A two-day event, a full infrastructure stress test
Amazon's Prime Big Deal Days ran October 6 and 7, and Adobe put total US retailer online spend over those two days at $9.86 billion, up 8.5% from $9.09 billion a year earlier. Numerator's household survey data showed average spend per household rose 9% to $114.01, with 54% of shopping households placing two or more separate orders across the event. Those are respectable but unremarkable growth numbers on their own. The traffic mix underneath them is the part that should get a CTO's attention before Black Friday, not the headline growth rate.
Mobile accounted for 52% of online sales during the event, roughly $5.1 billion, up about 13% from 2025. Social media drove 7% of revenue and was the fastest-growing channel, up 40% year over year, with affiliates and influencer partners holding a 21.9% share, up 11%. Adobe's data showed influencer-driven traffic converting shoppers at eleven times the rate of social networks overall. That is a meaningfully different traffic composition than most retailers' infrastructure was built and load-tested against even two years ago.
The category spikes show where capacity planning gets hard
Adobe's category data from the event is a useful proxy for where Black Friday traffic concentrates. Electronics grew 85% versus average daily September sales, toys grew 158%, video games grew 90%, and personal care grew 73%. Individual products spiked even harder: headphones and speakers were up 300%, smartwatches up 255%, and Halloween costumes up 610% as the holiday calendar overlapped with the event. Those are not smooth, predictable curves, they are near-instant demand spikes concentrated in specific product categories and specific hours.
Buy now, pay later made up 6% of online orders during the event, $692.3 million, up 4% year over year. That figure matters less as a payments trend story and more as an infrastructure dependency: a BNPL provider outage or slowdown during a concentrated spike window now takes out checkout for roughly one in seventeen online orders, not a negligible edge case anymore. Any retailer that has not load-tested its BNPL integration specifically under spike conditions, separately from its core payment processor, has an untested single point of failure sitting in its highest-volume sales window.
AI traffic growth changes what the spike even looks like
Separately from the Big Deal Days numbers, Adobe has tracked AI-driven traffic to US retail sites rising 62% year over year in July and roughly 1,200% since October 2024, with AI-driven visits during the upcoming holiday season projected to rise another 130% year over year. That traffic does not behave like traditional search or direct navigation traffic. It arrives in conversational bursts tied to whatever an AI agent or chatbot is recommending at a given moment, often without the referrer and session signals a retailer's analytics and bot-management tooling were built to parse.
Combine that AI traffic growth with the channel mix from Big Deal Days, mobile-first, socially amplified, increasingly BNPL-financed, and the resulting Black Friday traffic profile looks meaningfully different from the one most retailers' infrastructure teams are planning capacity around. A load test built on last year's Black Friday traffic logs will systematically underestimate both the volume and the burstiness of what actually arrives, because a growing share of that traffic is now agent-mediated rather than human-browsed in the conventional sense.
Product data readiness matters as much as server capacity
There is a second, less obvious implication of rising AI traffic for infrastructure planning: AI agents and conversational shopping tools can only recommend products whose data is complete and well-structured enough for the agent to parse with confidence. A product feed with gaps in materials, sizing, or use-case attributes is effectively invisible to conversational discovery, regardless of server capacity or checkout speed. Retailers investing purely in infrastructure scaling while leaving product data work for later risk discovering that their site handled the Black Friday load fine, while AI-mediated channels simply never surfaced their inventory to begin with.
That makes product data quality a capacity planning item, not just a merchandising one, ahead of this holiday season. A retailer that stress-tests checkout and payment integrations but has not audited whether its product feed meets the attribute depth AI shopping tools need is solving half the problem. The two workstreams, infrastructure scaling and product data readiness, need to run on the same timeline into November, because a holiday traffic surge that cannot find the product it was looking for never generates the load that would have tested the infrastructure in the first place.
What to load-test between now and Black Friday
Three specific things are worth testing before the holiday peak, based on what Big Deal Days exposed. First, BNPL and alternative payment integrations under spike conditions specifically, since they now carry real order volume and are rarely part of standard load-testing scripts built around the primary payment processor. Second, mobile checkout performance under the kind of concentrated category spikes seen in electronics and toys, since mobile now carries the majority of volume and a slow mobile checkout during a 300% demand spike compounds fast. Third, bot and agent traffic handling, since AI-driven visits are no longer a rounding error and conventional bot-mitigation rules risk either blocking legitimate agentic shoppers or failing to recognize the traffic pattern at all.
None of this requires new technology so much as it requires treating a mid-October promotional event as the dress rehearsal it actually is. Retailers that pull their own Big Deal Days or Fall Deals traffic logs and compare mobile share, social referral share, and payment method mix against last year's Black Friday baseline will find the gaps worth closing in the next six weeks. Retailers that wait for Black Friday itself to discover those gaps will be debugging live, during the only week of the year where the cost of downtime is highest and the attention span for apologizing to customers is lowest.
