When you build a bundle you choose a Strategy: Fixed, or Collection-based.
Fixed is exactly what it sounds like. You pick between 2 and 5 products and those are the products, every time, for every shopper.
Collection-based works quite differently, and the difference is worth understanding before you rely on it, because the products in the bundle do not exist until a shopper accepts it.
What collection-based actually does
You give it collections to compose from. At the moment a shopper accepts the bundle, Arbyn:
Reads up to 25 products from each collection you nominated.
Discards anything not currently available for sale.
Removes duplicates where the same product sits in two of your collections.
Shuffles what remains, at random.
Takes the first few from the shuffled list.
How many is "a few"? Three, always. There is no field on the bundle builder that lets you change this for a Collection-based bundle today, the draw size is fixed at three regardless of how many collections you nominate. (Internally the code that composes the draw supports a target between 2 and 5, but nothing in the builder's Collection-based form currently sets it to anything other than three, so treat three as the real number rather than a configurable default.)
The three consequences, plainly
It is genuinely random, and it is different every time. Two shoppers accepting the same bundle five minutes apart get different products. The same shopper accepting it twice in one conversation gets a different draw the second time. There is no stored set, no per-shopper consistency, and no way to see in advance what any given shopper will receive.
Arbyn itself does not know what is in the bundle before offering it. When the agent mentions the bundle, the products have not been chosen yet. It can name the bundle and describe the deal, and it cannot tell a shopper "this one contains X and Y", because at that moment nothing does.
Out-of-stock products are removed at the point of acceptance, not at the point of offering. Availability is checked when the shopper takes the bundle, so a bundle whose collections have gone thin can still be offered and then quietly assemble from a smaller pool.
When to use each strategy
Use Fixed when the combination is the point. A curated starter kit, a matching set, a "these three go together" recommendation, anything where the specific products are the reason someone wants it. Fixed is also the only choice if you want to describe the contents in your marketing, since you actually know what they are.
Use Collection-based when the category is the point and any few items from it constitute a good offer: a mystery box, "three from our sale rail", a discovery bundle for a large catalogue. It has one genuine advantage over Fixed, which is that it never goes stale. Add products to the collection and the bundle starts including them without you touching the bundle at all.
Do not use Collection-based if a shopper might reasonably expect two accepts to produce the same thing, or if your collections are small enough that the random draw could produce an odd combination. With a collection of four products and a target of three, you are shipping nearly the whole collection every time and the randomness buys you nothing.
What happens when a product is unavailable
Arbyn removes anything not currently available for sale before it draws, so an out-of-stock product simply does not enter the pool.
If the whole pool empties out, meaning nothing across your nominated collections is available, the bundle cannot be assembled and the shopper is told rather than being handed a broken offer.
On a Fixed bundle the behaviour is similar but more visible: unavailable products are skipped, the rest are added, and the shopper is told plainly which ones were left out and why. Arbyn does not silently hand over a smaller bundle and hope nobody notices.
A gap worth knowing about: availability is not checked at the moment the bundle is offered, only when it is accepted. So a bundle whose products have all sold out can still be mentioned in conversation, with the disappointment arriving a moment later. If a bundle matters commercially, check its stock rather than assuming an offer implies availability.
Keeping a collection-based bundle healthy
Give it enough to draw from. A pool of at least three or four times your target count keeps the draw meaningful. Below that, use Fixed.
Nominate more than one collection if you can. Duplicates across collections are removed, so overlapping collections cost you nothing and widen the pool.
Check availability periodically. The bundle will keep offering itself as long as it is active, regardless of how thin the collections have become.
Write the opening message for a category, not for products. Since you cannot know which products a shopper will get, copy such as "three picks from our winter range" survives the randomness; naming products does not.
The discount works the same either way
Whichever strategy you choose, the discount behaves identically, and it is a real Shopify discount rather than an honour-system promise. That is covered fully in the article on how a bundle's discount actually works. Two points that matter here:
The discount is scoped to the bundle's own products or collections, never store-wide.
On a collection-based bundle the discount is scoped to the collections you nominated, so it stays correct no matter which products the draw produces.
Good to know
The draw size is fixed at 3 for a Collection-based bundle. The builder has no field to change it, so unlike Fixed bundles (2 to 5 products, your choice), the count on this strategy is not currently something you set.
A collection-based bundle shows in the agent's own view as "composed at runtime", which is why it will describe the bundle by name and by deal rather than by contents.
All the other bundle settings behave identically across strategies: Show in, When to offer conditions, Cap, Priority and Status do not care which strategy you picked.
FAQ
Can I see which products a shopper actually received? The conversation shows what was added to their cart, so you can see it after the fact. There is no way to preview a draw in advance, because the draw does not happen until acceptance.
Will two shoppers get the same bundle? Almost certainly not, unless your collections are small. The selection is shuffled fresh for each acceptance.
If a shopper accepts twice, do they get the same products? No. Each acceptance draws again, independently.
Does it ever include an out-of-stock product? No. Availability is filtered before the draw. The gap is that stock is checked when the bundle is accepted rather than when it is offered.
Can I control which products are more likely to be picked? Not within one bundle, since the draw is uniform. Build a Fixed bundle for the products you want to push, or nominate a tighter collection.
Does adding a product to my collection update the bundle? Yes, immediately, with no edit needed on the bundle. That is the main advantage of this strategy.