Float Launches Agentic AI for Corporate Card Bookkeeping

Float, the Toronto-based FinTech founded in 2019, released Float Intelligence, an agentic AI automation layer that integrates with its corporate cards and expense stack to automate transaction coding and Canadian tax assignment. The offering bundles Float's existing automation with a new transaction coding agent that assigns general ledger lines and tax codes for HST, GST, and PST. Beta testing across 350+ Canadian businesses returned more than 90% accuracy on auto-coded transactions, with lower-confidence items routed for human review.

Float says it trained a LLM on hundreds of thousands of transactions from Canadian vendors and tax contexts to achieve high precision. Key capabilities reported include:

• •automatic assignment of general ledger account lines
• •automatic tagging of Canadian tax codes (HST, GST, PST)
• •confidence-threshold flagging that routes uncertain cases for human verification

Float positions the agent as precision-first rather than permissive automation, citing finance-specific correctness requirements. The company did not publish model architecture, latency, or on-prem/off-prem deployment options, and access is gated by customer service tier.

Expense coding and tax allocation are high-volume, repetitive tasks that traditionally consume bookkeeper hours and produce error-prone ledgers. By training on region-specific tax and vendor patterns, Float reduces the domain shift that generic models face when applied to localized finance workflows. The reported 90%+ auto-code accuracy on confident outputs is credible as a production threshold that materially reduces manual work while preserving auditability through human-flagging on low-confidence items.

Monitor real-world error modes in mixed vendor contexts and cross-border transactions, Float's transparency on model updates and drift mitigation, and whether the product expands beyond Canadian tax regimes. Also watch pricing and tier gating, which will determine adoption among SMEs and mid-market finance teams.

"Finance is not like programming or design, where you can sort of vibe code things together; you have to be really precise and accurate," said Ruslan Nikolaev, head of product, underscoring the product's precision focus.