En
← All articles

The shopping cart that no longer counts as abandoned

BigCommerce is adding storefront AI while changing how external agent carts are tracked. Merchants need to measure the two journeys separately.

ColumnE-commerceAuthorOpen Market NotesTypeArticle
Hand-drawn shop with two speech bubbles, one red.

An abandoned cart used to look like a simple event: a shopper assembled a basket and left without paying. Even then, a basket could mean comparison rather than purchase intent. With an external shopping agent, the same person might ask several stores for a quote, put an item in one cart while comparing another, or switch back to the merchant’s site to finish. What the store observes is now partly determined by which interface created the cart.

That distinction became concrete in November. On November 16, Commerce began rolling out Shopping Agent to eligible BigCommerce merchants. It is an assistant on the merchant’s own storefront, where it can help customers compare products and move toward checkout. The November 19 developer changelog describes a separate path: carts created through checkout on Universal Commerce Protocol (https://ucp.dev/specification/overview/) are no longer tracked as abandoned or followed by the corresponding email. Other cart sources retain their former treatment. The changelog also says BigCommerce is beginning to roll out the August UCP version alongside the April version over the coming weeks. It does not establish that the update is already present across all stores. The storefront assistant and the external protocol are different surfaces, not two names for one product.

First decide what a cart is evidence of

Take a shopper buying ink for an older printer. On the merchant’s site, the assistant can ask for the exact model and suggest compatible variants. The shopper can inspect the product page and move to checkout. An external agent, by contrast, might assemble a prospective basket as one step in comparing sellers. The presence of a cart is evidence that work occurred, not proof that the shopper intended to buy from that store. Treating every such basket as permission to send a reminder risks optimizing for events a human would never recognize as unfinished shopping. That is one possible operational rationale for keeping the paths separate, not an explanation BigCommerce gave for its change.

The opposite problem is a reporting illusion. If an analyst watches only the native abandoned-cart count, a fall can look like improved checkout performance even if customers are simply arriving via an uncounted path. Keeping a distinct count of checkout sessions by entry route and of paid orders lets the merchant test that interpretation. Where an external interface hands the buyer back to the store, linking the session to the eventual order may require a permitted non-personal reference. If the merchant cannot make that link reliably, it should report the gap rather than assign all unattributed orders to the agent. A reminder channel is a tool for recovery; it need not define the measurement population.

Use the checkout result, not the conversation, as the record

The UCP specification separates capabilities and makes compatibility part of the interaction. BigCommerce’s notes say stores that support the August release can continue serving the April release, with responses and order webhooks tied to the checkout version. For merchants and integration teams, that means an apparently identical request can yield differently shaped records; events should preserve enough context to be interpreted later. There is no reason to assume a newer protocol changes the checkout configuration of every store or enables every capability automatically.

The commercial boundary is simple. A shopping assistant can indicate that a product looks compatible, but the final SKU, price, stock, shipping and any promotion need to come from the applicable checkout record. Imagine a bundle that can be sent to one address but not split; a recommendation that sounds plausible is not evidence that fulfillment is possible. If a changed condition requires the shopper to review the order directly, track that handoff as a path, not automatically as failure. The most informative holiday-season tests measure paid and fulfilled orders, error reasons, customer support requests and returns by route. A polished conversation is valuable when it helps customers get the right thing. Cart counts alone cannot show that.

Information only. No investment, legal, tax, or financial advice.