Conversational commerce is buying and selling through a conversation rather than a catalogue: the customer describes what they need in chat, messaging or voice, and the seller, human or AI, helps them choose and complete the purchase inside that exchange. The definition is simple. The practice mostly is not, because most "conversational" commerce never asks the customer a single question, and without a question there is no conversation. There is a search box with manners.
Key facts
- Definition: commerce where the selection and purchase happen inside a dialogue (web chat, WhatsApp, Messenger, SMS, voice assistants)
- Channels: on-site chat, messaging apps, voice, and increasingly AI shopping assistants inside search engines and marketplaces
- The promised benefits: fewer abandoned choices, higher order value, 24/7 availability, lower support cost
- The missing piece: a conversation that only answers is retrieval; a conversation that asks is sales
- Evidence: in 238 conversations with a store's AI chat, the chat asked first 7 times and listed products 248 times
Conversational commerce, defined plainly
The term was coined around 2015 to describe shopping inside messaging apps. Today it covers any sale where the dialogue does the work a product grid used to do:
- A customer on a store's website types "need a quiet bathroom fan" and buys one without leaving the chat.
- A customer on WhatsApp sends a photo of a broken part and receives a replacement link.
- A customer asks a voice assistant to reorder what they bought in March.
- A shopper asks an AI shopping assistant inside a search engine to compare three models.
What these have in common is that the customer never browses. They say what they want, and the seller's side has to understand, narrow and recommend. That last step is where conversational commerce either earns its name or becomes a chatbot that lists products.
Why conversational commerce exists at all
A product grid assumes the customer knows how to filter. In most categories with any technical content, they do not. Someone buying a shower drain does not know whether their tray has a 50 or 90 mm outlet. Someone buying a toilet seat does not know hinge spacing is the thing that matters. A grid with filters cannot help them, because they do not know which filter to touch.
A conversation can. It can ask. That is the whole promise of conversational commerce: the seller takes over the narrowing that the customer cannot do alone, the way a shop assistant does on the floor.
The benefits of conversational commerce, honestly
Vendor pages list the benefits of conversational commerce as higher conversion, higher average order value, round-the-clock availability, lower support cost and more data about intent. All are achievable. None is automatic. Each depends on whether the conversation actually narrows the choice:
- Higher conversion happens when the customer ends the dialogue knowing which product is right. It does not happen when they end it with a list of nine.
- Higher order value happens when the seller asks about the installation and discovers a second part is needed. It does not happen when the chat answers only the question asked.
- 24/7 availability is worth something when the night-time conversation ends with a captured contact. It is worth nothing when it ends with "call us during working hours".
- Intent data is only produced by questions. A chat that never asks collects nothing but search terms.
The part most conversational commerce platforms skip
We exported 238 conversations from a well-rated AI chat in an online store before replacing it. The chat was fast, polite and knew the catalogue. In 379 customer inputs it asked a question first 7 times. In 248 it returned a priced product list. It produced 108 clicks, two carts and no attributed purchases in four months. The full account is in the case study.
Every conversational commerce platform can do that. Retrieval with a friendly tone is the default output of a language model attached to a catalogue. What none of them do by default is the shop assistant's first move: a question that reveals why the customer is here.
In the book Diagnostic Sales™ I describe that first move as the start of a diagnosis. The customer presents a symptom, "trap", "fan", "toilet seat". The seller does not prescribe from the symptom. They ask what it is for, where it goes, what went wrong with the last one. Only then do they recommend, and they recommend two things, not ten. The full structure is on the methodology page.
Conversational commerce without that move is a catalogue you can talk to. With it, it is a salesperson who never sleeps.
What a real conversation in commerce looks like
Three rules turned the chat in our store from the first kind into the second:
- One question before the list. "Trap" gets "For a washbasin, kitchen sink, shower tray or washing machine?" before a single product appears.
- No closest match. If the exact item is not in stock, the chat says so and asks about purpose, instead of offering a similar-sounding product.
- Take the contact. When the chat cannot confirm a detail or the customer is ready to buy, it asks for a number and promises a call. At night, always.
None of these needs a new platform. They need someone to write down how the best salesperson in the company runs a conversation and to give that to the chat. The seven practices in full are in AI chatbot for ecommerce: best practices.
Choosing a conversational commerce platform
The platform matters less than the vendors say. Before comparing features, check three things:
- Can you instruct it to ask a qualifying question before showing products, per product group?
- Can you forbid substitution when an exact item is missing?
- Can it capture a contact and hand it to a person, rather than only sending the customer to a phone number?
If the answer to any of these is no, the platform will do conversational retrieval, which is useful for support and does not sell. If the answers are yes, the remaining work is yours: writing the sales knowledge the chat will follow. How that knowledge moves from a salesperson's head into a system is described on the AI integration page.