The number of large enterprises or government processes “that have been really, truly reinvented with AI at the absolute core is basically close to zero,” Marc Warner, Accenture’s global chief technology officer, said during a Fortune Brainstorm AI virtual event. Accenture is a founding partner of Brainstorm AI and sponsored the event.

Warner and three other executives spoke with Fortune AI editor Jeremy Kahn about why many companies report little return from AI despite wide use. The other panelists were Matthieu Caillat of AXA, Greg Ulrich of Mastercard, and Marjorie Janiewicz of Mistral AI. Their answers pointed less to software than to organizational change.

Edges versus core

Warner said most companies have put AI in peripheral work such as marketing, meeting notes, and transcription. Coding is “perhaps a little different,” because returns there can be measured. Changing the economics of an industry, he said, means targeting core business processes, which is risky. He compared the moment to early factories that used electricity only for lighting, before redesigning production around electric motors.

Ulrich said the biggest returns come from redesigning whole workflows, not layering AI onto existing ones. He said Mastercard is “very, very advanced” in redesigning its product development life cycle, with smaller teams and faster turnaround. He also described a trap: speed up one step in an 80-step process and the next step simply develops a backlog.

“The AI agent to automate a piece of this isn’t … the long pole in the tent for a lot of this,” Ulrich said. “It’s the rethinking. It’s the remapping.”

Warner called AI transformations “politics projects” and said he had thought of them as technology projects until recently. He said he still sees the AI builds themselves as technology projects.

Caillat said that as technology improves, the work is less about technology and more about, “How do you bring people along?” Ulrich said Mastercard relies on CEO sponsorship, including weekly updates with the CEO’s direct reports, but added that no one has solved the politics of large organizations.

Accountability

Kahn asked an audience question about who is responsible when an AI agent goes wrong. Warner said there is no single clean answer. In general, he said, the business owner is accountable for an agent’s value, the technologist for its functionality, and the chief information security officer for its security.

He then said most companies shouldn’t dwell on fault yet. “I don’t think most organizations should be worrying too much about the blame game right now,” Warner said, urging companies to give people credit for trying things, knowing many will fail. He made an exception for critical incidents such as a major security breach.

Janiewicz, who said she spent 10 years in cybersecurity, agreed that finger-pointing isn’t useful. She said the better question is whether a company controls the technology it adopts and can set its own guardrails. Mistral sells open-weight models that customers can run on their own infrastructure.

Keeping pace

Executives also said their plans have a short shelf life. Ulrich said consulting tools Mastercard built about a year ago are now “quaint, and just not that useful today,” so the company is rebuilding them. Caillat said AXA used to “measure twice and cut once” and “can’t afford that anymore,” because conclusions expire before certainty arrives.

Caillat said AXA started with about 400 AI use cases to build enthusiasm, then narrowed its focus. He said the insurer’s plan for 2027 to 2029 targets €500 million to €700 million a year in recurring pretax value from AI by 2029, and that AXA has about 4 billion unstructured documents it is starting to use.

Warner said that, by Accenture’s calculations, the pace shows no sign of slowing until something approaching 2030. Energy constraints in the West could start to bite in the early 2030s, he said.

Accenture is the founding partner of Fortune Brainstorm AI events and hosted this session. Generative AI was used for research assistance and/or transcription for this article. The reporter independently reported and verified the factual claims in this article, and a human editor reviewed it before publication.