Voice AI enters the back office as OpenAI and Anthropic add agents

Voice AI back office systems are moving from basic question-answering toward tools that can carry out business tasks, according to recent product moves by Anthropic and OpenAI. Anthropic has introduced a voice mode for Claude that lets users work with connected apps such as Gmail and Google Calendar by speaking, while OpenAI has built Presence, a platform PYMNTS said enterprises are using for voice and chat agents in customer support and sales.

The shift matters for companies because back-office work often depends on information spread across email, enterprise software, spreadsheets and internal approval processes. A voice interface that can interpret a request, gather data and start a workflow could reduce the number of screens and handoffs required in areas such as procurement, accounts payable and contract management.

Consumer voice assistants have long handled relatively narrow tasks, including checking orders, resetting passwords, finding products or answering basic questions. The new enterprise focus is broader: systems are being designed to take action after understanding an instruction, rather than only retrieving an answer.

How will voice AI work in the back office?

In a business setting, voice AI works as a natural-language layer over corporate systems. A user gives an instruction by speech, the AI interprets the intent, draws on connected applications and starts the next step, such as drafting a message, updating a record or routing an approval.

PYMNTS described several possible uses. A finance manager could ask which invoices are becoming higher risk, request supplier outreach on payment extensions below $50,000 and prepare a liquidity summary for a CFO meeting. In contract management, an employee could ask for a summary of payment terms, a review of unusual clauses or an explanation of risk, with a more advanced system comparing the contract against company policy and updating procurement records.

That model requires links into the systems where company data and decisions already reside. PYMNTS said major AI vendors are building connectors into customer relationship management platforms, enterprise resource planning systems, document repositories, email, messaging tools and other business applications.

John Landy, chief technology officer at Billtrust, told PYMNTS that AI needs visibility across these systems and “cannot be added after the fact.” He said finance leaders and other business-unit owners should become “context architects” for their systems, working with human and digital coworkers to deliver results.

Data gaps remain a barrier for enterprise adoption

The risks are higher in B2B operations than in many consumer uses. A poor retail recommendation may be inconvenient, while an AI agent that mishandles supplier approval, payment terms or an invoice dispute can create financial and operational exposure.

PYMNTS Intelligence reported that more than eight in 10 CFOs at large companies either use AI already or are considering it. Separate PYMNTS Intelligence research found that 83% of companies have not fully automated accounts receivable operations, with fragmented data cited as a major cause.

Legacy systems are another constraint. Michael Younkie, vice president of product management at Billtrust, told PYMNTS in January that companies face challenges from older ERP systems with limited accounts receivable API capabilities.

For an operational AI agent, the standard is higher than producing a useful answer. It must use correct data, follow company policy, obtain approvals where required and create a record that can be audited.

PYMNTS Intelligence’s Enterprise AI Benchmark Report found that 71% of executives at companies with at least $1 billion in annual revenue see organizational readiness, rather than AI technology itself, as the main barrier to AI performance. Only 11% identified the technology as the main obstacle, suggesting that the next stage of voice AI in the back office will depend heavily on how well companies connect and govern the systems behind it.

This story draws on original reporting from PYMNTS.