The AI shopping surge is no longer a forecast, it is this week's traffic
A YouGov survey conducted in October 2026 found that 34 percent of shoppers who plan to participate in Black Friday or Cyber Monday say they will use AI to help with their purchases this year, more than double the 16 percent who reported the same a year earlier. That growth rate matters more to infrastructure planning than the absolute number, because it means whatever load-testing assumptions a retailer's technology team used for last year's peak traffic are very likely already out of date heading into this weekend.
The nature of that AI-driven traffic differs meaningfully from a human shopper clicking through a site at a human pace. HUMAN Security research into 2025 agentic AI activity on retail sites found that 8.8 percent of observed agent activity occurred specifically on account pages and 2.3 percent on checkout pages, concentrating automated traffic at exactly the two points in a retail site where a failure, whether a bug, a timeout, or a fraud attempt, causes the most direct damage to both revenue and customer trust.
The failure mode executives are actually worried about
Don Boxley, CEO and co-founder of infrastructure vendor DH2i, frames the central risk in a way that should reorient how retail technology teams think about Black Friday readiness entirely: a website can look completely functional to a shopper, loading pages normally and rendering product images correctly, while the database behind it has quietly become unable to confirm inventory or actually process a new order. That is a fundamentally different failure than the crashed, error-page outage most Black Friday contingency plans are built around, and it is far harder to catch with conventional uptime monitoring, since the monitoring tools themselves often just check whether a page loads, not whether the transaction underneath it actually completes correctly.
This distinction matters specifically because of where AI shopping agents concentrate their activity. An agent checking inventory across dozens of retailers simultaneously, or comparing prices in real time before completing a purchase, is generating exactly the kind of sustained database query load that exposes a backend confirmation failure long before it would show up as a visible site outage. A retailer whose database silently falls behind on inventory confirmation during a traffic spike may not notice the problem until customer complaints or chargebacks arrive days later, by which point the damage to both revenue and trust is already done.
The trust gap that makes database accuracy a customer-facing problem
Separately, the same YouGov research found that 56 percent of shoppers who already use AI tools for purchases say they worry about receiving inaccurate or outdated information during the process. That concern is not paranoia; it is a reasonably accurate read of how brittle AI-assisted shopping becomes when the underlying retailer data an agent is querying is itself stale, incomplete, or simply wrong, which is precisely the condition Boxley describes as a backend database quietly failing to keep pace with demand.
The compounding effect here is the detail retail technology leaders should sit with longest: a shopper's AI assistant is only as accurate as the retailer's own inventory and pricing data, so a database confirmation failure on the retailer's side does not just cause a failed transaction, it actively produces the exact kind of inaccurate information that already makes AI-assisted shoppers suspicious of the process. A backend infrastructure problem becomes, within seconds, a consumer trust problem that the retailer did not directly cause in the shopper's eyes but will absolutely be blamed for.
The security dimension layered on top of the capacity problem
Boxley also warns that not every automated visitor hitting a retail site during peak shopping periods is a legitimate shopping agent acting on a customer's behalf. Some automated traffic actively probes for vulnerabilities or attempts unauthorized access, traffic that looks superficially similar to legitimate AI shopping activity in raw volume terms but carries a fundamentally different intent. His recommended approach is to balance accepting legitimate automated traffic against protecting critical infrastructure, giving each connection access only to the specific resources it actually needs rather than broad backend access by default, a principle borrowed directly from zero-trust security architecture applied to an e-commerce context.
Richard Copeland, CEO of Leaseweb USA, adds the infrastructure-diversification half of the readiness argument: retailers should avoid depending on a single infrastructure environment, instead building hybrid setups that combine on-premises systems with public cloud capacity so a spike or failure in one environment does not take down the whole operation. Copeland's blunter framing is the one worth repeating to any executive still treating this as a purely technical concern: for a retailer without a plan, Black Friday and Cyber Monday will likely be a nightmare, and he notes some competitors began preparing for this specific dynamic as much as a year in advance.
What this means for your own Black Friday readiness check
If your technology team's Black Friday monitoring is built primarily around page-load uptime and error-rate dashboards, treat Boxley's warning as a direct prompt to add a specific check for transaction-completion accuracy, not just site availability, since the failure mode he describes is explicitly one that passes a standard uptime check while silently failing customers. Test specifically whether your inventory and checkout systems can confirm stock and process orders correctly under the kind of sustained, concentrated query load an AI shopping agent generates, rather than only load-testing for raw page-request volume.
With AI-assisted shopping roughly doubling year over year by YouGov's measure, and agent activity already concentrating on account and checkout pages specifically, this is also the moment to confirm your access controls distinguish legitimate shopping agents from exploratory or malicious automated traffic, rather than treating all non-human visitors as one undifferentiated category to either block wholesale or allow wholesale. Retailers that built hybrid, diversified infrastructure specifically for this dynamic months ago, per Copeland's timeline, are the ones least likely to discover a silent database failure in the middle of this weekend's peak traffic.



