Fintech Aurionpro has launched a trade finance platform built around AI agents, as banks explore how far automation can replace manual processing when it comes to core workflows.
The company said its new platform, called Fintra, uses AI agents to automate most of banks’ manual trade finance processes, including reading shipping documents, running compliance checks, recommending clauses and scoring risk.
It can also automatically complete letters of credit “in a fraction of the time”, Aurionpro said.
These AI agents are overseen by bankers at different stages so the system can learn from their human decisions and gradually expand their decision-making autonomy over time.
Fintra also integrates with messaging network Swift and banks’ general ledger systems, Aurionpro said, and includes an automatic limits management function that checks whether transactions breach the credit limits a bank has set on how much exposure it’s willing to take on any given deal.
Live pilots are currently underway with banks in India, the Middle East and Southeast Asia, according to the fintech. The pilots are “available for use in real operating environments, not just demos”, it told GTR.
To comply with regulatory governance, Fintra also has an embedded system called CGHP (confidence-gated handoff protocol) that evaluates every AI decision on four criteria: how confident the agent is, transaction value, whether a decision requires a human sign-off by law, and whether the system has seen the pattern before in similar scenarios.
The decision routes to a banker if any of those thresholds are not met, and the AI’s autonomy is earned gradually as it improves its so-called ‘agreement rates’.
“A bank cannot deploy an AI agent that auto approves a letter of credit without answering one question from the regulator: how do you know it is right?” said Aurionpro’s group CEO, Ashish Rai.
“Every agent decision is logged alongside what the human actually decided. The system tracks agreement rates. When the data shows 95% agreement sustained over months on a specific decision type, the bank can enable auto approval for that type.”
Rai added the system “is not configured into autonomy” but “earns it”, and creates a full audit trail for regulators to check.
Global fintech Aurionpro, which provides solutions for the banking and payments industry, acquired Hyderabad-based Fintra Software in April 2025 to bolster its trade finance offering, following several years of partnership between the two companies.
Under the new ownership, Fintra’s main software was rebuilt with AI as the core component, rather than adding it as a supplementary feature.
The platform “reflects Aurionpro’s conviction that the next generation of banking software will not be built by adding AI features to legacy platforms, but by rebuilding the core workflows of the bank on AI-native foundations”, Rai said.
Challenging market
Some of the different AI capabilities Fintra proposes “have been in the market for a number of years now”, noted Michael Vrontamitis, founding partner of advisory and consulting firm T3i Partner Network.
But integrating them all into a single platform could be a significant step forward and a potentially disruptive entry into a very competitive market, he said.
For instance, Fintra will be going up against incumbents such as Traydstream, Cleareye and Conpend, which offer document checking and compliance tools, while players like MyWave deploy AI agents that automate trade finance processes, including letter of credit generation.
Vrontamitis noted that entering the “very sticky” digital trade finance market is a challenge for any newcomer because “banks have invested in their systems, and they have long-term contracts that you can’t just change overnight, making change slow”.
Another key challenge with adopting automated processes is “how explainable it is to the regulator”, Vrontamitis pointed out, which Fintra’s CGHP attempts to address by embedding human governance and maintaining a fully auditable trail.
But regulation is evolving more slowly than the technology, Vrontamitis noted, and the impact of AI agents on roles within trade finance is likely to remain under scrutiny.
How Fintra’s AI agents work
Fintra has six autonomous workers that execute trade finance processes end-to-end with human oversight at defined “checkpoints”.
- Document extraction: the agent reads a purchase order and fills in all the required letter of credit fields automatically in about 45 seconds instead of 12 minutes, according to Aurionpro. The resulting application form is then reviewed by the bank worker.
- Document examination: when a seller submits paperwork to get paid, the agent checks every document against the contract terms, cross-checks figures across all invoices, bills of lading and letters of credit, and validates against international trade rules (UCP 600). “In our testing, the agent identified that ‘Nhava Sheva’ and ‘Mumbai’ refer to the same port complex,” Rai said. “An examiner handling 30 presentations a week across multiple geographies cannot hold every port alias in their head – the agent can.”
- Sanctions screening: the AI worker checks every party, vessel and cargo description against OFAC, EU and UN global sanctions lists instantly “before the bank officer opens the transaction”. Decisions are logged for a 10-year period, offering the “consistency and auditability that regulators demand”, Aurionpro said.
- Clause recommendation: Fintra’s agent builds the document checklist for each trade deal automatically, drawing on a database of over 30 standard clauses, and knowing that, for instance, a chemical shipment from Singapore to Mumbai needs specific insurance clauses and inspection certificates. The banker then reviews the proposed clause package.
- Risk scoring: before a banker reviews a deal, the system runs a full risk breakdown: counterparty exposure, country risk, facility usage and historical discrepancy rates and early “warning signals” from the applicant’s financial data, offering a factor-by-factor explanation, rather than a single final score.
- The orchestrator: the final agent governs all other five and decides whether the AI can act on its own, or whether a human must intervene.
