Trust over tech: The guarantees consumers need to make machine customers happen

From web traffic analysis to the first-hand experiences of practitioners, there are many stats that indicate the machine customer future has arrived. 

CX Network's 2026 research into the state of CX found AI-first customer journeys to be the third most influential trend shaping the work of practitioners to 2030 while consumers using AI emerged third most important customer behavior. When it analyzed web traffic to US retail sites and surveyed 5,000 US consumers, Adobe found a 693 percent year-on-year increase in traffic from AI sources during the 2025 Golden Quarter. 

But while conversations around consumer AI use focus on payment rails, experience and journey design or AEO, there's another element of readiness that CX needs to consider: whether consumers are psychologically and institutionally ready to let an agent act and spend on their behalf.

On 13 October, Michael Nevski, who most recently held the role of director of global insights for Visa's Business and Economic Insights team, will cover this point in a live discussion on the first day of Mastering CX: AI Agents 2026. 

Ahead of the session, he tells CX Network how brands can connect with customers using an AI intermediary, why consumers need more guarantees around trust not more technology and why trust in AI assistants will be earned incrementally and not granted. Via this link, you can register to watch the session live. 

Don't miss any news, updates or insider tips from CX Network by getting them delivered to your inbox. Sign up to our newsletter and join our community of experts. 

Learn More

CX Network: Your session is called The rise of agentic AI: Trust, control, and the North American consumer. What will it cover?

Michael Nevski: The premise is that the industry is asking the wrong question. Most of the conversation right now is about what AI agents will technically be capable of doing. 

The more consequential question is whether consumers are psychologically and institutionally ready to let an agent act, and spend, on their behalf.

I'll walk through longitudinal consumer research tracking how attitudes moved between last September and this April, across the progression from traditional AI to generative AI to agentic commerce. 

The through-line is a gap: the market is moving considerably faster than consumer understanding, and that gap is where adoption either accelerates or stalls.

The practical argument is that convenience does not automatically produce consent. 

Consumers already grasp the appeal, efficiency, personalization, less effort, potential savings. What they are waiting for is not better technology. They are waiting for a trustworthy system around the technology. So the session covers what the data says about readiness, where delegation actually begins, and what organizations have to build to earn the right to act on someone's behalf.

CX Network: What are some of the most surprising customer insights your work has uncovered so far this year?

Michael Nevski: Three insights genuinely reframed how I think about this.

The first is that a measurable group of consumers claim to be using agentic AI already, and claimed usage roughly doubled from about three percent to six percent over the study period, even though fully realized agentic commerce experiences barely exist yet. 

My read is that this is not measurement noise. Consumers are encountering increasingly autonomous AI without the vocabulary to distinguish generative from agentic. Which means adoption will not follow the clean technical progression the industry has drawn. People may behave as agentic users before they can define what agentic means.

The second is the split-brain mindset. 

Roughly 77 percent of consumers recognized clear benefits, and 83 percent simultaneously reported concerns. Those are the same people. This is not a market divided into enthusiasts and skeptics; it is a market of individuals holding both positions at once. That matters enormously, because it means the barrier is not persuasion about value. 

Consumers already believe the value. They are uncertain whether the institutions delivering it can be trusted.

The third is the most commercially instructive. In related research, consumers were comfortable letting an agent spend roughly US$10-20 on their behalf. I called it fun tickets in the session, an amount small enough to feel experimental rather than consequential. That number sounds like a failure. I think it is actually the roadmap. 

Trust here will not be granted, it will be earned incrementally, the way it is in a human relationship. Limited responsibility first, greater autonomy after repeated reliable performance. 

CX Network: From the growing adoption of AI assistants to economic pressures and demand for data privacy, a lot has happened in the consumer space in recent months. What do brands need to do to effectively connect and engage with their customers?  

Michael Nevski: Start by getting concrete, because abstraction kills interest. When agentic AI was presented as a general concept, sentiment was tentative and most people were unsure. 

When I showed specific use cases, the picture changed completely. 

Consumers were readily willing to delegate staple groceries, reordering household supplies, smart-home management, restaurant reservations, routine maintenance. 
The pattern is that people happily hand over what they already experience as chores. They are far more hesitant on banking, account-to-account payments and healthcare, where accuracy, recourse and accountability carry real weight.

So agentic commerce will not arrive as one universal behavioral shift. It will proceed task by task, starting with frequent, low-risk, reversible activities. Brands that try to lead with the most consequential use case will lose to those who earn trust on the mundane one.

On data, the finding I would most want people to absorb is that privacy is conditional, not binary. 

Asked generally whether they would share personal information with an agent, consumers were reluctant. Asked whether they would share location so the agent could find and book a nearby restaurant in a defined window, willingness rose sharply. 

They are not opposed to sharing data. They are opposed to sharing it without a clear purpose, a defined benefit, a limited scope and a reasonable duration. That argues for requesting data contextually and incrementally rather than asking people to build an unrestricted profile up front.

Economic value accelerates everything. General interest in using AI for an upcoming purchase was modest. Attach a guaranteed 10 percent saving and stated interest nearly doubled. Convenience is appealing; savings make the proposition measurable. But money alone will not solve trust, because consumers still need to believe the agent will operate within limits and that mistakes can be corrected.

And one uncomfortable finding for brands specifically: consumers showed greater interest in using a specialized, independent agent provider than an agent operated by a merchant. 

They understand that merchants are in the business of selling, and they question who a merchant-controlled agent is really working for. The message was essentially, merchant, stay in your lane. That should give every brand building its own agent something serious to think about, because the strategic question of the decade may be who consumers believe the agent actually represents. 

CX Network: For the purpose of proving the value of CX and insights to the wider organization, what are some of the non-negotiable tools (or techniques) practitioners should embrace for 2027 and beyond?

Michael Nevski: There are four, and only one is related to technology.

Connect insight to commercial outcomes, always. The most rigorous study is worthless if no one can trace a decision to it.

My test is simple: did this change where the business put its money, its people or its roadmap? Segmentation work I led at Schwab mattered because it moved roughly 30 percent of media spend and lifted conversion 32 percent, not because the clusters were elegant. Learn to state impact in the finance function's language, because that is the language budgets are set in.

Build integration, not another dashboard. Insight, measurement and analytics have to resolve into one view leaders trust. When leadership spends the first 15 minutes of a meeting adjudicating whose number is right, the function has failed regardless of the quality of any individual study.

Adopt AI across the research workflow, with rigor. I use AI for survey generation and validation, open-end coding, synthesizing prior research so we stop re-fielding questions we have already answered, and drafting executive summaries.

The productivity gain is real and large. The discipline is non-negotiable: validate against source, not plausibility. These models will confidently hand you a wrong answer. I once caught one attributing sales to the wrong company because two merchants shared an abbreviation and the model resolved the ambiguity toward what looked right. 

There is a certain irony in researching consumer trust in AI while maintaining strict guardrails on my own use of it, but that is precisely the point. 

Speed without rigor destroys the credibility you spent years building.

Keep listening, continuously. This is the one most people underestimate. Consumer expectations in this space are changing as fast as the technology. Attitudes toward privacy, control, spending authority and acceptable autonomy will look different in 12 months. A single study is a snapshot; what organizations need is a longitudinal read. 

Mine only produced its most interesting findings because I could compare September to April.

Join Mastering CX: AI Agents on October 13-14
to hear Michael cover all these insights and put your questions to him live. 

Quick links

• When your customer is a machine: Rethinking service design for AI agents
• The new rules of discoverability
• Agentic agility: The new 'normal' for CX in 2026