Trintech's finance agents execute inside your controls

Most "AI for finance" is a dashboard with a summary on top. Trintech went the other direction. On September 28 the company launched three new AI agents at its Connect 2026 conference — Data Access, Accruals Intelligence, and Exception Management — built to do close work rather than describe it. The design choice worth stealing is that every action lands inside the approvals and audit trail the finance team already runs on.

What actually happened

The three agents extend a set Trintech started earlier this year with its Flux and Variance Analysis agents, taking the count to five.

The Data Access Agent connects to, retrieves, cleans, and reshapes financial data from multiple systems — including the ones nobody wants to touch — and delivers it as a governed pipeline with visibility into what moved and what changed. The Accruals Intelligence Agent reviews historical activity and prior-period patterns to recommend accrual amounts, flags estimates that fall outside historical trend, and tracks accuracy over time by preparer, account, and methodology. The Exception Management Agent identifies and classifies exceptions as they surface, investigates root cause, attaches a confidence score, groups related exceptions, and keeps the trail.

Trintech CEO Darren Heffernan framed it plainly: finance teams "do not need more AI that simply tells them what to do. They need AI that does the work — in ways they trust." Every recommendation carries evidence, methodology, and approvals. Availability is through existing account managers; no pricing was disclosed.

Why governed finance agents matter for your business

You are not running Trintech. You are running QuickBooks or Xero, a spreadsheet nobody has rebuilt since 2019, and a close checklist that lives in someone's head. The lesson still applies, because the three agents map to the three things that actually eat your month-end.

Pull the pattern apart. Data access is the boring 60% — exporting, normalizing, reconciling formats across a bank feed, a payment processor, and a POS. That is deterministic work; script it before you point a model at it. Accrual estimation is judgment, and the interesting detail is that Trintech tracks accuracy by preparer and by methodology over time. That scoreboard is the control. An agent that estimates without a hit-rate record is just a confident intern. Exception handling is where the money is, and grouping related exceptions matters more than resolving them individually — twelve exceptions with one root cause is one fix, not twelve investigations.

The architectural rule underneath all three: the agent runs inside your approval flow, not beside it. If your agent writes a journal entry, the entry needs the same approver, the same evidence attachment, and the same immutable log as a human's. The moment the AI path is the fast path that skips review, you have built a control gap and called it automation. Track what the agent proposed, what a person approved, and what shipped — three separate fields, queryable by you.

Sources: Trintech press release (PR Newswire), Trintech agentic AI platform.