Why Customers Aren't Letting AI Agents Take The Wheel On Payments Yet

Rick Watkin is President & CEO of KUBRA, a leading provider of customer experience management solutions for utilities and other industries.

gettyCustomers are saying “yes” to a lot of what AI has to offer these days, whether that’s an AI-generated quote, a snippet to help make a purchase decision or a pre-filled form. But those turn into “nos” the moment AI renews a policy or pays a bill on their behalf.

When autonomous AI agents get involved in money matters, comfort and trust fall off a cliff the moment the system stops suggesting and starts acting. Today, only 23% of U.S. consumers trust AI to handle a payment transaction on their behalf.

In essential services, that cliff is steeper still. Six in 10 utility customers primarily pay through their utility’s app or website, yet only 27% strongly trust their utility to protect their personal data, and only 28% strongly trust the accuracy of their bill. If they don’t fully trust the provider with their data and their bill, giving an AI agent the authority to act on that bill is a much bigger leap.

Trust is a threshold. Right now, customers trust generative AI to assist long before they trust AI agents to act. Delegation is trust measured by how much control you hand over. Each step up the ladder, from awareness to permission to reliance to full autonomy, requires a higher threshold than the one before.

There’s also a structural mismatch. AI is built to produce the most likely or most helpful answer, not a guaranteed one. For financial execution, customers want an exact, deterministic outcome every time. That gap is a design problem to solve.

Stakes matter too. If an AI agent overpays for a pair of shoes, it’s an inconvenience. An underpaid insurance premium or an overpaid utility bill can disrupt a household. That’s why essential services face a higher trust bar than retail.

In government, misplaced trust costs privacy and control, which is why the bar is high. Pew found that 76% of Americans say it’s important to be able to tell whether content was made by AI or by people. Knowing the source matters, and Canadian data points to the same. Forty-seven percent of Canadian respondents would trust a custom government AI chat over Google or Bing when grounded in official government content.

Trust in insurance is built on two things: a path to a human and a clear explanation of how AI is used. In a GlobalData survey, 49% said nothing would improve their confidence in an AI-determined quote, yet 28% said the option of a human referral would. A Swiss Re global survey found that people who felt they understood how insurers use AI were more likely to trust them with their data.

​Utilities are working to earn back customer confidence. A 2025 study by J.D. Power put residential electric satisfaction at 499 out of 1,000, the lowest across its residential utility studies, and EY reports that 49% of customers believe their utilities make bills confusing on purpose. That belief is also an opening; AI earns trust when it brings clarity, but loses it when it moves from informing to acting. A recommendation can be checked before a customer acts on it, while an action that changes what a customer owes needs a clear way to review and reverse it.

Across all three industries, trust comes from what surrounds the AI, not the AI alone. While only 23% trust AI with a payment, 61% said they would trust Visa to handle agentic transactions, showing how much a familiar name matters. This suggests consumers are looking for a brand they already trust to stand behind AI agents.

The industry is already acting with new initiatives like “Know Your Agent” (KYA), offering standards that verify an agent’s identity, the person it acts for and the limits of its authority. For organizations still rebuilding customer confidence, the faster route is to borrow it by running AI payments on rails customers already know.

First, run on trusted rails: Route agent-initiated payments through payment networks and brands customers already recognize, not new, unproven pathways. Next, keep a visible human checkpoint by requiring approval before a charge posts or a bill amount changes, because in payments, an unreviewed action costs real money from a real account.

Make consent easy to grant and revoke, so customers can expand or withdraw an agent’s authority in one step, at any time. Finally, explain the specific account, and keep the record. If a customer asks why this month’s bill is higher, the agent should give a one-sentence answer based on that customer’s own account, not a generic explanation. In regulated industries, that same answer must hold up afterward, so a support rep or regulator can reconstruct why the agent acted.

When it comes to AI-powered payments, customers are waiting for proof that someone trustworthy is accountable and that they stay in control. For utility, government and insurance, the task is clear: Give customers a familiar brand, a human checkpoint, easy consent and answers specific to their account.

The promise of AI in essential services is bills people no longer have to think about. Customers will only let AI take them there if they stay in control along the way. The organizations that build those guardrails first will be the ones customers trust to act on their behalf.

Forbes Technology Council is an invitation-only community for world-class CIOs, CTOs and technology executives. Do I qualify?