AI in Retail in 2027: Agents on the Shop Floor and in the Chat Window

How retailers actually use AI in 2027 — support agents, visual search, demand forecasting and the unglamorous integrations that decide whether any of it works.

During the 2025 holiday season, traffic from generative AI tools to US retail sites rose 693.4% year over year — and those AI-referred shoppers converted 31% better than everyone else (Adobe, 2026). Read that again, because it rearranges this whole article. The most consequential retail AI of 2027 is not the agent a retailer installs on its own site. It is the assistant the customer consults before arriving — ChatGPT, Gemini, Perplexity — which now does the shortlisting that search results and shop assistants used to do. Retail AI is now two jobs: be recommended by machines you don't control, and then resolve, sell and serve on the properties you do.

Where the money is

Support that resolves, not deflects

Order status, returns and exchanges are structured, verifiable tasks — ideal for agents with tool access to the order system and a hard rule: act only on data you can cite from a system of record, escalate everything else with context attached.

Guided selling

The highest-converting retail AI pattern is boring: ask three clarifying questions, narrow a catalogue of thousands to three options, and explain the trade-offs. It works because it mirrors a good shop assistant, not because the model is clever. Keep the human holding the final buy decision: consumers are comfortable letting AI advise, compare and shortlist, but most do not want it completing the purchase for them (Qualtrics XM Institute, 2026).

Forecasting and replenishment

Demand models that blend weather, local events and social signals now run at SKU-store level. The wins show up as fewer stockouts, not as keynote demos.

Failure modes to avoid

• Invented policy. An agent that hallucinates a returns policy creates legal exposure. Ground every policy answer in retrieved, dated source text.
• Channel amnesia. Customers start on WhatsApp and finish on web. If the agent cannot carry context across channels, you have built three worse chatbots instead of one agent. Zendesk's CX Trends 2026 names the fix "memory-rich AI": 81% of consumers want an interaction to continue without backtracking (Zendesk, 2026).
• Success measured in deflection. Containment rate without resolution and re-contact rates is how teams automate their way into silent churn.

Retail is one chapter of a wider pattern — see the AI customer support agents guide for the operating model.

The storefront you don't own

The original edition of this article treated the retailer's own channels as the whole board. The 2025 holiday season — the first measured at full scale — moved the opening move elsewhere. Adobe Analytics, which tracks over a trillion visits to US retail sites, recorded a 693.4% year-over-year increase in traffic arriving from generative AI tools during November–December 2025, inside a record $257.8 billion online season (Adobe, 2026). More important than the volume is the quality: AI-referred visitors converted 31% more than other traffic and stayed 45% longer. The assistant pre-sells; the site has to close and serve.

McKinsey's agentic-commerce research puts the trajectory plainly: by 2030, up to $1 trillion of US B2C retail revenue could be orchestrated by agents, with $3–5 trillion globally (McKinsey, 2025). Salesforce already estimated AI influenced more than 20% of global online retail sales in the 2025 holiday season (Digital Commerce 360, 2026). None of this requires the retailer's permission. What it requires is machine-readable truth: complete product data, current prices and stock, policies in plain text an assistant can quote without inventing. The failure mode from the original article — the agent that hallucinates a returns policy — now happens on other companies' surfaces, in front of customers you never see.

There is a hard limit, and it is trust. Qualtrics XM Institute's consumer research finds shoppers gladly delegate research, comparison and shortlisting to AI, but not the purchase itself; 53% worry about privacy in AI-enabled support, and only two in five believe organisations use their data responsibly (Qualtrics XM Institute, 2026). So the 2027 operating split is: let the assistants advise and refer; on your own properties, let agents resolve and assist; and keep the final buy decision visibly human-approved. Retailers who respect that line get the referral traffic. Retailers who try to automate past it will discover the line was load-bearing.

Sources

1. Adobe. (2026). "Adobe: Holiday Shopping Season Drove a Record $257.8 Billion Online with Consumers Embracing Generative AI Tools". Adobe Newsroom.
2. McKinsey & Company. (2025). "Are you — or your AI agent — shopping on Cyber Monday?" (The Agentic Commerce Opportunity). McKinsey.
3. Digital Commerce 360. (2026). "Generative AI shifts online holiday shopping traffic in 2025". Digital Commerce 360.
4. Customer Experience Dive. (2026). "Consumers trust AI to advise, but don't consent to it purchasing for them". CX Dive.
5. Retail Dive. (2026). "NRF forecasts 4.2% holiday sales growth, muted seasonal hiring". Retail Dive.
6. Zendesk. (2026). "Contextual Intelligence Becomes the New Standard for Exceptional Customer Experience in 2026". Zendesk.