Shopify · Conversational commerce

The only storefront surface
that answers back.

A product grid is a broadcast: it shows the same thing to everyone and learns nothing. A conversation moves value in both directions — the shopper gets an answer, and you finally hear the question.

A pollination exchange: a bee at an opened bloom with pollen travelling in both directions
The asymmetry

Grids record clicks. Conversations record questions.

Analytics tells you what shoppers selected from what you already decided to show them. It cannot tell you what they came in wanting and failed to find, because a grid gives them no way to say it. That gap is where most catalog decisions should come from, and almost no store collects it.

What a grid can tell you

Which of your existing products got attention, in which order, at which price. Useful — and entirely bounded by choices you already made. A grid can never report demand for something you do not stock.

What a conversation adds

The unanswered ask. Fifty shoppers a month requesting a size you do not carry is a buying decision, and it is invisible to every other surface in your store.

Be honest

Where conversational commerce does not belong

It is a fit for a specific shape of decision, not an upgrade every store should take. The fastest way to discredit it is to bolt it onto a catalog that never needed it.

Skip it — narrow catalogs

Three products and good search? A conversation inserts a step between a shopper and a thing they could already see. You will measure that as friction, correctly.

Skip it — pure replenishment

Someone rebuying the same coffee wants one click, not a dialogue. Reorder flows beat conversation every time for a decision already made.

Use it — wide or technical catalogs

Hundreds of SKUs where fit, compatibility or occasion decides the purchase, and the shopper lacks the vocabulary your filters assume.

Use it — considered and gift purchases

High deliberation, low expertise. The shopper knows the constraint but not the product, which is precisely what a filter rail cannot accept as input.

The failure mode

A wrong answer costs more than a lost sale

This is the part most write-ups leave out. A grid cannot lie about a fabric composition. A generative surface can, and the bill arrives as a return, a refund and a review — not as a conversion metric you would notice.

Bound the answers

Responses come from the ingested catalog — your variants, prices, stock and copy. Not from a general model's impression of what a product like yours is usually like.

Prefer refusal

"I cannot confirm that — here is a human" is a good outcome. Confident invention is the only genuinely expensive one.

Escalate with context

Orders, returns and account questions hand to a person with the thread attached. The customer should never have to start again.

Both
directions — the shopper asks, you learn
Catalog
bounded answers, refusal over invention
Human
handoff with the conversation attached
Questions

Conversational commerce, answered

01What is conversational commerce?+
Selling through a dialogue rather than a layout. Instead of a shopper navigating categories and filters to find a product, they describe what they need and the store narrows it. The interface is a conversation; the catalog, checkout and fulfilment stay exactly as they were.
02Does it actually convert better than a normal storefront?+
Not universally, and anyone claiming otherwise is selling something. It converts better where the shopper cannot self-serve: wide catalogs, considered purchases, gifting, technical fit. On a narrow catalog with strong search and three SKUs, a conversation adds a step and makes things worse. Match it to the decision, not to the trend.
03What does a conversation give a merchant that a grid does not?+
The question. A grid records what was clicked; a conversation records what was asked and not answered. That is the most valuable merchandising data most stores never collect — the gap between what shoppers arrive wanting and what the catalog carries.
04Is this just a chatbot with better marketing?+
The difference is whether unscripted questions are handled. A scripted flow answers what someone anticipated and collapses on everything else. A store agent reasons over your actual catalog, so it can answer questions nobody wrote down — and refuse the ones it cannot support.
05What is the real risk?+
A confidently wrong answer. A grid cannot misstate a fabric composition or invent a delivery date; a generative surface can, and the cost lands as returns, refunds and trust rather than as a bad conversion number. Bounding answers to ingested catalog data and escalating everything else is not a limitation — it is the whole design.
06Where should a store start?+
With the questions your support inbox already answers repeatedly before purchase. Those are proof of demand for a conversation, they are bounded, and they are checkable. Start where you can verify the agent is right, then widen.

Start where you can check the answers.

The pre-purchase questions your inbox already handles are bounded, provable, and proof the demand exists. Begin there, then widen.