Is your chat selling, or just answering?
Most stores already have a chat widget. Very few have anything that sells. What separates them is what happens in the ten seconds after the shopper's question gets answered.
A shopper lands on a product page for a $180 jacket. They type: is this warm enough for winter?
One version of the answer: "Yes, this jacket is rated for temperatures down to -5°C and features an insulated lining. Let me know if you have any other questions."
The shopper reads it, thinks okay, and closes the tab.
The other version: "Down to about -5°C, so it's good for most of winter. Where are you, and is this for commuting or being outside for hours? If you're regularly below freezing for long stretches, the Alta is the same cut with a heavier fill, and it's $30 more."
The shopper types: Chicago, mostly walking the dog.
"Then the Alta. Chicago winters will get past this one on the cold days, and you'll be out in it daily. Want me to put the Alta in your cart? What size are you in the one you're looking at?"
Both threads answered the question correctly. One of them made a sale and raised the order value. Accuracy did not separate them, and neither did response time or the quality of the language model. The second one had a goal.
Support tools and sales tools are different products
This distinction gets blurred constantly, because both of them are a box in the corner of your screen that types words. They were built to do opposite jobs, and they are measured on opposite outcomes.
A support tool exists to close tickets. Its success metric is deflection, meaning how many questions got resolved without a human. It answers, then stops, because stopping is the goal. Every additional exchange is a cost.
A sales tool exists to close orders. Its success metric is what ended up in the cart. It answers, then does the thing a good salesperson does next: asks the question that narrows the choice, names a recommendation instead of listing options, and offers to take the action. Every additional exchange is an opportunity.
A support agent hands over information and a sales agent produces a decision, which is why one of them can finish a conversation by explaining how to order and the other finishes it by putting the thing in the cart.
That is why "we already have chat installed" is usually not the answer people think it is. The honest test is not whether you have a widget. It is whether, when a visitor hesitates on a product page at eleven at night, anything on your site recognizes the hesitation and does something about it. If the box only opens when someone clicks it, and only responds when someone types, it is behaving like a receptionist rather than a salesperson.
The four things that separate them
The gap shows up in four specific places, and you can check all of them on your own store in about five minutes.
The first is whether it acts or only advises. "You may also like" is a shelf, not a recommendation. A salesperson says get this one, and here is why for you specifically, and then carries it to the till. If your chat cannot put the correct variant in the cart, apply an approved discount, and hand the shopper to checkout, then the shopper has to do all the work of buying immediately after being told what to buy, and a meaningful share of them won't.
Second, does it ever start the conversation? The shoppers most worth talking to are the least likely to open a chat window. They're the ones comparing two products for the fourth time, or hovering near the back button with a full cart. A reactive tool never meets them. Something has to notice the behaviour and open with one specific, relevant line. A generic "Can I help you?" does not count. Everyone has learned to dismiss that without reading it.
Third, does it know the actual catalog, or a scraped snapshot from last Tuesday? Live prices, live stock, the real policy, the actual variants. A tool that confidently recommends something out of stock does more damage than one that says nothing, because now the shopper doesn't trust anything else it said either.
The last one is language. "Supports multiple languages" as a settings toggle is a different thing from actually replying in the language the shopper typed in, with the prices and currency of their market. If you sell across borders, this is either working on every message or it isn't working.
"But they can just message us on WhatsApp"
This comes up constantly and it deserves a real answer rather than a dismissal, because WhatsApp genuinely does work for a lot of merchants.
The problem is where it puts the conversation. A shopper is standing in your store, holding the product, ready to ask a question. Sending them to WhatsApp tells them to step outside, walk to a different building, and talk there. Then, once they've decided, they have to come back, or you build the order manually as a draft, or you send them a link and hope they follow it. Every one of those steps is a moment where a distraction can win.
The second problem is quieter and more expensive. Those conversations are trapped. A thousand WhatsApp threads contain a detailed record of what your customers want, what confused them, what they nearly bought, and what stopped them, and none of it is aggregated, counted, trended, or connected to whether an order actually happened. You are generating very good data about your own demand and storing it in a format nobody can read at scale.
An agent that lives on the product page has neither problem. The question gets answered where the product is. The cart action happens in the same breath as the recommendation. Every conversation is structured data from the moment it happens.
WhatsApp is a fine place to talk to a customer who already bought. It is a poor place to sell to one who hasn't.
The five-minute audit
Open your own store in an incognito window and behave like a confused customer.
Ask something a real shopper would ask. Not "what's your return policy," which any tool handles, but something that requires judgement. I'm between these two, which one for a beginner? Watch what happens. Does it pick one? Does it explain why for your situation? Does it offer to add it?
Then hesitate. Put something in the cart and sit there for ninety seconds without clicking anything, and see whether anything happens.
Then ask in another language. Then ask about something you don't stock.
Whatever your current tool does in those four moments is what it does for every visitor you're paying to send there. If it answered all four well, you're in better shape than most stores and you should keep what you have. If it answered politely and then waited, you don't have a sales problem you can fix with a better prompt. You have a shop with nobody on the floor and a very responsive information desk.