AI Agents in E-Commerce: Field-Level Revenue in Shopify Sidekick, BigCommerce's MCP Protocol, and Ozon's 2036 ForecastThematic roundup · digest for November 22
Surrealist still life with stone catalog blocks, digital product cards, and an MCP port
Thematic roundup3 sections≈ 5 min

AI Agents in E-Commerce: Field-Level Revenue in Shopify Sidekick, BigCommerce's MCP Protocol, and Ozon's 2036 Forecast

When launching a new product on a marketplace, it is tempting to pour your budget into studio shoots and infographics while filling in specification fields at the last minute before shipping. Shoppers buy with their eyes anyway, so why spend time on boring lines of text? In practice, that habit is getting expensive. Three fresh signals from the US and Russia show how AI already works with catalogs: for analytics and auto-selection, a machine needs clean structured fields, prices, and stock counts, and it will simply skip right past a pretty cover.

Shopify Sidekick: Revenue Breakdown by Custom Product Fields

Alongside basic prices and titles, online store and marketplace listings store custom fields (metafields) and composite objects (metaobjects). Sellers use them for niche details like fabric type, sole material, collection, or supplier.

Previously, figuring out which material drove the most revenue meant exporting a spreadsheet to Excel and merging columns by hand. Since November 17, 2026, the built-in AI assistant Shopify Sidekick removes that routine. You just ask in plain language: 'Show total sales by fabric for last quarter.' The service returns a revenue breakdown by values of the 'fabric' field right next to your sales metrics.

There is one catch: in the Shopify admin, the 'Filter or group data in Analytics' option must already be enabled for that field. If you filled the field inconsistently, the assistant has nothing to count. It supports common data types across products, variants, customers, and orders: text, numbers, product references, URLs, dates, colors, booleans, and lists of text or numbers. Some complex types and reports are not supported yet. Shopify has not published accuracy benchmarks for Sidekick or numbers on its sales impact, but the logic is clear: clean attributes save hours of manual reporting.

BigCommerce Storefront MCP: Direct AI Access to Catalogs and Carts

While Shopify let AI into seller reports, BigCommerce opened the door straight onto the storefront. Previously, AI assistants pulled data from store sites blindly—scraping HTML and tripping over banners and pop-ups. Now, Storefront MCP is available in Early Access. MCP (Model Context Protocol) is an open standard that lets assistants like ChatGPT or Claude connect to a catalog directly. Instead of wandering a dark warehouse with a flashlight reading box labels, the assistant walks up to a service window and talks to the database in clear commands.

A store owner turns the feature on in the control panel (Settings → API and MCP). Deployment takes up to 10 minutes, after which the platform spins up a managed endpoint at https://<store-url>/api/mcp over Streamable HTTP (separate from the admin Management MCP). The assistant gets four tools: searching products and variants (Search products), reading specs and prices (Get product details), managing the cart (Manage cart), and generating a checkout link (Create checkout).

In practice, it is simple. A shopper tells their assistant they need hiking boots for a weekend mountain trip. Through MCP, the AI finds matching models (say, a pair for $129.00), confirms details, adds an option to the cart, and returns a secure Proceed to checkout link—the human finishes paying on the store's site. For now, the basic B2C flow only supports guest mode: the assistant cannot see past order history or personal discounts (scoped permissions and session sync exist only for logged-in buyers in B2B Edition).

Ozon Survey: 60% of Russians See AI-Agent Shopping as Realistic by 2036

It is easy to think Western storefronts on Shopify and BigCommerce are far away, while bright photo stickers will decide sales on WB and Ozon for years to come. A fresh Ozon survey for Platform Economy Day (survey results) quickly shakes that assumption. The company did not disclose the exact sample size, but the numbers stand out: 60% of Russians believe it is realistic that by 2036 AI agents will select and order everyday goods on their own, leaving the human only to tap 'confirm'.

Not everyone wants to wait ten years. Already, 61% of respondents are willing to let AI buy without confirming every transaction. However, 56% of them would set a strict limit, such as up to 3,000 ₽ per order, and 71% of all respondents named fraud protection as the top condition. Age gaps show up right away: young adults aged 18 to 29 are more open to limits above 10,000 ₽ or no cap at all, while among those aged 46 to 61, 45% oppose charges without confirmation, even though 60% are still open to delegating routine orders to AI in the future.

The takeaway for your next listing audit is straightforward. When an algorithm shops for a buyer, it compares hard facts: specs, price, rating, reviews, and delivery times. An agent won't even glance at a slick gradient or a '100% cotton' badge on your third infographic slide—it will check the 'Composition' field. If that field is blank or lazily says 'textile', your listing gets skipped before a human ever sees the checkout link.

Want a deep diveVotes help us choose what to break down in detail. This material has 0 votes.
Sources · 6