Agentic AI is pushing finance systems beyond automation and into decision-making territory. As these systems begin acting across multiple workflows, the boundaries of trust, auditability, and human accountability become less clear.
In this interview, Jeremy Ung, Chief Technology Officer at BlackLine, explains why finance teams should resist full autonomy in favour of stronger human-led governance, how data foundations shape the reliability of agentic AI, and what CTOs and CFOs need to align on now to avoid hidden risk as AI-driven decision-making expands.
How ready are finance teams to trust agentic AI with complex workflows?
Building trust in agentic AI isn’t a static event, it’s an ongoing process. Right now, finance teams are moving from isolated, task-specific automation to more complex, multi-step workflows. However, complete autonomy is not the immediate goal. The real priority is building a framework of human-led governance for AI.
Industry data shows a natural reticence among finance leaders. Deloitte has noted that 21.3% are wary of full autonomy, which reflects responsible oversight rather than resistance. Leaders need to be able to explain how AI reaches its decisions, placing the emphasis on data accuracy, control, and complete auditability.
What safeguards are needed to prevent audit and compliance risks in autonomous finance systems?
Mitigating risk is the most critical challenge in finance and accounting, where the adage “95% accurate is 100% wrong” highlights how even small errors can result in significant financial and reputational damage. AI used in autonomous finance systems must be designed specifically for financial contexts and grounded in a clear understanding of financial processes, data, and controls. Every decision made by an AI agent must be fully auditable, which depends on data quality and ongoing human oversight.
Comprehensive policies that define standards for data quality, consistency, and reliability across the organisation are vital. The development of these policies needs to be under human control, with AI used to augment human expertise rather than replace it. For high-risk decisions in particular, an AI agent can support analysis and provide recommendations, but the ultimate decision rests with a finance professional. In this sense, autonomous AI changes how accountability is exercised but does not remove it.
To address hidden risks, safeguards are required across governance, transparency, data integration, and the use of recognised international certification standards. Together, these measures support auditability and credibility while ensuring robust human oversight remains in place.
How can agentic AI integrate finance more deeply with enterprise systems?
The conversation is fundamentally shifting from automation to intelligence. Agentic AI’s true potential lies in making finance the data-driven nerve system of the enterprise, breaking down silos with the rest of the organisation.
Integration with enterprise resource planning (ERP), risk, compliance, and treasury systems allows AI agents to support predictive forecasting, anomaly detection, and narrative reporting that aligns finance more closely with other business functions.
One example is procure to pay. An AI agent can monitor inventory levels within ERP systems to anticipate supply shortages, initiate sourcing activity through procurement tools, and manage the procurement process end to end. The system operates autonomously across these steps, escalating to a human only when it identifies exceptions such as unfamiliar vendors or unusual pricing.
By using large language models alongside structured reasoning and planning, AI agents can also address multi-step scenarios that require contextual judgement. This includes adapting workflows to changing market conditions and tailoring interactions based on situational inputs, moving finance closer to higher levels of process autonomy.
In order-to-cash scenarios, for example, an AI agent can track customer payment behaviour, assess credit risk, adjust payment terms to support cash flow, and manage collections based on real-time risk signals.
What data inputs matter most for agentic AI in forecasting and scenario planning?
It is more about the agent’s ability to pull together different data and ingest, contextualise, and reason with it. A platform is vital in connecting disparate data sources, from ERP systems to banking platforms, to provide the holistic, intelligent view needed for dynamic forecasts that AI can apply appropriately.
This allows agentic AI to incorporate market indicators and place forecasts within a broader economic context. Inputs such as interest rates, inflation indices, commodity prices, and geopolitical developments can be combined to generate scenario planning and stress tests that model both best- and worst-case outcomes. This helps finance teams assess potential impacts and prepare responses, supporting informed decision-making while maintaining compliance and control.
What technical decisions should CTOs and CFOs align on to avoid AI lock-in and capability gaps?
There are three key elements to prioritise: a unified data strategy, AI designed for financial use cases, and talent management.
- A unified data strategy: This goes beyond cleaning data. It involves joint investment in robust, integrated data structures that allow accurate, contextual data to move reliably across systems.
- An AI architecture designed for finance: CFOs and CTOs need to align on embedding AI directly into operational software rather than treating it as a separate layer or add-on.
- An investment in talent: While technology can improve efficiency, its effectiveness depends on humans. This requires shared ownership of building a finance function that can operate with higher levels of autonomy and adaptability, across both the tools in use and the teams responsible for them.
