Ask most finance software what its AI does and the answer is a chatbot on top of a dashboard. Pac O’Shea, co-founder and chief executive of Round, a finance operations platform for UK companies below the size at which a treasury function exists, thinks that misses the point entirely.
In a written interview with The Fintech Times, O’Shea sets out where AI genuinely earns its place in finance teams, where the hype still lives, and why the line that matters is not what AI can say but what it may safely do.
For O’Shea, the test is simple. “AI earns its place where there is a high volume of repetitive work, lots of structured financial data, and a clear outcome you can measure,” he says. Extracting data from invoices, matching payments, reconciling accounts, forecasting cash, preparing payment runs and monitoring treasury positions all qualify. “The value isn’t that AI can produce a clever answer. It’s that it can actually complete the work.”
The hype, he argues, sits around “AI simply putting a chatbot on top of financial data and calling that automation”. Giving a CFO a box to ask questions of their ERP is useful, he concedes, but it does not fundamentally change how the finance team operates. “The real opportunity is moving from AI that tells you what happened to AI that can safely take action.”
That distinction between reporting and doing runs through Round’s product, which carries out day-to-day finance workflows rather than just surfacing dashboards. “A dashboard tells you that something needs to happen, an agent can actually do it,” O’Shea says. In accounts payable, instead of showing a finance manager a list of invoices, Round can ingest the invoice, extract the relevant information, check it against the business’s rules, route it for approval and ultimately execute the payment. In treasury, the platform can continuously monitor balances and upcoming obligations and move cash according to rules the finance team has defined, rather than waiting for someone to log in and do it manually.
“The line is around authority,” he adds. “We don’t think an AI should have unlimited discretion over a company’s money. The finance team defines the rules, limits and approval requirements, and the AI operates within that framework.”
Autonomy, on this account, does not mean removing controls. “Every action needs to be attributable, explainable and constrained by policy. You want to know what the AI did, why it did it, what information it used and which rule or approval allowed it to happen.” Higher-risk actions can carry approval thresholds or a human sign-off, while lower-risk, repeatable actions can run automatically. “We’re not trying to replace the finance director’s judgement. We’re trying to remove the mechanical work around that judgement.”
It is also why Round is built as a workflow layer rather than a general-purpose AI assistant. “Financial workflows have explicit rules, permissions and consequences, so the system needs to reflect those constraints,” O’Shea says.
Regulation shapes the architecture too. Round operates under the UK regulatory perimeter as an appointed representative of WealthKernel, which is authorised and regulated by the Financial Conduct Authority, and O’Shea treats that as a reason to build differently rather than an obstacle. “There are clear requirements around safeguarding, client assets, permissions, controls and oversight. The principle is that AI can automate an activity, but it doesn’t remove the underlying regulatory obligation or accountability.” Round does not hold client funds on its own balance sheet and is not in the flow of funds, he says. Client cash is
held with established, regulated Tier 1 custodians. “The aim isn’t to use AI to bypass financial controls. It’s to make those controls programmable and then automate everything that can safely sit within them.”
The evidence O’Shea offers is the company’s own. Round says it is now processing over $900 million through the platform, and that customers are reducing per-invoice processing time by around 75 per cent, with payment runs that previously required significant manual effort automated within days rather than weeks. The bigger metric, he suggests, is what happens to the finance team’s operating model. “Instead of someone spending the morning checking balances, chasing approvals, downloading invoices and preparing payments, those processes can run continuously in the background. The finance team gets to spend more of its time on cash strategy, forecasting and the decisions that actually require judgement.”
Asked what a finance director should ask before trusting any AI tool with live financial workflows, O’ Shea lists five questions. What exactly can the AI do, beyond the “AI-powered” label? What are the controls, and can limits, approval thresholds, permissions and rules be set around those actions? Can I see what happened, with a clear audit trail for every financial action? What happens when the system is uncertain? “A good system should know when to stop and ask for a human rather than confidently doing the wrong thing.” And who is ultimately accountable, since “AI doesn’t change the underlying responsibility of the finance
team or regulated provider”.
The most important question, he suggests, is the bluntest. “What happens if the AI gets it wrong? If the answer isn’t very clear, you shouldn’t give it access to live financial workflows.”
Round has already expanded from treasury into accounts payable and payroll, and the ambition now is what O’Shea calls the operating layer for the finance function, with cash position, upcoming payments, payroll, invoices, FX exposure and treasury strategy informing one another rather than existing as separate workflows.
“Ultimately, I think treasury becomes much more dynamic,” he says. “Instead of a finance team periodically looking at a dashboard and deciding what to do with its cash, you have an intelligent system continuously managing liquidity within parameters set by the finance team. The human moves further up the stack, from executing transactions to setting strategy, policies and risk appetite, while the software handles more of the execution.”
The post Round Treasury: Where AI Actually Earns its Place in Finance appeared first on The Fintech Times.
