Shoptalk Spring asked retailers a question most weren’t ready to answer: what has your AI investment produced? The conference, held March 24–26 in Las Vegas, marked a clean break from the announcement-driven AI panels of prior years. Speakers were expected to name production systems, cite metrics, and defend ROI. PwC’s Global CEO Survey, covering 4,454 business leaders, provided the backdrop: only 12% report that AI has delivered both lower costs and higher revenue, while 56% report neither benefit. The permission-to-experiment phase is over, and what follows will separate retailers running AI in production from those still circulating slide decks.

The retailers with results brought specific numbers. Tecovas, a Western boot brand, reported a nearly 10% revenue lift in test categories from AI-driven inventory optimization. The Vitamin Shoppe’s COO Andrew Laudato described his team’s approach: “We didn’t set out to use AI — we set out to provide a better customer experience.” Gap Inc. applied AI to product discovery, checkout, and sizing. Sephora announced a collaboration with OpenAI to extend its conversational shopping experience into emerging AI platforms. Home Depot, which generates roughly $25 billion in online revenue, disclosed parallel AI experiments with Google, OpenAI, and Microsoft.

The demand side of AI commerce is moving faster than most retail organizations. Adobe reported at Shoptalk a 700% year-over-year increase in AI-driven site traffic during the 2025 holiday season, along with a finding that 45% of consumers now use AI somewhere in their buying journey. eMarketer forecasts that AI platform-driven ecommerce will reach $144 billion by 2029, roughly 8.8% of total retail ecommerce sales. Pinterest told the same audience that 39% of Gen Z users now reach its platform before Google for product discovery. Google disclosed that its AI Overviews already reach two billion people. If the traffic is arriving through AI channels and the retailer has no presence there, the sale completes somewhere else.

The conference rewarded speed to production over strategy documents. PebblePost CEO Jacob Ross put it plainly: “Marketers want to find proof points to build confidence in a world that feels like maximum hype.”

Shoptalk also exposed what production-ready AI demands from the retailer’s own operations. Investor panelists at a Shoptalk session — Matt Nichols of Commerce Ventures and Janie Yu of LFX Venture Partners — identified four infrastructure gaps blocking agentic commerce: universal APIs for agent-placed orders, multi-merchant cart solutions, standardized product identifiers, and clean product data. None of these are AI problems; they are data-architecture problems that have sat on roadmaps for years. The retailers showing results at Shoptalk had solved them before deploying a model. Tecovas, the Western bootmaker, runs seven to ten AI agents simultaneously in its stores and reports that training staff takes five times longer than building the software.

Early movers always look impressive on conference stages, and Shoptalk selects for success stories. This is plausible, but only if you assume the external environment is holding still. Merkle estimates, as cited by Omnitalk, that 50% of ecommerce could involve agentic or answer-engine transactions by 2029. If that forecast is even directionally correct, the companies without production AI systems are not being cautious — they are falling behind a distribution channel that is forming without them.

Twelve percent of CEOs reporting measurable AI value is the gap that every other retailer now has to explain to its board.

Shoptalk’s shift from “what are you planning?” to “what have you shipped?” will compress timelines across the industry. Boards that approved open-ended AI budgets in 2024 will expect auditable results by the end of this year. The conference did not name a winning AI application. It established that the burden of proof has moved from the skeptics to the spenders, and two quarters is not a long time.