A Chat upsell rule says: when a shopper shows interest in these products, recommend these other ones, optionally with a discount attached.
The mechanism behind it is worth understanding, because an upsell rule behaves differently from a discount condition or a trigger. Those are gates that either open or do not. An upsell is closer to a strong recommendation handed to the agent at the right moment.
The two "When to offer" settings
After a pick
This watches the shopper's bag inside chat, not their Shopify cart.
A "pick" is an item a shopper has marked as interesting while browsing with Arbyn in the conversation. The field's own help text says it plainly: it fires when one of your triggering products is marked in the shopper's bag.
That is a genuinely earlier moment than adding to cart, and it is the reason this rule is useful. A shopper who has just picked out a snowboard is thinking about the snowboard. Recommending boots then lands better than recommending them at checkout, and it does not require the shopper to have committed to anything yet.
It also means the rule can fire for someone whose Shopify cart is still completely empty. If you are used to thinking of upsells as a cart feature, this one is not.
When asked about
This one watches the product the shopper is currently looking at or asking about, rather than what they have collected. Use it when the recommendation makes sense on interest alone, before someone has picked anything.
It guides the agent, it does not force it
This is the most important thing to know about the feature.
When a rule matches, Arbyn checks everything deterministically: the trigger products, the cap, whether the shopper already has the recommended item, and whether the attached discount is genuinely usable. All of that is decided by rules, not judgement.
What it then does is tell the agent about the recommendation and instruct it to prefer your products over its own ideas. The agent is strongly directed toward your rule. It is not compelled by it.
In practice that means a matching rule will usually produce the recommendation, and occasionally will not, because the conversation had gone somewhere that made a product suggestion inappropriate. A shopper mid-complaint about a late delivery is not going to be sold goggles, and that is the correct outcome even though a rule technically matched.
If you need a guaranteed, unconditional offer, an upsell rule is the wrong tool. Use a Proactive trigger with a discount attached, which presents deterministically, or the Perks tab, which is a list the shopper opens themselves.
Upsells only appear in chat
A Chat upsell reaches the shopper in the chat widget and nowhere else. It is not offered over email, Instagram or Facebook Messenger, regardless of your configuration, because those channels cannot render the product row an upsell is built around.
It is called Chat upsell for that reason. If most of your conversations arrive by email, this feature will look far quieter than the number of matching rules suggests.
More than one rule can match at once
Unlike discount offers and bundles, where exactly one wins, every eligible upsell rule is handed to the agent together.
Priority affects the order they are listed in, so a higher-priority rule is presented first. It does not eliminate the others. The agent may act on one, on several, or on none.
The practical guidance is to keep overlapping rules few. Three rules all matching the same pick gives the agent a menu rather than a recommendation, and a shopper shown eight products has been shown nothing.
Max offers per conversation
The cap counts how many times that specific set of recommended products has been put in front of the shopper in this conversation.
The count is tied to the exact combination of products you chose. Two rules recommending different products each have their own allowance, so the cap limits repetition of one recommendation rather than limiting upselling overall.
It is counted across the whole conversation, not per session or per day, and a new conversation starts fresh.
How the follow-up discount decides someone declined
There is no "no thanks" button on an upsell. It appears as a row of products, not as a question. So a decline has to be inferred, and here is exactly how:
A shopper is treated as having declined when the recommendation has already been shown at least once and the product still is not in their bag.
On that reading, Arbyn switches to the Follow-up discount if declined if you configured one, so the second mention carries a better deal than the first. If you did not set one, the primary discount is used again.
Two things follow from this:
Silence counts as a decline. A shopper who simply moved on to another topic looks the same as one who considered and passed. That is usually fine, since both are people who did not take it.
The follow-up is a genuine second chance, not a nag. It only comes into play within the cap you set, so with the cap at 1 there is no second mention and the follow-up discount never gets used. If you configure a follow-up discount, set the cap to at least 2, or you have configured something that cannot happen.
That last point is the most common misconfiguration on this page.
Attaching a discount
The Discount offer to attach dropdown only lists discounts that are genuinely usable: live in Shopify, unexpired, and set to show in chat. If it is empty, the page tells you why, and the fix is to enable Show in: Chat on a discount offer first.
This means you cannot accidentally attach a dead or hidden code. Anything you can pick will work.
Good to know
Recommend up to 5 products per rule. Fewer is usually better; a row of two or three reads as a suggestion, five reads as a catalogue.
The rule name is internal. Shoppers never see it, so name it for your future self.
A shopper who already has the recommended product is skipped, so a rule will not recommend something already in their bag.
The trigger products and the recommended products should not overlap. A rule that recommends the thing that triggered it does nothing useful.
FAQ
Does "After a pick" mean added to cart? No. It means marked in the shopper's bag inside chat, which happens earlier and does not require a Shopify cart at all.
A rule matched and no recommendation appeared. Is it broken? Probably not. The rule directs the agent strongly but does not compel it, and there are moments in a conversation where a recommendation would be wrong. If it never appears across many matching conversations, check that the conversation is in chat rather than email, and that the cap has not been used up.
Why do I see no upsells on email conversations? Chat upsell is chat-only. Email, Instagram and Messenger cannot render the product row it uses.
Can two upsell rules fire in one conversation? Yes. All eligible rules are offered to the agent together, with priority setting the order rather than picking a single winner.
My follow-up discount never gets used. Why? Almost certainly because Max offers per conversation is 1, so there is never a second mention for the follow-up to appear on. Set it to 2 or more.
How does Arbyn know the shopper declined? It infers it: the recommendation was shown at least once and the product is still not in their bag. There is no decline button on an upsell row.