Extraction · reconciliation · reporting

Bank statement converter with reconciliation checks

Convert PDF bank statements into structured Excel/CSV data and check whether the numbers reconcile before you rely on them.

Supported pilot PDF intake See sample report Feasibility scoping (other banks)

PDF intake below is founder-led pilot access on supported bank families and validated statement classes—not open public self-serve. Outside that scope, use feasibility scoping—not the same path as live pilot processing.

Request pilot conversation Why teams use it

Live pilot path

Structured extraction & checks in working pilot deployments

Review-first verdicts

Normalized status & review reasons

Branded HTML + JSON

Reports and machine-readable case outputs

Privacy-conscious flow

No default use of files for model training

Conversion, reconciliation, and reporting

Documentation for each step of the workflow—exports, checks, coverage, and how data is processed. No substitute for professional judgment on material decisions.

Who uses GrassLedger

Teams that need structured figures and explicit reconciliation context—not a black box score replacing analysts, underwriters, or counsel.

For accountants and bookkeepers
Convert PDF bank statements into Excel/CSV for cleanup, review, and accounting workflows.
For lenders and review teams
Check whether borrower bank statement packages are complete, consistent, and ready for deeper review.
For forensic and legal review
Generate review-ready statement reports with reconciliation status, limitations, and extracted data.

Founding pilot

Early access, founder-run—not open self-serve

The founding pilot is a limited early-access offer: capped monthly volume, selected institutions, and founder-supported onboarding so expectations stay aligned with what the system actually does today.

Limited pilot capacity each month
Focused on supported bank families first
Guided workflow and review of real outputs with the founder
Room for scoped evaluation requests—by agreement, not implied scale

No public volume guarantee and no implied broad-market coverage beyond what we validate together. Institutions outside the supported list go through a separate feasibility path—see scoping intake.

Coverage

Currently supported

Live pilot processing today—coverage is by institution and product / statement class, not by logo alone. Each entry is a supported bank family with selected, validated statement types—not an assertion about every document the brand might issue.

Bank of America
U.S. Bank
Wells Fargo

Everything else is out of scope until validated. Use feasibility scoping for other institutions or statement types—not the supported PDF intake path.

Pilot intake

Send statements through the pilot

For supported bank families and validated statement classes only. When your app points at the pilot API, submit a PDF below for live pilot processing—founder-led access on an agreed path, not a general public upload utility. Set NEXT_PUBLIC_API_BASE_URL (see repo docs) to match your FastAPI server.

Institution or statement type not on the list? Feasibility scoping intake —not the PDF path above. Need onboarding first? Request a pilot conversation.

Product workflow

How it works

End-to-end flow on the live pilot pipeline—aligned to underwriting and legal review.

1
Upload statement
PDF ingested under controlled handling.
2
Extract structured data
Deterministic paths—not probabilistic fill-in.
3
Run checks
Reconciliation and consistency signals surfaced explicitly.
4
Surface review reasons
Normalized verdicts and reasons for triage.
5
Generate report
Branded HTML plus machine-readable case output.

Why teams use it

Built for credit files—not chat-style document Q&A

Deterministic extraction, explicit reconciliation, and review-first workflows—so outputs stay defensible under challenge, not probabilistic “best guesses.”

Repeatable results

Same inputs yield the same structured outcome—what you need for audit-ready records and repeat exams.

Review-first by design

Normalized verdicts and review reasons—built for committees and counsel, not silent auto-approval.

Evidence-aligned reporting

Human-readable reports stay aligned to structured case data. When an outcome requires review, the HTML report can include a selective visual evidence appendix— trace-backed context alongside the structured case, not a substitute for full forensic analysis.

What you receive

Outputs from each processed statement

Each run returns extraction, checks, review posture, HTML, and structured JSON—built for credit files and professional review. For review-oriented outcomes, reports may also include a selective visual evidence appendix to support analyst review—not full-document forensics, and not generated for routine pass cases by default.

Hard cases stay explicit: they surface with clear review reasons and recommended next actions—they are not silently auto-passed when signals warrant scrutiny.

Structured extraction

Deterministic fields and sections from defined paths—repeatable for downstream systems and file notes.

Reconciliation checks

Summary- and section-level signals when numbers do not reconcile within stated tolerances.

Review reasons

Normalized verdicts, review reasons, and recommended actions for underwriting and legal triage.

Evidence-backed review

For review-oriented outcomes, the HTML report can include a selective visual evidence appendix—context for surfaced issues, not only the status label. Routine pass outcomes do not include this by default.

HTML report

Branded, human-readable report aligned to the structured case—suitable for professional review.

Downloadable output

Machine-readable JSON case bundle alongside the report—for storage, integration, and audit trails.

Contrast

Generic extraction vs. GrassLedger

Same document type—different posture toward consistency and reviewability.

Dimension

Typical AI extraction

GrassLedger

Dimension

How numbers are produced

Typical AI extraction

Model-inferred totals; confidence often opaque to reviewers.

GrassLedger

Deterministic extraction paths; same file yields the same structured result.

Dimension

When totals disagree

Typical AI extraction

Limited, inconsistent signals for reconciliation gaps.

GrassLedger

Explicit reconciliation and finding-level signals for review workflows.

Dimension

Outputs under scrutiny

Typical AI extraction

Often text or classifications with limited context on flagged items—weak auditability when outcomes require review.

GrassLedger

Structured case object, human-readable report, and—for review-oriented outcomes—a selective visual evidence appendix in HTML where helpful. Built for file documentation, not tamper or fraud detection.

Dimension

Fit for professional review

Typical AI extraction

Hard to defend line-by-line in disputes or exams.

GrassLedger

Normalized verdicts and review reasons suited to underwriting and legal review contexts.

GrassLedger assists structured review; it does not replace professional judgment, credit decisions, or legal advice.

Sample output

Case JSON (illustrative)

Representative shape: verdict, findings, guidance, limitations—plus branded HTML on live pilot runs. Illustrative only; field names vary by deployment.

Evidence traces for flagged cases. When an outcome requires review, the HTML report can include a selective visual appendix tied to surfaced issues—not always present, and not a tamper-detection or full redlining product.

Example — REVIEW

{
  "case_meta": {
    "statement_id": "stmt_8f2a…",
    "bank_name": "Example National Bank",
    "statement_type": "checking",
    "period_end": "2024-09-30",
    "page_count": 5
  },
  "verdict": {
    "status": "REVIEW",
    "review_reason": "REVIEW_FINANCIAL",
    "recommended_action": "request_manual_review",
    "report_summary": "Verdict REVIEW: Summary reconciliation requires attention before disposition."
  },
  "primary_issue_summary": "Summary totals do not reconcile to extracted activity at the stated tolerance.",
  "findings": [
    {
      "code": "SUMMARY_RECONCILIATION",
      "severity": "high",
      "title": "Summary reconciliation",
      "detail": "Printed closing balance differs from calculated close based on extracted credits and debits."
    },
    {
      "code": "SECTION_CONSISTENCY",
      "severity": "medium",
      "title": "Section consistency",
      "detail": "Withdrawals subtotal does not match sum of parsed withdrawal lines for this statement."
    }
  ],
  "next_step_guidance": {
    "action_label": "Manual review",
    "primary_instruction": "Obtain a clearer copy or validate line items against the custodian record before credit action."
  },
  "limitations": [
    { "code": "COVERAGE_PARTIAL", "message": "OCR-assisted pages may reduce line-level completeness." }
  ]
}

Outside current coverage

Other banks or statement types

Not on the supported list? Use the scoping form below. We triage feasibility and fit— not implied production support or a roadmap commitment.

Prefer conversation first? Request a pilot conversation—or return to supported PDF intake if your institution is already covered.

Feasibility scoping intake

For institutions or statement types outside current coverage. We assess whether a scoped evaluation is realistic—this is not a promise of support, timeline, or production readiness.

Redacted samples are optional, explicit, and evaluation-only—not default product use, not broad training data, and not generalized product development without a separate written agreement.

Trust

Security, privacy & data use

Sensitive PDFs are handled for professional review—not consumer-app defaults.

Use of your data

Customer documents are processed strictly for verification and reporting.

By default, they are not used for model training or broad product development.

Optional data sharing for improvement

You may provide selected, appropriately redacted material for institution- or format-specific support— explicit, scoped, and under separate agreement, not inferred from ordinary use.

Retention & deletion

Retention windows and deletion procedures are designed around sensitive financial documents. Exact terms belong in your contract and data-processing documentation—not on this page.

Operational posture

Defaults assume PII and account identifiers need controlled handling.
Deterministic verification workflows—not generic document “AI” products.

Attestations

Specific certifications (e.g. SOC 2) are not claimed here. Request trust materials under NDA if procurement requires them.

Next step

Request a pilot conversation

For teams that want founder-led onboarding: walkthrough of real outputs, review workflow fit, and how the founding pilot is run—not a generic sales demo.

This form is for conversation and onboarding. If your institution is outside current coverage, use feasibility scoping instead so we can triage fit without mixing paths.

Pilot conversation request

Business email required. Replies come from hello@grassledger.com within a few business days.

Bank statement converter with reconciliation checks

Conversion, reconciliation, and reporting

Try the converter→