Bank statement automation
Azure AI Document Intelligence and Claude extract transactions from bank statements, check them against the statement balances, and categorize them into Accounting CS import files. Bookkeeping entry time dropped by 90%.
We designed and built an AI agent platform on Microsoft Azure that automates document intake, bookkeeping and tax prep review for an accounting firm, integrated with Canopy, Microsoft 365 and Accounting CS.
Every tax season, staff keyed bank statement transactions by hand, sorted and filed incoming client documents, and checked draft returns against the source paperwork. The firm wanted AI to take on that work without sensitive client data leaving its control, and without giving up review by its own people.
Azure AI Document Intelligence and Claude extract transactions from bank statements, check them against the statement balances, and categorize them into Accounting CS import files. Bookkeeping entry time dropped by 90%.
Each season’s tax documents are classified by AI, matched to the right client, and filed through the Canopy API. 99% are filed with no human touch.
An agent ties draft 1040s back to their source documents and flags discrepancies, so problems are caught before a return reaches senior review.
Staff review and approve agent actions in Microsoft Teams through Adaptive Cards, so people stay in control of what the agents do.
Security & compliance
We designed a PII protection layer using Azure AI Language plus custom rules. It removes SSNs, account numbers and client identities inside the firm’s own Microsoft tenant before any call to a language model, in line with the FTC Safeguards Rule and IRS Publication 4557.
The platform went live in phases, and no agent acted on its own until it had proved itself.
Platform architecture on Azure, the PII protection layer, and the integrations with Canopy, Microsoft 365 and Accounting CS.
Bank statement processing, document classification and client matching, and the tax prep review agent.
Agents ran alongside staff in shadow mode and had to pass accuracy gates before any autonomous action was turned on.
Phased rollout to the firm, with staff training on reviewing and approving agent work in Teams.
If your team spends its busy season on data entry and document handling, we can show you what AI agents could take on, and how to keep client data protected while they do.