Unilever’s AI agents process orders in 90 seconds, while Nestlé uses them to prepare collections

Unilever says its AI system now processes orders in about 90 seconds across 11 pilot markets, down from 15 to 20 minutes. The deployment shows AI handling a defined chunk of routine business paperwork, while people review exceptions rather than handing over the whole process.

Watch Desk analysis

What happened

Unilever’s NOVA system uses four agents to read incoming messages and extract order details, followed by a fifth that checks their work, according to PYMNTS. The company says 95% of orders in the 11 markets pass through without manual handling. Missing information and exceptions are referred to employees, and the system is also being used for claims validation.

The report also describes Nestlé using SAP tools to help prioritise collections and recommend actions. SAP’s 26 October product update lists seven receivables agents for tasks including account analysis, preparing collections, correspondence and dispute handling, with phased general availability under way.

Why it matters

These examples put AI into the unglamorous paperwork between a sale and payment: interpreting orders, checking records and preparing collections work. That can save time, but the systems are not shown making every decision unaided. Unilever’s fifth-agent check and referral of exceptions to employees are part of the practical picture, not decorative footnotes.

The potential gains sit beside a stubborn obstacle. PYMNTS Intelligence research with JPMorgan found that 90% of surveyed CFOs cited integration complexity as a barrier to further accounts-receivable automation, up from 67% in the previous year’s survey. The research surveyed 100 US CFOs in July 2026 and compared their answers with a separate survey of 375 CFOs in 2025.

Our read

The most useful detail is not that companies have agents, but what they do and where people still step in. Unilever’s reported processing time is striking, although it comes from the company’s own account of pilot markets, not an independent comparison. Integration with existing systems and a sensible route for exceptions may matter more than the agent count.

What to watch

• Whether Unilever shares results beyond the 11 pilot markets, including accuracy and exception rates.
• How Nestlé and SAP evaluate the recommendations and actions from receivables agents.
• Whether companies can overcome the integration barriers CFOs identified.

Discussion spark: For routine finance work, should businesses let AI agents process most cases automatically and send exceptions to staff, or should a person check every case before it moves on?

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.