What conversational commerce is
The term conversational commerce was coined by Chris Messina in 2015, describing sales through
messengers and chats. Since then the technology has travelled from simple FAQ bots to LLM-based
systems that can hold a full conversation about a product.
The core idea is replacing a filtered catalogue with a conversation: the shopper describes the
need in words (“I’m looking for wireless headphones for sport, under $120”), and the system finds
and recommends.
How conversational formats have evolved
| Generation | Technology | What it can do | Limits |
|---|---|---|---|
| 1.0 | Scripted bots | FAQ, order tracking | Only the scenarios someone wrote |
| 2.0 | NLP + intent classification | Understands phrasing, basic matching | Needs labelled training data |
| 3.0 | LLM + RAG | Free-form dialogue, accurate matching | Inference cost, hallucination |
| 4.0 | Agentic commerce | Autonomous purchases | Needs protocols and trust |
Conversation as a competitive advantage
A classic catalogue expects the shopper to already know what they are looking for and to be
comfortable with filters. A conversational interface lowers that barrier: the shopper states the
need the way they would to a person, and the system takes on the searching and comparing.
That matters most in categories where the choice is genuinely hard — electronics, beauty, sports
nutrition — and for gift purchases, where the buyer is not the end user.
Tip: conversational commerce does not replace the familiar catalogue — it widens the set of
entry points. The natural place for a conversational interface is next to search, or as an
alternative path on the homepage.
Conversational commerce and agentic commerce
Conversational commerce assumes a human in the conversation: the shopper takes an active part
and makes the final call. Agentic commerce is the next step, where an agent acts autonomously and
completes the purchase without confirming each action.
For retailers that means different integration requirements: a conversational assistant slots into
the existing site, while an agent needs dedicated protocols and authorization mechanics.