AI-powered crypto accounting and Xero reconciliation console

The Challenge

Closing the books on a web3-active client is the single largest cost driver in every engagement at firms that take on crypto. Wallets across multiple chains, multi-asset transactions, FX pulled from five different sources, internal transfers that read as expenses unless someone catches them. The work is manual, error-prone, and impossible to scale without burning analyst hours.

Our Approach

Crypto reconciliation is hard because every wallet address is an unlabeled black box by default. We solved that with on-chain identity correlation: a counterparty resolution model maps each wallet to its real-world identity (centralized exchange, DEX router, custodian, internal wallet, known DeFi protocol) by combining on-chain signals (address labels, transaction patterns, contract interactions) with client-provided context on wallet identity and purpose. AI categorization then suggests Xero account codes per row, internal transfers auto-flag, and the accountant just confirms. Behind the scenes, the pipeline ingests across EVM (6+ networks), Solana, and TRON through a unified provider, splits multi-asset transactions into one row per movement, and converts every figure to the client's base currency using daily central-bank reference rates. Selective publish to Xero with duplicates blocked and a public audit page linked from every entry back to the on-chain proof.

Where AI does the work

• On-chain identity correlation maps every counterparty wallet to a known entity (CEX, DEX router, custodian, internal wallet, DeFi protocol) using address labels, transaction patterns, contract interactions, and client-provided wallet context.
• AI categorization suggests Xero account codes and tax codes based on counterparty type and transaction shape, so the accountant confirms instead of types.
• Internal transfer detection auto-flags wallet-to-wallet movements within the same client's portfolio so they don't book as expenses.
• Per-client learning turns every reviewer override into training signal, so the model gets better at each client's chart of accounts over time.