries that exposure.
Example workflow
One case, escalation to filing
- 1Escalation receivedFraud-alert triage, KYC review, branch referral or monitoring system
- 2Case assembledTransactions, counterparties, account history and prior filings, each with its source system
- 3Narrative draftedPart V draft, Field 2 key terms, open questions and confidence
- 4Controls appliedCompleteness checks, 12 CFR 21.11 tier checks, key-term currency and confidence threshold
No human action required
Stages 1 to 4 run unaided, and nothing is determined at any of them — the agent is assembling, and the officer's lane opens at the confidence gate.
Goes to the BSA officer to determine.
Adds a second investigator read first.
Officer determination
The case is held with its evidence, its open questions and the confidence.
Approve · Amend · Send to second review
Determined — filing authorised
- 6Case system updatedOnly where write access and approval policy allow it
- 7Outcome evaluatedNarrative completeness, amendment rate, days to file and examiner findings
Amendments
A filing is amended, a closure reopened — both land on the officer's record.
What should not run autonomously
Human approval stays in control
Outside the boundary — human approval required
8 items
- Determining that the bank knows, suspects or has reason to suspect.
- Setting the initial-detection date the clock runs from.
- Signing and submitting the report to FinCEN.
- Deciding not to file — the safe harbour does not cover that.
Automation boundary
Agent acts unaided
- ✓Assemble the case from the transaction, customer.
- ✓Draft a Part V narrative that reads without the attachments.
- ✓Run the filing clock forward from the date the officer entered.
- ✓Check the draft for completeness and hold it for the BSA officer.
Writes stay inside the boundaries set at implementation, and inside the confidentiality perimeter.
- Notifying the board or its designated committee.
- Answering a subpoena or any external request for a filing.
- Opening or answering a 314(b) exchange.
- Changes to review, documentation or escalation rules.
Example output
One case, annotated
Everything the agent assembles is attached to the case it was drawn from.
SAR assembly output · single case
Illustrative example- Case
- Cash structuring referral
- Drafted line
- Deposits sit below the cash reporting threshold across several branches
- Tier cited
- 12 CFR 21.11(c)(2)
- Source of date
- Officer-entered
- Confidence
- 89%
- Clock
- Runs from the entered date
As received
Taken from the case, payment and customer systems — nothing on this side is written by the agent.
Evidence used
- Branch deposit history
- Prior filing on subject
- Customer risk record
Why this tier
It is the tier the banking-agency rule sets and a person still decides.
What the score decides
Below the configured threshold the case picks up a second investigator read before it.
Action
ApproveAmendSend to second reviewValue
Where AI adds value
The same four claims, placed at the point in the workflow where each one applies.
Where the value lands
Value 01 – 04
Every escalation
From the case record
Take what the other agents defer
Pick up the escalations the fraud-triage and KYC agents hand over, and build the file a determination needs.
Detection is not determination
The clock starts when a review concludes, not when a flag fires. That rule has not moved.
Documented to your rule
FinCEN's October 2025 FAQs require no no-file record; the FFIEC manual still asks for one; an order can compel it.
Build an evidence trail
The activity, the review it went through and the officer who determined stay on the filing.
Integrations
Typical integrations
Five system groups connect to the same agent. Which of them are in scope is decided in discovery.
Monitoring systems
- Actimize
- Verafin
- SAS
- Oracle FCCM
Case management
- Unit21
- Hummingbird
- Sardine
- Alloy
Core banking
- FIS
- Fiserv
- Jack Henry
- Temenos
- Finastra
Agent
Suspicious-activity reporting
- Reads the case
- Drafts the narrative
- Holds for determination
Filing and records
- BSA E-Filing portal
- Records archive
- retention
Observability & evaluation
- OpenTelemetry
- Langfuse
- Supported monitoring/evaluation sources
Integration availability depends on the client's existing systems and API access.
Agent controls
Six layers between the model and the filing
Six controls, outermost first. What the whole set lets past is written into the map below.
L6 · Outermost — last line of defence
Inward → L1 · closest to the model
- L1 – L2
- Test whether a draft may stand
- L3
- Puts the determination in the officer's hands
- L4 – L5
- Keep the activity and the review behind it
- L6
- Holds the filing at undecided when signals degrade
How Nestack evaluates it
Evaluate the assembly workflow — not only the finished narrative.
Coverage runs the whole depth of the workflow, and every layer is cut by slice.
Surface — the narrative the officer signs
- E1Final-output evaluationDid the narrative carry who, what, when, where and why?
- E2Step-level evaluationDid the agent use the right case, the right tier rule and the current key-term list?
- E3Tool evaluationDid it read and write the correct case and the correct field?
- E4Confidence calibrationDo low-confidence cases actually attract more officer amendments?
- E5Slice evaluationHow does performance change across specific typologies?
- E6Business outcomeHow many filings were amended, and how many closed cases were reopened?
Floor — the outcome the bank answers for
Failure modes
Where each failure originates in the agent
Seven failure modes, placed at the stage each one originates.
Agent lifecycle
Direction of processing →
01 · Retrieval
1 mode
- ES-03
Stale key-term list
The advisory key-term list was never refreshed.
Stage gathers
Case evidence, prior filings and the key-term list
02 · Reasoning
2 modes
- ES-04
Detection date inferred
The alert timestamp is read as initial detection.
- ES-06
Narrative not standalone
Part V leans on documents FinCEN never receives.
Stage proposes
The draft, the tier, the clock and the confidence
03 · Tool / write
2 modes
- ES-02
Disposition by throughput
Review capacity is sized to what the agent emits.
- ES-05
Continuation filed early
Built on a filing rule instead of a review period.
Stage writes
Only where write access and approval policy allow it
04 · Output
1 mode
- ES-01
Wrong subject reported
The narrative names a subject the resolution got wrong.
Stage returns
The narrative the officer signs and FinCEN reads
05 · Change / Version
1 mode
- ES-07
Silent review regression
A model or rule change narrows what is escalated.
Stage tracks
Model, prompt, key-term list and review rules
- Sev-1 · disclosed outside the perimeter
- Sev-2 · a wrong narrative reaches FinCEN
- Sev-3 · evidence degrades, case holds
Affected slices
One typology can carry most of the failures
A programme-wide filing rate can look settled while a few typologies absorb most of the amendments and late filings. Nestack reports the amendment rate by typology, not only in total.
Slice performance — reported separately, not only in aggregate
Illustrative example| Slice | Failure rate | Lift | Lift vs. threshold | Status |
|---|---|---|---|---|
| Continuing-activity filings | 9.7% | 3.8× | Review | |
| Correspondent and cross-border | 7.8% | 3.1× | Review | |
| Elder financial exploitation | 4.6% | 1.8× | Watch | |
| Structuring and cash | 2.0% | 0.8× | Normal |
Bar: amendment-rate lift vs. the structuring-and-cash baseline · scale 0–4.0× · tick marks the 2.0× review threshold
2 of 4 slices over threshold
Evidence-linked improvement
No cycle closes without a new test
A cycle closes when the late filing is a case the next release has to catch. That suite is what the next narrative drafted is measured against.
Improvement cycle · five stages
Switchback — the path turns at Improve and returns at Learn
- 01
Detect
No-file records fall short of the bank's own rule.
- 02
Diagnose
The case nobody called suspicious for weeks is read back until the cause narrows to one.
- 03
Improve
The fix carries a number, and the filings that forced it travel with it.
- 04
Verify
Nothing goes out until every touched narrative case has passed again.
- 05
Learn
It is kept for good, and the review rules are amended in the same commit.
Learn → Detect
The return edge. The next detection is measured against a suite one case longer.
Typical build scope
Twelve workstreams across six weeks
The build scope read against the delivery timeline. Week structure follows the six-week plan — discovery, sources, review workflow, evaluation, integration, then production validation and handover.
Workstream
Week 1Week 2Week 3Week 4Week 5Week 6
- 01SAR workflow discovery and boundary definition.Week 1
- 02Case and monitoring source assessment.Week 1, Week 2
- 03Tier, clock and key-term rule mapping for filings.Week 1, Week 2
- 04Case ingestion and evidence normalisation.Week 2
- 05Narrative drafting and source binding.Week 2, Week 3
- 06Confidence scoring and exception routing.Week 3
- 07Officer determination workflow.Week 3, Week 4
- 08Case-system and filing integration.Week 5
- 09Narrative and clock cases.Week 4, Week 5
- 10Guardrails and determination controls.Week 4
- 11Case-trail instrumentation.Week 5, Week 6
- 12Deployment, documentation and Agent Care handover.Week 6
12 workstreams · 6 weeks · bar shows the weeks a workstream is active — several run in parallel
Final scope and sequence confirmed in discovery
Engagement tiers
What each tier includes
Rows are the capabilities named in each tier's scope. Higher tiers include everything below them.
| Capability✓ in scope · — not at this tier | PilotOne typology, one team | ProductionProduction case systems | AdvancedMultiple entities / regulators |
|---|---|---|---|
| Introduced at Pilot | |||
| Assembly to your case record | ✓ | ✓ | ✓ |
| Officer determination | ✓ | ✓ | ✓ |
| Narrative-completeness baseline | ✓ | ✓ | ✓ |
| Introduced at Production | |||
| Reporting by typology | — | ✓ | ✓ |
| Determination workflow in your systems | — | ✓ | ✓ |
| Approved write-back | — | ✓ | ✓ |
| Case-system integration | — | ✓ | ✓ |
| Introduced at Advanced | |||
| Multi-entity rule sets | — | — | ✓ |
| Multi-stage BSA approvals | — | — | ✓ |
| High case volume | — | — | ✓ |
| Multi-entity filing controls | — | — | ✓ |
| Build price | From $5,000 | From $8,000 | Custom quote |
- Final build price
- Confirmed after discovery based on integrations, workflow complexity, transaction volume, approval controls and deployment requirements.
- Separate from build
- Build pricing is separate from recurring Agent Care, which covers managed monitoring, evaluations, incidents and verified improvements after launch.
What we need from you
What you bring, and what we build with it
Each input maps to a piece of build scope and a week in the delivery timeline.
You bring
We build with it
01Your escalation path and case record structure
→Case ingestion and evidence mapping
Week 102Representative filed and closed cases
→Assembly baseline, narrative extraction and source binding
Week 203Your review, tiering and documentation policy
→Review-policy mapping and automation-boundary definition
Week 104Access to relevant APIs, feeds or exports
→Case and monitoring assessment, then integration setup
Week 205Filings you would not want read back
→Timeliness cases and failure-mode testing
Week 406What no narrative may leave out
→Confidence scoring, exception routing, guardrails and approval controls
Week 307Named BSA officers to determine on cases
→Officer determination workflow, then pilot and production validation
Weeks 5–6
Nothing else is required
Deployment, documentation and Agent Care handover are ours.
Delivery timeline
Four phases across six weeks
The bands are drawn over the weeks each phase genuinely needs, which is why the fifth carries two.
Phase
W1
W2
W3
W4
W5
W6
- DiscoveryW1
- BuildW2 – W3
- EvaluateW4 – W5
- Pilot & LaunchW5 – W6
Week focus
W1SAR workflow discovery, policy mapping and the automation boundary
W2Case-system integration and the assembly baseline
W3Assembly workflow, confidence logic and determination controls
W4Evaluation suite, guardrails and failure-mode testing
W5Filing integration, pilot cases and targeted corrections
W6One review cycle run under the BSA officer, then Agent Care handover
- Reading the band
- Each bar spans only the weeks its work is named in. The week 5 overlap is two phases running, not padding.
- At the end of W6
- Validation closes on live cases, and Agent Care picks up monitoring.
- Duration
- Six-week plan shown · typical delivery 4–6 weeks depending on scope confirmed in discovery.
Next step · Banking AI agent
Build a SAR agent around your bank's determination chain.
Show us your escalation path, your case system and who signs the filing. If your fraud and KYC teams already stop at reporting, this is the page that picks it up.
- Typical delivery
- 4–6 weeks
- Deployment
- Your case system
- Officer decides
- Pre-filing
- After launch
- Agent Care available
Nestack Agents · Suspicious-activity reporting
AGT-BK-13 · Agent Care available after launch