Trained exclusively on real Canadian transactions, Float’s first AI agent outperforms general-purpose LLMs by 28 percentage points, giving businesses back hours they can’t afford to lose in the current economic climate.
TORONTO, April 21, 2026–(BUSINESS WIRE)–Float today launched Float Intelligence, an AI and automation layer embedded across the finance platform where more than 7,000 Canadian businesses manage payments, optimize cash flow and close their books. Its first capability, a transaction coding agent, automatically assigns GL codes and Canadian tax codes—HST, GST and PST—to corporate card transactions, turning what used to require hours of line-by-line coding into a process that takes minutes.
“We didn’t set out to build an AI product. We set out to remove the everyday friction caused by Canadian businesses being handed infrastructure that was never designed for them,” said Rob Khazzam, Co-Founder and CEO at Float. “We spent six years fixing that foundation. Float Intelligence is what becomes possible once it’s right.”
The launch comes as Canadian businesses face a deepening efficiency squeeze. A recent Float report found that while revenue grew 5% last year, profits declined, cash reserves dropped nearly 5% and debt stayed flat. Businesses are burning reserves to stay operational, not borrowing to grow. In a country where half of SMBs spend up to 40 hours a month on payments and reconciliation alone, Float Intelligence is designed to give these businesses the advantage they need now, not by adding capital, but by giving finance teams back the operational capacity to deploy it more strategically.
Float Intelligence first tackles transaction coding
The first Float Intelligence capability is a transaction coding agent, which automatically assigns GL codes and a range of tax codes to card transactions. Designed specifically for the Canadian context, Float’s agent handles the complexity of provincial tax structures, the variance in how Canadian businesses categorize spend and the accuracy threshold finance teams require before trusting a system to act on their behalf.
-
Confidence-gated coding: The system only auto-codes when confidence meets Float’s 90% accuracy threshold. When it’s not sure, it routes the transaction for human review rather than guessing.
-
Self-improving and hyper-personalized: Instead of a one-size-fits-all approach, the model powering the agent is calibrated to each business’s specific chart of accounts and coding history, and learns from every correction or modification made by the user.
-
Built for privacy: Float runs all inference in its own secure AWS instance, not through third-party LLM providers. No transaction data or coding decisions are shared between businesses. Each customer’s model is calibrated on their data alone.
