Conversational artificial intelligence (AI) started out handling simple tasks. For much of its decades-long commercial life, from Siri to Alexa and beyond, consumers used voice AI tools to check an order, reset a password, find a product or ask about the weather.
Historically, the experience hasn’t been great. After all, you likely aren’t having a smart home device read this article to you.
But two recent moves by Anthropic and OpenAI show voice AI shifting from answering questions to taking action. Anthropic’s new voice mode lets Claude users complete tasks in connected apps such as Gmail and Google Calendar by talking instead of typing. OpenAI, meanwhile, has built Presence, a platform enterprises are already using to deploy voice and chat agents for customer support and sales. Together, the launches point to the same opportunity: a natural-language interface capable of executing work itself, not just retrieving information about it.
Instead of just answering questions, these systems interpret business intent, pull information from multiple applications and kick off the workflows needed to complete a task.
That could make conversational AI more important in business-to-business commerce than it ever became in consumer technology. But it also raises the stakes. A chatbot that recommends the wrong pair of shoes is inconvenient. An AI system that approves the wrong supplier, misinterprets payment terms or mishandles an invoice dispute can create financial and operational risk.
See also: The 7 AI Terms Every CFO Needs to Understand
OpenAI and Anthropic Are Building Voice AI for Action, Not Conversation
For years, voice assistants competed on whether they could understand what people said. They’ve mostly solved that problem. The new competition centers on whether AI can do something useful after it understands its instructions.
Many business processes already function like conversations, even when they occur through email threads, spreadsheets and enterprise platforms. A warehouse supervisor can ask about inventory shortages while walking the floor. A field technician can diagnose equipment without stopping to type. A procurement manager can negotiate, retrieve supplier history and update an ERP system during the same conversation.
For B2B firms, the implication is workflow compression: voice AI agents collapse multiple corporate interactions into a single conversation. Consider accounts payable. Instead of navigating multiple ERP screens, a finance manager could simply ask: “Which invoices are becoming high risk? Contact suppliers requesting payment extensions under $50,000 and prepare a liquidity summary for tomorrow’s CFO meeting.”
In contract management, an employee might ask an AI system to summarize payment terms, identify unusual clauses or explain the risks in an agreement. A more capable system could then compare the contract with company policy, route an exception for approval and update the appropriate procurement record.
More like this: Agentic B2B Is Here. Are Your Contracts and Invoices Ready?
The interface becomes conversational while the underlying execution becomes increasingly autonomous. That explains why nearly every major AI vendor is racing to build connectors into CRMs, ERPs, document repositories, email, messaging platforms and business applications.
“AI has to have visibility into all of it,” John Landy, chief technology officer at Billtrust, told PYMNTS in an interview this week. “It cannot be added after the fact.”
“Every business unit owner, including finance leaders, should move toward becoming a context architect for their solutions,” Landy added, “working with human and digital coworkers to deliver meaningful results.”
The PYMNTS Intelligence report “Smart Spending: How AI Is Transforming Financial Decision Making“ found that more than 8 in 10 CFOs at large companies already use AI or are considering it.
You may like: Black-Box AI Forces CFOs to Write a New Audit Playbook
Enterprise AI Still Has an Integration Problem
Still, that frictionless vision collides with a messier reality: B2B operations are rarely as clean as a conversation, and voice-enacted, autonomous workflows run straight into it.
Research from PYMNTS Intelligence has shown that 83% of companies have yet to fully automate their accounts receivable operations, with data fragmentation a key culprit.
B2B workflows depend on fragmented data, established approval rules, contractual obligations and software systems that rarely work together. An AI agent may understand that a buyer wants to change a supplier, but completing the request could require data from procurement, finance, compliance, inventory and risk-management systems. After all, a traditional chatbot succeeds largely on whether its answer is useful. An operational agent has to clear a higher bar. It must use the correct data, follow the right policy, obtain the required approval and leave an auditable record.
“We see challenges around legacy ERP systems with limited AR API capabilities,” Michael Younkie, vice president of product management at Billtrust, told PYMNTS in January.
Research from PYMNTS Intelligence’s “The Enterprise AI Benchmark Report” shows that 71% of executives at companies with at least $1 billion in annual revenue believe organizational readiness, not AI technology, is the primary barrier to AI performance. Only 11% pointed to the technology itself. For voice AI to fulfill its promise as the interface to the B2B back office, in other words, the constraint isn’t the AI. It’s everything the AI still has to plug into.
