AI-Powered Bank Statement Reconciler

Problem This Solves

Reconciliation often breaks down when bank statements, payment platforms, and orders do not line up: Chinese bank exports from BOC, ICBC, CCB, and ABC use different CSV/Excel columns, while Alipay, WeChat, PayPal, Stripe, Amazon, Shopify, and Temu each have their own schema. Delayed settlements, batch payments, and inconsistent counterparty names can defeat simple “same date plus same amount” checks, and manual review easily misses unclaimed bank entries and unmatched orders.

How It Works and What to Watch

The skill ingests bank statement and order files, identifies key fields such as transaction date, transaction amount, counterparty account, balance, and memo, then applies matching modes:

• Exact Matching: same date and exact amount, best for real-time transfers;
• Fuzzy Matching: configurable via date_range_days and amount_tolerance, useful for delayed settlement;
• Semantic Matching: approximate counterparty-name similarity, for cases like “Alibaba / Alibaba Cloud” or “Zhang San / Zhang San (individual)”.

Results are grouped into matched, differences, unclaimed, unmatched, and summary, with Excel export. Discrepancies can be marked processed, pending collection, or bad debt. Keep in mind that unsupported formats may need conversion, and missing columns, amount parsing issues, date format mismatches, or tier limits can trigger COLUMN_NOT_FOUND, AMOUNT_MISMATCH, DATE_PARSE_ERROR, or TIER_LIMIT_EXCEEDED. Semantic matching, PayPal/Stripe support, and higher-tier outputs are capability-dependent.