Global ecommerce is projected to hit $6.56 trillion in 2025, and retailers are racing to close the AI readiness gap

Global ecommerce is on track to reach $6.56 trillion in 2025, according to Forbes, and online channels are expected to account for nearly a quarter of all retail sales by 2027. Those numbers set the ceiling. What is becoming clearer in mid-2026 is how wide the gap remains between the market retailers want to capture and the AI infrastructure most of them actually have in place.

Digital Commerce 360 identified three distinct AI-related problems retailers are actively trying to solve right now: data quality issues that block agentic deployments from working as intended, AI assistant performance that requires significant tuning before it moves business metrics, and cross-border commerce complexity that most existing stacks simply were not built to handle. Each is a different failure mode, and each requires a different fix.

The data problem is upstream of everything else

Before any AI assistant, recommendation engine, or agentic workflow can function reliably, it needs clean, structured, and consistently formatted product and customer data. That prerequisite is the problem most retailers are hitting first. According to Digital Commerce 360's trend analysis published July 23, the data readiness gap is one of the three central AI challenges retailers are working through this year. Incomplete product catalogs, inconsistent attribute tagging, and siloed customer records all degrade model performance in ways that are invisible until they hit the customer experience.

The operational implication is direct: teams responsible for product information management, MDM, or catalog ops are now de facto blockers or enablers of AI initiatives. Investment decisions that looked like back-office hygiene projects six months ago now carry a different business case.

Michaels goes live with Gemini and shares the numbers

Michaels, the arts and crafts retail chain, is one of the more concrete examples of a large-format retailer moving from AI experimentation into production. The company deployed a Google Gemini-powered AI assistant and, as reported by Digital Commerce 360 on July 22, is now sharing early results from that rollout. The Michaels case is notable not just for the deployment but for the transparency: publishing performance data, even early-stage, sets a benchmark that peers can compare against and procurement teams can use to frame their own vendor conversations.

The Gemini integration positions Michaels within a broader shift among specialty retailers to move AI assistants beyond simple search and into guided shopping experiences. Getting to measurable output requires investment in both the model layer and the underlying product data that feeds it, reinforcing the upstream problem described above.

Agentic AI takes aim at cross-border complexity

Cross-border commerce is the third problem, and in many ways the hardest. Duty calculations, localized payment methods, country-specific compliance requirements, and last-mile logistics variability all compound in ways that rule-based systems struggle to keep current. Sazo, a platform covered by Digital Commerce 360 on July 23, is building agentic AI specifically for this infrastructure layer. The approach uses AI agents to handle the dynamic, multi-variable decisions that cross-border transactions require, decisions that change with regulatory updates, currency fluctuations, and carrier availability.

The agentic model matters because it does not just automate a fixed workflow; it can adapt when conditions change. For a retailer selling into a dozen markets, that adaptability is the difference between a scalable international operation and a manual exception-handling queue.

What this means for your team

• Audit your product data infrastructure before scoping any agentic AI project: the model is only as reliable as the catalog and customer data feeding it.
• Use Michaels' Gemini deployment as a benchmark when evaluating AI assistant vendors, ask any shortlisted provider to show comparable early-stage performance metrics, not just case study claims.
• If cross-border revenue is a growth priority, evaluate whether your current ecommerce stack can handle dynamic duty, compliance, and logistics decisions at scale, or whether a purpose-built agentic layer like Sazo is worth a proof-of-concept.
• Treat data readiness investment as a prerequisite in AI budget planning, not a parallel workstream, teams that skip it are likely to restart the AI project after the first production failure.