Enterprise & Work

Coupa survey: 72% of leaders would let AI agents handle transactions

A Coupa poll of 606 executives finds 72% would let AI agents run transactions while under 1% report agents doing it today; data quality blocks 29%.

Business leaders back AI agents for transactions, survey
Business leaders back AI agents for transactions, surveyRaten-Kauf / pixabay
By Marcus Bennett5 min read

Why it matters

  • 72% of business leaders would let AI agents create or update transactions, while fewer than 1% say agents do that work today.
  • Coupa's poll covered 606 procurement, finance, supply chain and IT executives across North America, EMEA and APJ, published 18 November 2026.
  • Data quality or systems integration was the top obstacle for 29% of respondents, against 17% citing agent accuracy or reliability.
  • 94% want a complete audit trail and 90% want defined approval thresholds before granting agents greater autonomy.
  • Procurement teams piloting or using agents are expected to roughly double, from 28% to 59% in contract analysis and from 26% to 55% in spend and category strategy.

72% of business leaders say they would let AI agents create or update business transactions. Fewer than 1% say that work happens today.

Coupa published the research on 18 November 2026, drawing on a poll of 606 procurement, finance, supply chain and IT executives across North America, EMEA and APJ. The distance between stated willingness and actual deployment is the report's central result, and it defines the commercial stakes for enterprise agentic AI — software agents that execute tasks inside a workflow rather than only recommending actions to a person.

The sponsor's own Chief Customer Officer put a number on today's usage. John Frank, Chief Customer Officer at Coupa, said: "AI agents handle about 4% of the work today, and leaders expect that to triple within a year."

That tripling is a forecast of intent, not a measurement of deployed agents.

How wide is the gap between intent and deployment?

The survey separates ambition from operations:

  • 72% of business leaders would let AI agents create or update transactions; fewer than 1% report agents doing that work today.
  • 53% would allow agentic AI to handle transactional work only with safeguards in place.
  • 94% want a complete audit trail; 90% want defined approval thresholds before agents receive greater autonomy.
  • Frank puts current agent coverage at about 4% of the work, with expectations of a tripling within a year.

Respondents did not describe the barrier as reluctance toward the technology itself. Many identified underlying operational issues, rather than the technology itself, as the main barrier to wider use.

What is actually blocking deployment?

Data quality or systems integration was the single biggest obstacle for 29% of respondents. Agent accuracy or reliability drew 17%.

The ordering carries consequences for vendors selling into enterprise accounts. The friction sits in customer data and legacy integration, not in the accuracy of the models themselves. And 95% said access to trusted, reliable data would make them more willing to let agents act.

Even the reluctant group softens under stated conditions. Among leaders initially reluctant to let AI agents create or update transactions, 47% said they would allow it under clear safeguards.

What controls do executives demand before granting autonomy?

Governance, not capability, governs adoption decisions in these results. Some 94% of executives said they want a complete audit trail. And 90% said defined approval thresholds are necessary before agents can be given greater autonomy.

Both figures sit above the 72% willingness number, which makes governance a precondition rather than an optional extra. An agent without an audit trail or a threshold structure fails the test even where a buyer has already conceded the principle of delegation.

Confidence tracks with the same pattern. For tasks with meaningful business consequences, 45% of respondents said they were very confident in platform-embedded agents — agents running inside a system that already holds the data and the rules — compared with 12% for general-purpose AI tools. The 45-to-12 gap favours software embedded where spending decisions already live, the category Coupa occupies.

Early adopters are further ahead

Executives at organisations with agents already in production show stronger conviction than their peers. Among them, 93% expect agents to become standard within three years, compared with 68% of those not yet using them.

Budget allocation follows the same line. Early adopters allocate about 20% of their AI budget to agents, versus 7% among organisations not yet using them.

Procurement wants agents, and lacks the process discipline to run them

Procurement leaders registered the most openness to a larger role for agents. The survey found 86% expect agents to make their role more strategic, compared with 70% across other functions.

Practical limits are equally visible in that function. Among procurement respondents, 35% named data quality or integration as the biggest obstacle to adoption. Just 12% said their policies, approval thresholds and exception-handling processes are fully documented and consistently followed.

Procurement therefore combines the highest appetite with the weakest documentation, a 16-point strategic-role premium over other functions sitting beside a policy readiness rate of 12%. The function most willing to hand over routine decisions is the one least equipped to define what decisions an agent may take alone.

Which procurement workflows are expected to double

The projected growth concentrates in specific procurement processes:

  • Contract analysis: from 28% of teams piloting or using agents today to an expected 59%.
  • Spend and category strategy: from 26% to 55%.
  • Supplier support and workflow orchestration: from 27% to 53%.
  • Intake, guided buying and requisitions: from 26% to 51%.

Frank said current usage remains limited despite strong interest from management teams, and tied the pace of adoption to data and rules rather than to model performance.

"We believe how fast that happens comes down to trust in the data and the rules an agent follows. Our agents work from trusted community data, with permissions and audit trails built in to support the oversight leaders say they need. As a result, we are starting to see broad adoption of our purpose-built AI agents across our community," Frank said.

He also located demand inside systems customers already run. "Our customers want AI agents working inside the platform they already run, where their data, rules and approvals live. This report shows that's where leaders trust agents most," he said.

On the commercial path forward: "We're committed to every customer's success with agents. That starts with one process workflow, measured results and room to expand together," Frank said.

What to watch

The survey captures intent and expectation from 606 executives, and the under-1% figure for agents creating or updating transactions is the harder number to argue with. If the projected doubling in contract analysis, from 28% to 59%, lands, and if the 4% of work handled by agents triples within a year as Frank expects, the binding constraint stays where the survey placed it: 35% of procurement teams naming data quality or integration, and only 12% with fully documented approval policies. Vendors arriving with embedded data, audit trails and threshold-based approvals will capture that demand; those arriving with standalone models will compete for the 17% who cite accuracy as their concern.

Source: itbrief.com.au

Share this article:

More from Marcus Bennett

Marcus Bennett

Show full bio

Senior reporter covering companies, deals and strategy in artificial intelligence and machine learning at AI In Context.

718 articles

Related articles

  1. Only 10% of Enterprises Let AI Agents Decide Without Human Approval
  2. 51% of AI projects miss targets, yet 85% of firms raise budgets
  3. Australia Leads AI Agent Adoption at 46%, but Managers Pay the Price
  4. AI Agents Become Coworkers—and Managers of the Bots
  5. Behind the AI Agent Boom, Humans Are Still Making the Phone Calls

« Previous articleNext article »