Float launches AI tool to automate Canadian tax coding

Float has launched Float Intelligence, a financial AI tool embedded in its platform, used by more than 7,000 Canadian businesses.

The first feature is a transaction coding agent that assigns general ledger and Canadian tax codes to corporate card transactions. It is designed to handle HST, GST, and PST coding, a task that often requires finance teams to review transactions one by one.

The launch comes as Canadian businesses face growing pressure to reduce manual work in core finance processes. Float cited its own research showing that while revenue grew last year, profits declined, cash reserves fell, and debt levels remained flat. That, it argued, suggests businesses are relying on reserves to stay operational rather than borrowing to expand.

Against that backdrop, time spent on payments and reconciliation has become an increasingly significant issue. Float said half of Canada's small and medium-sized businesses spend up to 40 hours a month on those tasks alone.

First use

The transaction coding agent is the first part of a broader AI and automation layer across Float's finance platform. It auto-codes transactions only when the system reaches a 90% accuracy threshold, routing lower-confidence cases for human review.

The approach is intended to avoid automated errors in a function where finance teams typically want a clear audit trail and a high degree of accuracy. The model is also tailored to each customer's chart of accounts and coding history and adapts to user corrections.

Float processes the system's inference within its own AWS environment rather than through third-party large language model providers. Transaction data and coding decisions are not shared between customers.

Canada focus

Float is positioning the tool around the complexity of Canadian accounting and tax treatment, rather than as a general-purpose AI assistant. Broad, large language models, it argued, do not reflect how Canadian businesses code spending across provinces and categories.

On a benchmark of 5,000 real Canadian transactions, the transaction coding agent achieved 90% precision, compared with 62% for a leading general-purpose large language model tested on the same dataset, according to Float. The company attributed the gap to training on Canadian transaction data and a per-business calibration layer that reflects customer-specific vendor relationships, chart structures, and tax assignment patterns.

"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 Chief Executive Officer, Float. "We spent six years fixing that foundation. Float Intelligence is what becomes possible once it's right."

The emphasis on local market design reflects a broader pattern in business software, as suppliers seek to demonstrate that AI tools trained on sector- or jurisdiction-specific datasets can outperform broader consumer-facing models on narrow finance tasks.

Beta results

During beta testing with more than 350 Canadian businesses, the transaction coding agent reached 95% measured precision on eligible auto-coded transactions, according to Float. Users shifted from manual coding to a review-and-approve workflow.

"Accuracy and trust are non-negotiable for finance teams," said Ruslan Nikolaev, Co-Founder and Head of Product, Float. "Because Float already sits at the centre of how these businesses manage their spend, the model learns from how each customer actually works. If it isn't confident, it stays out of the way."

The product is now available to customers on Float's Professional plan. It activates automatically and begins learning coding patterns from the outset, according to the company.

Float offers corporate cards, expense management, bill payments, and business accounts, and has built its business around finance operations for Canadian companies. The latest launch adds AI-driven workflow automation to that offering at a time when many finance software providers are trying to make automation a practical selling point rather than a standalone product category.

One early user described a reduction in routine month-end work. "Month-end close used to mean four hours of manual GL coding. Now Float's transaction coding agent handles the first pass and it's right over 90% of the time. I review what actually needs my attention and move on," said Wilford Lam, Finance Lead, Layerzero. "We're getting three hours back every single month, time we can use to drive strategic impact across the company. It's one of those features that makes you wonder how you managed without it."