Until now a person visits your store, reads the product page and clicks buy. In agentic commerce an AI agent takes over those steps. It queries catalogue, price and stock through an interface, compares with other stores and builds the cart. It never sees your theme. It reads data.
That moves the work. What sits in images, tabs and running text today has to exist as a field. What a member of staff settles by phone today has to exist as a rule in the system: delivery time, returns, minimum quantity, shipping abroad.
This page covers purchases made by agents. Being named as a source in assistants' answers is covered under Shopify GEO. Indexing, canonicals and redirects for classic search are described under Shopify SEO.
Marketplaces have ordered by machine for years. The requirements match those of an AI agent: identifier, price and stock have to be right, or a wrong order is created. Kanela is not an agentic commerce project. It shows how that foundation is built.
The Swiss online pharmacy carries more than 40,000 products and also sells through Zur Rose, Galaxus and Manor. We merged identifiers across SKU, EAN, GTIN and Pharmacode. The feed to Zur Rose covers 27,302 checked items, with price and stock updates every 15 minutes. Galaxus runs on five separate feeds. A dedicated application evaluates own stock and supplier availability together and stops unusually large stock changes automatically. 1,820 stock levels in Shopify were corrected in the process. The figures are in the Kanela case study.
At Hairshop Pro, Plentymarkets stayed in place as the PIM after the move to Shopify Plus. An interface synchronises product master data, stock, prices, orders and fulfilment status with the store, and nobody reconciles by hand any more. The Hairshop Pro case study describes the setup. The service behind it is system integration.
Four things: a complete catalogue with unique identifiers, stock and prices that are correct at the moment of the query, shipping and returns rules held as data, and a checkout that runs through without a question back to the buyer.
An agent matches your product through GTIN, SKU and variants. If the identifier is missing, or a variant is described only in the title, it cannot match your offer to its user's request. Selling across borders adds currency, tax and duties. An agent ordering from Switzerland needs the Swiss price and the landed cost, which is a matter of how Shopify Markets is set up.
In the checkout, purchases fail on mandatory fields only your store knows, on shipping rules that apply in the last step, and on payment methods an agent cannot trigger. We test the flow with test orders per market. The page on Shopify checkout describes what can be adapted.
With the Storefront MCP, Shopify provides an interface through which an agent queries the catalogue, cart and policies of a store. MCP stands for Model Context Protocol. The agent no longer depends on what a crawler happens to find on the page.
Two protocols cover the purchase itself. The Agentic Commerce Protocol, ACP, comes from OpenAI and Stripe. The Universal Commerce Protocol, UCP, was introduced by Shopify and Google. Both are young and their terms are changing, including the countries they cover. We therefore state no fees and no country lists here and check the status at the time of your project.
NICCOS is an OpenAI partner and has been a Shopify Plus Partner since April 2022, with 19 reviews and a 5.0 rating in the Shopify Partner Directory. Our recommendation does not depend on the protocol: keep the data in a state that can serve any channel. The protocol then remains a question of reach and margin.
Orders from agents have to be labelled on arrival, with channel and source on the order in Shopify. Without the label they appear as direct traffic in reports, and you cannot judge whether the channel pays off.
We define the labelling together with the rest of the measurement plan, so that GA4, Shopify and the ERP assign the same order to the same channel. The page on tracking setups describes how that plan is built. Sphen.ai, a joint product of getSichtbar and NICCOS in the Shopify App Store, checks with buyer questions whether AI assistants name your store.
It is not relevant while stock and prices drift apart between ERP and store. An agent exposes those errors faster. It does not fix them. Integration comes first.
It is also not relevant for ranges that need individual advice or a quote before a price exists. We say so in the first conversation. How much revenue agents generate today depends on industry and market, and we give no forecast for it.
An agent stops as soon as a fact is missing or contradicts another. These are the four places we check first.
GTIN and SKU are missing or assigned twice. The agent cannot match your offer to a product.
ERP and store hold different values. The order arrives and has to be cancelled.
Shipping, returns and delivery time sit on a text page and not as data on the product or in the store.
Mandatory fields, late shipping rules or payment methods stop an agent from completing the purchase.
The work is in the data and in the checkout. The protocol comes after that.
Identifiers, variants and the properties that matter for the purchase are complete in fields and metafields.
The ERP or PIM owns the values, and Shopify receives them through a tested interface.
The checkout runs through without a question back, and every order carries channel and source.
We check identifiers, variants and mandatory details across the whole range and list the gaps.
For price, stock and product copy we define which system owns the value.
Details from images and running text move into product fields and metafields.
Shipping, returns and delivery time are stored so that an interface can return them.
We query the store through the interface and run test orders through to the ERP.
Channel and source are recorded on the order and reported separately.
The agent searches the catalogue for properties its user has named.
It queries the current price and stock per variant, in the buyer's market.
It builds the cart and hands over to the checkout.
It reads the store's rules and passes them on to its user.
Your catalogue is maintained
Identifiers, variants and properties are largely complete, and you want to close the remaining gaps.
ERP and store are connected
Stock and prices come from one owning system and are correct in the store.
You already sell through feeds
Google Shopping or marketplaces are running. The requirements of an agent are similar.
You sell in several markets
Prices, currencies and duties are set per market, and an agent should receive the right values for each.
Stock levels drift apart
If ERP and store hold different values, the data flow has to be right first.
To system integrationYou want to be named in answers
If your aim is to be cited as a source and not yet the purchase, GEO is the fitting service.
To Shopify GEOThe checkout loses buyers today
If human buyers already drop out in the checkout, the work starts there.
To Shopify checkoutYou want to know where your store stands first, build the data basis, or connect a channel. That gives three shapes of collaboration.
For stores that want to know whether an agent could order correctly today.
Ends with a list of gaps, sorted by their effect on an order.
For stores whose product details and rules sit in text and images.
Ends when catalogue, stock and rules can be retrieved in full through the interface.
For stores with a clean data basis that want to open a channel for agents.
Ends when test orders arrive correctly in Shopify and the ERP and are attributed to the channel in reports.
Every figure appears in the case study, where you can verify it.
Groundies, MYCR7 and Gusti Leder are three of thirteen projects we have published with starting point, delivery and result.
We review the catalogue, stock and checkout of your store and show you where a purchase would fail. The first assessment is free and gives you an initial evaluation, not a full audit or a worked-out concept.
NICCOS
The page could not be loaded.
Please reload the page. If an update has just gone live, this will load the latest version.
