A roundtable convened specifically because nobody has settled this yet
Mastercard brought 40 senior executives from across retail and payments into a closed-door session at its New York technology hub specifically to work through a question the industry has not resolved: how much agentic AI belongs in holiday shopping this year, and where exactly the line sits between helpful automation and automation that makes customers uneasy. The fact that this required a dedicated roundtable, rather than a settled playbook retailers could simply follow, says something about how unresolved the agentic commerce question still is heading into the highest-stakes shopping season of the year.
PYMNTS' own framing of the research underlying the discussion captures the tension precisely: AI may be ready to shop, but consumers are still deciding how much freedom they want to give it. That gap between technical readiness and consumer comfort is the central problem every executive in that room is trying to solve before Black Friday, not after it.
Where consumers draw the line, and where they do not
The research PYMNTS Intelligence brought to the roundtable draws a clear distinction in what consumers will and will not delegate to AI. Shoppers are generally comfortable letting AI help find the right gift, compare prices across retailers, or locate a deal, tasks that involve research and narrowing options. They are considerably less willing to let software choose the seller, select the payment method, or complete the purchase without their explicit approval, tasks that involve an actual financial commitment rather than information gathering.
That distinction maps onto a practical design principle retailers can act on immediately: AI assistance earns trust fastest in the discovery phase of a purchase and loses trust fastest the moment it starts making decisions about money. Retailers building holiday AI experiences around product discovery and comparison are working with the grain of consumer comfort. Retailers trying to push AI all the way through checkout without an explicit confirmation step are working against it.
The category effect nobody can ignore
The research surfaces a category-dependent pattern that complicates any one-size-fits-all agentic commerce strategy: a consumer may happily automate a repeat grocery order, a purchase with low financial stakes and high predictability, while wanting to personally choose a jacket, a sofa, or a luxury item, purchases where personal taste, fit, and higher cost all raise the bar for how much control a shopper wants to retain. The same person can sit at opposite ends of the automation-comfort spectrum depending entirely on what is in the cart.
That category effect means a retailer selling across multiple product types cannot deploy a single agentic commerce policy uniformly. A grocery or consumables retailer has more room to automate aggressively than a furniture or apparel retailer, and a retailer spanning both categories needs a genuinely different AI posture for each, not a single assistant tuned to a company-wide default comfort level that fits neither category well.
The gap between what AI can do and what shoppers will allow
The uncomfortable fact underlying this entire roundtable is that the technology has already outpaced the trust required to deploy it fully. AI agents are technically capable of researching, comparing, selecting a seller, and completing a purchase entirely without human involvement at any step, and several retail and payments vendors have shipped exactly that capability already. Consumer comfort with that full chain, per PYMNTS' research, has not caught up, particularly for considered or higher-stakes purchases where a shopper wants to retain the final say.
That gap is the actual holiday planning problem for retailers this year, more than any specific feature decision. Deploying the full capability the technology allows risks alienating a meaningful share of shoppers who are not ready to cede that much control, while under-deploying it risks losing ground to AI-native competitors and the discovery-and-comparison use cases where consumers have already shown clear enthusiasm.
What a sensible holiday rollout looks like given this split
The practical answer coming out of sessions like this one is a tiered approach: let AI do the work consumers already trust it to do, product discovery, price comparison, deal-finding, aggressively and visibly, while keeping a clear, unskippable human confirmation step before any payment method is selected or a purchase is finalized. That structure captures the conversion benefits of AI assistance without forcing shoppers past the trust boundary the research identifies.
Retailers should also expect that trust boundary to shift over the course of the holiday season itself as shoppers gain more exposure to agentic tools through repeated use. A rollout designed around today's comfort levels should be built with room to extend automation further as trust builds, rather than locking in a fixed experience that will look conservative by the time similar tools have had a full season to normalize with the broader shopping public.
Why this roundtable format itself is a signal worth noting
That Mastercard convened competitors and partners across the payments and retail ecosystem into one closed-door room to work through this together, rather than each company making its own independent bet on where the trust line sits, suggests the industry views miscalibrating this wrong at scale as a bigger risk than any single company's competitive advantage from getting there first. A retailer that pushes automation too aggressively this holiday season and visibly loses consumer trust creates a problem for the entire category's credibility, not just its own.
For retail technology leaders not in that room, the signal is to treat holiday agentic commerce planning as an industry-wide trust calibration exercise rather than a purely internal product decision. Watching how peers and competitors position their own AI shopping tools this season, and how consumers respond, is now a meaningfully useful input into getting your own deployment right, since nobody, including the 40 executives who spent a day working through it, has a fully settled answer yet.
