AI by IndustryRetail

AI for retail and e-commerce: discovery, support, and inventory

AI-referred shopping traffic grew 4,700% from a tiny base. What independent retailers should do with that signal—and why product data, support, and stock discipline come before a shopping bot.

By Adi Huric, founder of Most AI LabsAugust 20269 min read

On this page

AI shopping traffic is real—and still small

Product truth is the new storefront

Support is the first measurable automation

Forecasting needs a baseline it can beat

The honest bottom line

Sources

The first retail AI project is usually not a virtual stylist. It is making sure every channel agrees on what the product is, what it costs, whether it is in stock, and what happens when the customer sends it back.

AI shopping traffic is real—and still small

Adobe analysed more than one trillion visits to US retail sites and surveyed over 5,000 consumers. In July 2025, traffic from generative-AI sources was up 4,700% year over year. Thirty-eight percent of consumers said they had used generative AI for shopping. Visitors arriving from AI sources were 10% more engaged and stayed 32% longer than visitors from non-AI sources.

Growth percentages explode when the starting point is tiny. Adobe explicitly describes AI-referred traffic as modest relative to channels such as paid search and email. The finding supports preparing for AI-mediated discovery; it does not support moving a proven acquisition budget into an unmeasured “AI channel.”

The National Retail Federation’s 2025 AI report surveyed 56 AI leaders at US retailers. It is useful for priorities and concerns, but the respondents are specialized leaders, not a representative sample of independent shops. Statistics Canada found 16% of retail businesses planned AI software adoption in 2025, while 23.8% of retail workers reported workplace generative-AI use by March 2026. Again, business deployment and individual tool use are different measures.

Product truth is the new storefront

Search engines, marketplaces, chat assistants, ads, and customer-service tools all consume the catalogue. Fix titles, identifiers, variants, dimensions, materials, compatibility, availability, price, shipping, returns, care, and warranty at the source. Give each field an owner and update rule. AI can flag inconsistencies and draft descriptions; it should not invent an attribute because the copy looks incomplete.

If an assistant cannot tell whether the blue medium is in stock, the problem is not the assistant. It is the product record.

Support is the first measurable automation

Order status, return eligibility, size guidance from published charts, care instructions, and store information are bounded questions. Connect the assistant to the order and policy systems, expose the source it used, and hand off when identity, payment, fraud, damage, a chargeback, or an exception appears.

Measure self-service resolution, repeat contact within seven days, incorrect-answer rate, refund and appeasement cost, escalation time, and customer satisfaction. “Containment” is a dangerous goal by itself: a system can keep customers away from staff by making them give up.

Forecasting needs a baseline it can beat

Forecasting and replenishment tools may help where the retailer has clean SKU history, promotions, stockouts, returns, lead times, and seasonality. Compare the model with the existing method using forecast error, in-stock rate, aged inventory, markdowns, and lost sales. Hold out a product group or location when possible; otherwise nobody knows whether a warm season or successful campaign deserves the credit.

Generative AI is good at summarizing why the forecast changed. It is not automatically good at producing the forecast. Keep the numeric model, source data, constraints, and purchasing approval separate from the conversational explanation.

Watch out for this

What we could not verify: there is no universal independent return for an SMB retail chatbot, recommender, or forecasting tool. Retail results depend heavily on catalogue size, traffic, repeat purchase, margins, and data quality.

The honest bottom line

Make product data reliable. It improves search, marketplaces, support, and AI discovery at the same time.
Automate bounded support. Orders and policies are safer than open-ended product promises.
Test forecasting against the current method. Measure stock and margin outcomes, not model sophistication.
Treat AI traffic as an emerging source. Track it separately without confusing a large growth rate with a large share.

Our free 7-day audit starts with catalogue, support, and inventory data—not a demo.

The lead calculator is intentionally not an e-commerce revenue calculator. Retail sales require traffic, conversion rate, average order value, returns, and margin; a lead-cost model would give false precision.

Sources

Where this leads

AI & AutomationThe MOST service this article supports Lead CalculatorModel what a lead is worth in your industry.

Next step

Find the first project worth testing.

The 7-day audit maps where AI earns its place in your operation, and where it does not.

Start with the 7-day audit Or talk to the founder first