EY Principal: ‘Design for Zero’ to Optimize AI’s Benefits in Finance

Companies deploying AI across finance departments are reaching an inflection point after discovering that simply adding the technology to existing workflows often falls short of delivering the gains executives expected, according to Amanda Donohue, principal of finance consulting at EY.

After more than two years of experimentation, many finance departments are finding that automating isolated tasks speeds up work but rarely transforms how they operate.

“That’s the inflection point we’re at now,” Donohue said in an interview with The AI Innovator. “Companies are saying, ‘This AI thing came. I’ve automated some tasks. I’ve done some things. I’m not really getting either the benefit or the headcount reduction or whatever expected benefit they had.'”

The shift represents a broader change in enterprise AI strategy. Rather than viewing AI as another productivity tool, organizations are increasingly evaluating which parts of finance should be handled by AI agents and where human judgment remains indispensable.

Donohue argues many early AI initiatives amounted to little more than what she described as “RPA 2.0” — the next chapter of robotic process automation that accelerates a handful of steps without fundamentally redesigning the process itself.

Designing for zero

Instead, she advocates what she calls a “design for zero” methodology.

Rather than asking how AI can improve an existing multi-step process, companies should begin by defining the desired business outcome and designing the workflow as though no human intervention were required. Humans are then added back only where judgment, oversight or regulatory requirements demand it.

“Forget your process,” Donohue said. “What is the outcome you’re trying to achieve?”

That represents a significant departure from how finance systems have traditionally evolved.

Over decades, companies accumulated enterprise resource planning systems, procurement software, billing platforms and many other applications as business needs emerged. In this complex setup, data management often became an afterthought, she said, making it more difficult for AI systems to produce reliable results.

Because AI depends on accurate, well-governed data, organizations now have to adopt what she calls a “data-first” mindset before large-scale AI deployments can succeed.

FP&A: A common use case

One area where companies are seeing early returns is financial planning and analysis (FP&A).

Finance teams frequently spend weeks gathering data from multiple ERP systems, reconciling spreadsheets and investigating why financial results differed from forecasts. Donohue said AI can automate much of that historical analysis, allowing finance professionals to focus instead on interpreting the results and advising business leaders.

In one EY solution, six AI agents work simultaneously to analyze data and cross-check one another’s outputs, helping reduce the risk of hallucinations. Clean data and extensive testing remain essential before organizations can trust AI-generated analysis, she said.

Even so, Donohue cautioned that finance is unlikely to become fully autonomous.

Finance involves approvals, compliance obligations and spending decisions that require human judgment, making complete automation less practical.

“I don’t think companies will want to be in a place to say, ‘let’s just let the AI decide how much money is going to go out the door,'” she said.

Setting the baseline

The biggest obstacle, however, may not be the technology itself.

Donohue believes many companies are making a strategic mistake by jumping directly into AI pilots without first determining where automation would generate the greatest business value. Organizations need to establish baseline performance metrics, decide whether they prioritize lower costs, faster decision-making or improved forecasting, and then focus AI investments accordingly.

Some companies, she noted, have deliberately chosen to delay adoption, preferring to become “fast followers” while competitors absorb the costs and risks of early experimentation. Others conclude that AI simply does not yet provide sufficient return on investment, particularly if they have already moved finance operations to low-cost offshore locations.

The transition will also reshape the duties of finance professionals.

Rather than spending hours collecting and reconciling data, they will increasingly be expected to understand data quality, evaluate AI-generated insights and challenge model outputs. Universities and employers will need to prepare workers for those new responsibilities, she said.

Asked whether AI agents will eventually absorb the finance function entirely, Donohue said, “I think there’s still going to be a finance department” with a treasurer, chief accounting officer and other group leaders. “I think there will be people. Their skill sets will be different and I think (finance departments) will probably be smaller than what we see today.”