Conversational commerce is buying, selling, and customer service that happens through conversation - chat, messaging, or voice - instead of through menus, search bars, and forms. The shopper asks; the system answers, recommends, and acts. The term is a decade old, but the thing it describes only recently started working, because the conversation layer only recently became good enough to be trusted with a catalogue, a policy, and a customer's patience.
Where did conversational commerce come from?
The phrase dates to the mid-2010s messaging boom, when commerce was expected to move into chat apps the way it had moved onto phones. The first generation was keyword bots with decision trees: they could route, collect an order number, and paste a help-center article. What they could not do was hold a real conversation about a real catalogue or execute a real policy, so "conversational" mostly meant "a chat window in the corner." The current generation changed both halves: large language models handle the conversation naturally in any language, and a deterministic decision layer executes the business rules. That combination is what finally separates conversational commerce from the chat widgets it used to be confused with.
What is conversational commerce NOT?
It is not a chat widget with canned replies, and it is not a bot whose job is to keep customers away from your team. A system that answers questions but cannot check an order, apply a return policy, or complete a purchase is a FAQ with typing indicators. And a system measured on how many customers it turned away is a cost project wearing a service costume. Production conversational commerce is measured on resolution (how many customers got their actual matter handled, end to end) and increasingly on revenue per conversation. It is also not the same as agentic commerce, the newer term for AI agents that act autonomously on both the consumer's and the brand's side; conversational commerce is the broader umbrella, agentic is where it is heading.
What are the components of real conversational commerce?
Four layers, in practice:
The conversation layer. A language model that understands intent in the customer's own words and language, without forcing them through button trees.
The decision layer. Deterministic execution of everything with money or policy attached: prices, discounts, return eligibility, age checks. The model talks; it does not decide. In Zowie's architecture this is the Decision Engine.
The integration layer. Order management, CRM, loyalty, carriers. Without it, the conversation can only apologize; with it, the conversation can act.
The oversight layer. Quality evaluation on every conversation rather than a sampled few percent, with a readable record of what the AI did and why. This is what makes the system governable at retail volume.
What does conversational commerce look like in production?
Monos, the Canadian travel and luggage brand, runs roughly 70% of its customer enquiries through automation, with 70% of tickets handled in chat and cost per ticket down 75%. Mike Wu, Monos's Senior Director of Ecommerce and CX, described the starting point bluntly: "We knew we needed automation, but most AI platforms felt like black boxes." The fix was a platform his team could see into and steer.
At European scale, MODIVO (the fashion platform behind eobuwie.pl) runs conversational service in 13 languages across 17 markets, resolving 46% of chats overall and 55% in several markets, with average resolution time down 47%. One platform, one set of rules, every market. That multilingual point is easy to underrate: for a cross-border retailer, conversation is the only service channel that scales into a new language without hiring into it.
Does conversational commerce actually sell?
Yes, and this is the part the last decade's chat widgets never delivered. When the conversation can access the catalogue and the customer's context, it becomes a discovery surface: shoppers learn about products they would never have found through search. Total Wine's published numbers show shoppers who engage the AI agent converting at four times the rate of a traditional session and spending about 20% more when they buy. The scoping is precise (engaged shoppers versus traditional sessions, not sitewide), and the mechanism is covered in depth in how AI agents grow AOV. The practical sequencing in production is service first: the AI agent proves itself on resolution quality, then selling switches on.
How do you evaluate a conversational commerce platform?
Six questions do most of the work: Can it act, or only answer (what integrations run in production)? Who decides about money, the model or a deterministic layer? Can it hold your languages and your catalogue size? How is quality evaluated at volume, every conversation or a sample? Can your team change policies and answers themselves, without an engineering sprint? And can the vendor show published, scoped production numbers rather than a demo? We keep a maintained comparison of the major platforms in the retail platform guide.
The fastest way to understand conversational commerce is to watch one real conversation. See it run at getzowie.com/commerce, or start with the Monos story.
