Check fraud intelligence and check-review operations.
Explainable check fraud intelligence for financial institutions
See exactly why a check was flagged, route it to the right reviewer, and preserve a complete audit trail — without a black-box fraud score.
- Signal-level
- score decomposition
- Named reviewer
- on every decision
- Full audit trail
- per case
- Routing checksumABA valid · 0 pts
- Amount consistencyNumeric ≠ written · +22
- Duplicate presentmentImage match 0.94 · +18
- Image qualityLegible · unscored
Illustrative example. Every signal carries its own evidence, weight and confidence, and a named reviewer records the binding decision.
How it works
Submit → Analyze → Score → Review → Decide
- 1
Submit
Check, amount, channel and capture enter the system.
- 2
Analyze
Check-specific modules inspect signals and supporting evidence.
- 3
Score
Signals produce an explainable 0–100 risk assessment.
- 4
Review
Analyst sees what was flagged, why it mattered and what needs attention.
- 5
Decide
Reviewer records the disposition and rationale, kept in audit history.
What the analyst sees
Anatomy of a flagged check
Every flagged case opens onto the same structure. Nothing in it is generated prose — each line is the evidence the analysis modules actually recorded.
Risk score
One 0–100 number, with the risk level and the queue it routed to.
Signal decomposition
Each signal that was present, and the weight it contributed to the score.
Why each signal mattered
The supporting evidence recorded by the module, in plain language.
Which signals counted
Scored signals are listed separately from findings held out of the score.
Which signals were experimental
Experimental and never-scored categories are labelled and contribute zero points.
Module and version provenance
The module key and version that produced each signal, plus the scoring model version.
Human reviewer decision
The named reviewer, the outcome and the rationale — visually separated from the automated recommendation.
Immutable audit history
The full event trail, including superseded analysis runs retained rather than overwritten.
Why institutions can trust it
What is proven, and what is not
Each claim below carries its own status. Nothing is marked verified because it sounds good — only because the product or a regression run demonstrates it today.
Explainable scoring
Every point of the 0–100 score traces to a named signal with its evidence, confidence, weight and the module version that produced it.
Human decision authority
Analysis recommends. A named reviewer records the binding outcome with rationale, and the recommendation and the decision are shown separately.
Check-specific intelligence
Image-derived MICR checksum, amount agreement between the image and the submitted value, and duplicate-image presentment are calibrated into the score. Layout and forensic findings are reported but deliberately unscored.
Fast sandbox evaluation
A synthetic check can be submitted and analysed in the sandbox in minutes, using the same pipeline production uses.
API-ready integration
Versioned REST surface with scoped machine credentials, idempotent submission and signed webhooks, sandbox-first.
Audit-ready case history
Submission, analysis, assignment, escalation, disposition and re-analysis are all recorded with actor and timestamp, append-only.
What you can put in front of an examinerSee details
- The signals that were present on the item, and the evidence recorded for each.
- The contribution each scored signal made to the 0–100 score.
- The scoring model version and the module version behind every signal.
- The limitations that applied — including capture quality and missing intake information.
- The reviewer actions taken: claim, escalation, information requests.
- The final human decision, its rationale and who recorded it.
- The complete append-only audit history, including superseded analysis runs.
This is evidence availability, not regulatory certification. CheckGuard claims no examiner approval, regulator endorsement or compliance certification.
Why institutions choose CheckGuardSee details
- Explainable scoring: the decomposition is the product, not an export.
- Check-specific intelligence rather than a generic anomaly model.
- Human-in-the-loop by design — automation never closes a case.
- Strong auditability: append-only case history and retained superseded analysis.
- Fast sandbox evaluation on synthetic data, no production access required.
- API-first integration with scoped credentials and signed webhooks.
- Transparent assurance status, including what is not yet verified.
- Modular architecture: detection modules plug in without changing the workflow.
Who it is built for — and what CheckGuard is notSee details
Built for
Check fraud intelligence and check-review operations.
- Community banks
- Regional banks
- Credit unions
- Fintechs
- Sponsor-bank programs
- Deposit / remote-deposit-capture operations
Not what CheckGuard is
- Not an AML suite
- Not a transaction-monitoring platform
- Not a wire or ACH fraud platform
- Not an identity platform
- Not a replacement for every fraud system you run
Integration, capabilities, FAQ & pilot pricingSee details
CheckGuard integrates through a versioned API with scoped machine credentials, sandbox and production environments, idempotent submissions and signed webhooks. Scope, limits, pricing structure and readiness questions are documented in full.
Design-partner programme
Run a sandbox pilot on your own check scenarios
Institutions start in a sandbox institution with synthetic captures, then evaluate explanations, routing and audit history before any production traffic.