Skip to main content

​
Submissions

A submission is the top-level container — typically one per loan application, deal, or merchant under review. It holds documents, transactions, and analytics. Lifecycle: new → ready → processing → processed → reviewed Two statuses sit outside that happy path: failed when processing errors out, and review_requested when a submission is flagged for manual attention. The SDKs’ waitForProcessing / wait_for_processing helpers stop on processed, reviewed, or failed.

​
Documents

Documents are files attached to a submission — bank statements, tax returns, P&Ls, balance sheets, ISO applications, voided checks, IDs. Processing pipeline:
  1. Upload — File stored, basic metadata extracted
  2. OCR — Text extraction from PDF/image (Tesseract, Google Vision, or custom)
  3. Classification — ML model identifies document type
  4. Parsing — Type-specific parser extracts structured data (fields, tables, transactions)
Document classes:
  • bank_statement — Monthly/quarterly bank statements
  • tax_return — Federal/state tax returns (1040, 1120, 1065)
  • pnl — Profit & loss / income statements
  • balance_sheet — Balance sheets
  • iso_application — ISO/MCA applications
  • voided_check — Voided checks
  • identity — Government-issued ID
  • financial_statement — Generic financial documents
  • other — Unclassified

​
Transactions

Transactions are the atomic financial records extracted from bank statements or imported via Plaid/API. Each transaction goes through enrichment:
  • Cleaning — Normalize description, remove noise
  • Categorization — Assign to category taxonomy (revenue, expenses, transfers, etc.)
  • Merchant detection — Identify merchant name, MCC code
  • Recurring detection — Flag subscriptions, rent, payroll, loan payments
  • Duplicate detection — Find matching/duplicate transactions across accounts

​
Enrichment

The enrichment pipeline transforms raw transaction descriptions into structured, categorized data:
  1. Clean — Strip check numbers, reference codes, trailing digits
  2. Classify — ML model assigns primary + secondary categories
  3. Merchant — Fuzzy match against merchant database, extract MCC
  4. Recurring — Pattern detection for periodic transactions
  5. Anomaly — Flag unusual amounts, patterns, or timing

​
Categories

Moneyline uses a hierarchical category taxonomy:
  • Revenue — Sales, service income, refunds, interest
  • Expenses — Operating, payroll, rent, utilities, marketing
  • Transfers — Internal transfers, owner draws, loan proceeds
  • Fees — Bank fees, NSF/overdraft, wire fees
  • Other — Uncategorized transactions
Each category has a slug, name, and group (revenue/expense/transfer/fee).

​
Analytics

Analytics are computed from enriched transactions:
  • Scorecard — Multi-metric financial health assessment with grades
  • Cashflow — Revenue, expenses, net income by period (day/week/month/quarter)
  • Balances — Daily ending, minimum, and average balances
  • Risk Score — ML-generated risk assessment (0-100)
  • Anomalies — Flagged unusual transactions
  • Forecasts — Category-level projections
  • Recurring — Identified recurring transaction patterns
  • Merchant Summary — Spend breakdown by merchant
  • Statistics — Summary metrics (totals, averages, growth rates)

​
Policies

Policy rules are configurable conditions evaluated against analytics:
Rules are evaluated per-submission and produce pass/fail/warn results with severity levels.

​
Integrations

  • Plaid — Direct bank account connection, transaction import
  • Finicity — Alternative bank data provider
  • Yodlee — Additional bank data provider
  • Decision Logic / Ocrolus / Inscribe — Document verification
  • Salesforce / HubSpot — CRM sync (push analytics + status)

​
Webhooks

Real-time event notifications for:
  • submission.created, submission.processed, submission.failed
  • document.uploaded, document.classified, document.parsed
  • transactions.enriched
  • policy.evaluated
All webhook payloads include HMAC-SHA256 signatures for verification.

Next

Put it to work

The data model makes more sense once you have run a document through it end to end.

Run the quickstartAPI reference