Agentic commerce uses AI systems to help carry out shopping tasks, from understanding a need and comparing products to taking an authorised action. For a merchant, the first useful investment is dependable product and policy information that remains consistent through discovery, checkout and fulfilment.
Why is product readiness a current business issue?
AI shopping is creating another route into the buying journey, but discovery and completed autonomous purchases are different things.
In July 2026, Salesforce reported 200% year-over-year growth in the use of agentic search as a shopping starting point. Its State of Commerce research combines surveys and platform activity. The figure is a reported growth rate within that research, not a claim that most purchases are now autonomous or a forecast for an individual store.
Shopify's 2026 guidance describes AI shopping channels with different discovery and checkout capabilities. Availability depends on the channel and merchant. Our recommendation is to prepare the information and operational handoffs that remain useful across those channels, then evaluate an integration against its actual eligibility and economics.
Sources: Salesforce — Shopping's New First Step: agentic search research, July 2026; Shopify — Agentic Commerce: Benefits & How To Get Started, April 2026
Which product facts need an accountable source?
Separate stable product specifications from changing offer and fulfilment information.
A product description, a stock count and a delivery promise are different kinds of information. Give each an owner and decide which system is authoritative. Copying the same field manually into several channels makes disagreement harder to diagnose.
The following is a suggested audit matrix. Adapt it to the catalogue and the claims shoppers need to evaluate; it is not a universal feed specification.
| Information | What to check | Suggested owner |
|---|---|---|
| Identity and variants | Stable identifiers; size, colour and model relationships | Catalogue team |
| Specifications | Materials, dimensions, compatibility and supported uses | Product owner |
| Offer | Price, currency, market, promotion dates and conditions | Commerce team |
| Availability | Sellable stock and the timestamp of the latest update | Inventory operations |
| Delivery and returns | Region, exclusions, applicable costs and policy version | Operations and support |
How should variants and structured data work together?
Keep the visible product, its machine-readable description and any submitted feed consistent.
Google supports ProductGroup markup for related variants, alongside Product information for individual items. This helps describe relationships such as sizes and colours. It does not authorise a purchase, reserve stock or make every AI platform consume the same data.
As an illustrative check, take a jacket with two sizes. Open the small variant, compare its identifier and stock status with the catalogue and checkout, then repeat for the large variant. If a shared description says both are available while one cannot be bought, fix the source and synchronisation path rather than adding more promotional copy.
Test at least a normal item, an unavailable variant and a product with regional restrictions. Those cases expose different failures. Keep public specifications visible in readable text as well as relevant structured fields.
Sources: Google Search Central — Product variant structured data
What changes when an assistant can take an action?
The transaction system must validate the current request independently of the assistant's recommendation.
Imagine an assistant recommending that jacket while the last unit sells on another channel. The recommendation is now stale. The transaction path needs a current availability check and a clear response when the item cannot be supplied. A confident conversation is not a stock reservation.
Define who authorises an order, which terms the shopper sees, and how the order is reconciled with payment and fulfilment. If a request is retried after a timeout, the receiving system needs a way to recognise it without creating a second order. Preserve a route to support when the final state is uncertain.
These are operational design recommendations, not a description of every shopping platform. Check the integration's documented behaviour before deciding which responsibilities belong to your store.
How can a small team start without a rebuild?
Audit one commercially useful product group and follow its data through the existing systems.
Choose a category with recurring customer questions or frequent catalogue corrections. Record missing specifications, inconsistent policies, variant mismatches and delayed updates. Fix the source fields first, then check the storefront and each enabled channel. Keep a dated record of what was verified.
Measure field completeness and mismatch frequency alongside qualified referrals, order exceptions and returns caused by inaccurate information. Discovery is only part of the result: extra orders with incorrect expectations can add support cost. Review the programme using the whole customer journey.
Common questions
Do we need to replace our ecommerce platform?
Not simply to improve product data. Start with the current catalogue, feeds and integrations. A platform change needs a separate business case based on constraints the existing system cannot reasonably resolve.
Is agentic commerce the same as conversational commerce?
Conversational commerce describes a shopping interaction through dialogue. Agentic commerce describes software carrying out parts of the task. They can overlap, but a chat interface alone does not establish authority to place an order.