Will digital ERP agents reduce administration or create more exception work?

It’s a tough question and the answer depends on each business case. There is every indication that digital ERP agents can take the repetitive work out of finance and HR. That is certainly the idea behind Unit4’s Ava (Advanced Virtual Agent), a digital agent designed to help automate routine enterprise tasks.

The proposition is appealing. 

After all, nobody wants to spend an entire morning chasing an expense claim, rekeying payroll data or answering the same policy question for the umpteenth time.

Unfortunately, automation, especially when not well planned, has a habit of moving work elsewhere instead of eliminating it.

When digital ERP agents like Ava deal with standard cases, someone still has to deal with the awkward cases. For example, an incomplete invoice, the worker with an unusual contract, the expense that falls outside policy, the payroll adjustment that does not fit the template, and so on. The question is not whether digital agents will reduce administration. They probably will. But, the question is who inherits the exception work left behind.

The exception is the work

Routine transactions rarely tell the whole story in people-centric organizations. People in HR and finance work with policies, but they also work with judgment, context and the sort of one-off circumstances that do not slot neatly into a workflow.

An AI agent may flag a missing purchase order or identify an expense-policy breach. That is useful. But, if a finance manager authorises an exception because a supplier relationship, emergency purchase or contractual commitment warrants it, accountability does not disappear. It moves.

The danger is that organisations measure success by the number of tasks automated, as opposed to the quality and volume of exceptions created.

This is particularly relevant to Unit4’s public-sector, higher-education and professional-services customers, where processes often combine standard rules with complex people and funding arrangements. Unit4 has said Ava forms part of its wider push to meet customer demand for AI-enabled innovation.

Automation needs an owner

Every automated process needs a clear answer to four questions:

• Who reviews an exception?
• Who corrects inaccurate data or an inappropriate recommendation?
• Who approves the final action when policy and context conflict?
• Who can reconstruct what the agent did and why?

These are operational questions, not theoretical AI-governance concerns.

As we argued in our examination of why agentic AI needs process intelligence, AI works best when it has a structured view of how work really moves between teams. That includes the delays, hand-offs and workarounds hidden behind an apparently simple process.

The real measure of value

Ava could reduce the grind that bogs down finance and HR teams. 

That matters. 

But organisations should monitor whether automation shrinks the exceptions queue or simply makes the remaining cases harder and more time-consuming to resolve.

Human work will not vanish. It will become more concentrated around judgment, correction and accountability.

That is not a failure of digital agents. It is the job they leave behind.