Ung says BlackLine is seeing time savings of more than 90 per cent in some work involving accruals, account reconciliations, account preparation and collections.
That can mean AI performing work around the clock while finance professionals spend less time on repetitive processes. Ung says the longer-term opportunity is not simply to add AI features to existing software but to build a “digital workforce” in which agents can carry out complex, long-running tasks alongside human teams.
But giving AI agents more work also raises the standard of governance required around them.
Research published by ACCA and Chartered Accountants Australia and New Zealand in July found 93 per cent of finance professionals were concerned about the integrity and verifiability of AI-generated insights.
“Issues include AI hallucinations, inaccuracies, incomplete data sets, lack of transparency and bias,” ACCA says in a release accompanying the report.
For CFOs, the question is becoming less about whether AI can perform financial work and more about what must be in place before they can trust it to do so.
From probability to proof
Generative AI is probabilistic, producing outputs based on patterns in the data it processes. For financial work, BlackLine argues that this needs to operate within deterministic controls: rules that define what a system can do and provide a consistent framework for how work is performed.
An answer that looks plausible is not enough when the result feeds into financial processes. Finance teams need controls around what a system is allowed to do and evidence that allows its work to be examined afterwards.
Ung describes that requirement as “defensible AI”.
Ung says defensible AI requires companies to understand not only the work an AI system performs but the instructions it received, what it was permitted to do and the model used to perform the task. Those details must be retained so the result can subsequently be interrogated.
Ung says conversations with regulators, auditors and major audit firms have reinforced the high standard of control and transparency expected around financial information.
“You need to know exactly what goes into it so you can be audited later,” he says.
He compares the principle with a familiar instruction given to children learning mathematics: show your working.
Getting the answer right is only part of the exercise. The steps used to reach it matter too.
For AI agents, that means creating an evidence trail as they work. A finance professional or auditor should be able to return to a result and establish what information was considered and how the system reached its conclusion.
The distinction is what Ung describes as the difference between a “glass box” and a black box.
“For finance and accounting, that is not an optional thing,” he says. “Being able to understand how the AI arrived at a result or a conclusion, what data it looked at, what evidence it used, how it constructed its reasoning is critical.”
That requirement becomes more important as AI moves beyond assisting employees and begins performing longer and more complex financial tasks. Controls that govern human work cannot simply disappear when the work is automated.
Ung says AI agents should operate within the safeguards expected of people performing the same activity, while providing additional traceability around the models, instructions and information involved.
Beyond the month-end crunch
Finance and accounting teams can spend long hours at month-end gathering information, reconciling accounts and working through manual processes before the books can be closed.
Agentic AI creates the prospect of shifting more of that work into continuous processing. Instead of concentrating activity around monthly or quarterly deadlines, financial information can be processed as transactions occur.
Ung says the longer-term goal is a continuous close, giving businesses access to financial information closer to real time.
For Australian companies dealing with exchange-rate movements, tariffs, international trade or intercompany transactions, waiting for the next reporting cycle can leave managers making decisions using information that is already weeks old.
Ung says substantial sums can be at stake when conditions move between reporting periods.
The opportunity is therefore broader than reducing the workload at month-end. Faster processing can change when financial information becomes available to the rest of the business and allow finance teams to spend less time monitoring manual processes and checklists.
But automation at that scale also raises a governance question. A system performing the same task repeatedly across large volumes of transactions can amplify the benefits of good decisions, but it can also amplify errors.
That makes the ability to trace an agent’s actions and understand the controls governing it fundamental to expanding its role.
For Ung, accountability is built into the profession itself. Accounting depends on processes and controls that allow businesses to stand behind their financial statements. An opaque system that produces a result without providing a way to establish how it got there runs against that principle.
He compares the challenge with commercial aviation, where complex systems operate at enormous scale because procedures, controls and lines of accountability are established before the work begins.
“You want to have those frameworks, you want to have those principles, you want to have that accountability, so that you can do this successfully every single time,” he says.
Building the hybrid finance team
The next phase of AI in finance may therefore be less about replacing people than changing the division of work between people and technology.
Ung sees AI agents becoming part of a digital workforce, taking on more repetitive and continuous financial processes while finance professionals retain oversight and concentrate on work requiring human judgment.
That model could also change the role of the finance function. If agents can perform more processing as transactions occur, finance professionals can spend less time assembling information and more time interpreting what it means for the business.
AI agents do not operate according to the working day and can repeat tasks across large volumes of transactions. That helps explain why CFOs have an incentive to explore a model in which automated systems take on more of the processing work.
The more work an organisation entrusts to agents, the greater the need to define what they can do, preserve evidence of what they have done and keep people accountable for the financial outcomes.
Ung says the potential of AI in finance is high, but the accountability required is just as high.
The result could be a finance function in which people and AI agents work alongside each other, with technology providing scale and continuous processing while people provide judgment, oversight and responsibility.
For Ung, that is the promise of a digital workforce: financial work performed faster and at greater scale without abandoning the control, auditability and trust on which the finance function depends.
Learn how CFOs are building trusted AI governance at https://www.blackline.com/agentic-financial-operations/