Three Upsells, One Purchase
Multi-step funnel sequencing is live for buyers. A single offer can carry up to three order bumps on the checkout page, then a post-purchase chain of up to three one-click steps, shaped upsell → downsell → upsell. The downsell only appears if the upsell before it was declined. One-click charge authority lasts 30 minutes from the main purchase; after that the remaining steps are skipped and the buyer lands on the receipt. You build the chain by asking your AI. The dashboard shows it read-only.
A buyer who has just paid you is a different person from the one who landed on your page ten minutes ago. They have already decided you are worth money. That state does not last long, and everything in this post is about what you are now allowed to do inside it.
Multi-step sequencing is live. Here is exactly what a buyer can walk through, exactly how long they have, and exactly how you build it — including the part where the way you build it is not what you expect.
Three bumps, before the payment
A single offer can carry up to three order bumps on the checkout page. They are tick-boxes. The buyer ticks them before they pay, so they ride the same purchase — one card entry, one payment.
A cart offer cannot carry bumps. It can carry a funnel, but not bumps, and that is deliberate rather than missing: a cart already asks the buyer to choose between several products on the page, so bolting tick-boxes onto tick-boxes just makes the page harder to read. If you want the "several things, one page" shape, I wrote about that separately in Sell Several Things On One Page.
Then up to three one-click steps
After the payment goes through, the buyer can be shown a chain of up to three one-click steps, shaped upsell → downsell → upsell. One click accepts. There is no second card entry, because the charge rides the authority from the purchase they just made.
The part worth understanding properly is the middle step. The downsell only appears if the upsell before it was declined. It is not a step in a queue that everybody sees. It is a branch, and it is the branch that gets taken when the answer to step one was no.
That conditional is the whole reason to bother with three steps rather than one. A downsell shown to somebody who already said yes is not a downsell, it is an insult with a price on it — you have just told a paying customer they could have had something cheaper. Because the step is gated on the decline, you can write it as what it actually is: a smaller version of the thing they just turned down, aimed only at the people who turned it down.
Thirty minutes, then it stops
One-click charge authority expires 30 minutes after the main purchase. That is not a soft limit or a nudge. When it is gone, the remaining steps in the chain are skipped and the buyer lands on the receipt.
Read that as a design constraint rather than a limitation to work around. It bounds how long a charge can land on a card without the buyer being on the page, which is the correct behaviour for a one-click charge. Nobody should be able to bill a card off the back of a decision made an hour ago in a tab that got left open.
It does have a practical consequence for how you write the steps: they have to be answerable quickly. Put a four-minute video in front of the button on step one and you have spent four of your thirty minutes before anybody can say yes, on a page the buyer did not ask to see. Short, specific and skippable beats persuasive here, because the buyer already bought — you are not re-selling them, you are offering them one more thing.
How you actually build one
This is the part I do not want to be vague about, because vagueness here is how software ends up with documentation describing screens that do not exist.
There is no offer editor for this. Two levels of dashboard editing for funnel chains sit behind flags, and both are off — for every account, including mine. The dashboard shows you the chain on an offer, and it shows it read-only. It is a window, not a workbench.
You build the chain by asking your AI. The bumps and the funnel are fields on the offer — bumps[] and funnel[] — and your agent writes them over MCP or the API when you describe what you want:
I know how that sounds if you came here from a tool with a drag-and-drop funnel canvas. But this is not a workaround for a builder we have not finished. It is the shape of the product. GoCushy is AI-first: the place you describe a funnel is a sentence, and the agent is the thing that turns a sentence into a live checkout. The dashboard's job is to show you what got built, so you can check it against what you meant.
Given that, I would rather ship the buyer-facing half now with an honest read-only view than put a builder in front of you that quietly writes something different from what the checkout runs. That failure mode is expensive and it is silent, and the way you find out is a buyer's email.
Three and three is a ceiling, not a target
Six places to add money to one purchase, if you use all of them. The maximum is not the recommendation, and I want to be blunt about that because the number is the most quotable thing in this post and the easiest thing to misuse.
A buyer who has to say no four times to reach their receipt has learned something about you, and it is not that you have a great catalogue. That feeling has a cost, and the cost shows up as refunds, chargebacks and the emails you get instead of the second purchase. A chain that grinds every buyer is how you earn refunds — not a risk of it, a method for it.
What I would actually do: one bump and one upsell. Both genuinely useful to somebody who just bought the main thing. Run it, live with it for a few weeks, and see whether refunds move before you reach for the downsell. Add the third step only when you have something specific to say to the people who declined — not because there is a slot for it.
Where this is the wrong choice
Four cases where I would not use it, and I would rather say so than have you find out.
- You want to click a funnel together yourself. Today you cannot. Editing is off for every account and the dashboard is read-only. If you are not going to describe your funnel to an AI, this feature is not available to you in a form you will enjoy.
- Your follow-on is a considered purchase. A 30-minute one-click window is the right shape for something a buyer can decide about in fifteen seconds. If they need to look at the calendar, talk to a partner or get a budget signed off, the chain is the wrong container — sell it in an email a week later instead.
- The buyer is choosing between things, not adding to one. That is a cart offer, and a cart takes a funnel but not bumps. The cart post covers which shape fits which sale.
- You do not have three things worth offering. This one gets skipped a lot. Three steps with nothing real in steps two and three is worse than one step, because the first weak offer discounts the strong one you led with.
What this changes
Before this, one purchase could carry one extra ask. Now it can carry a sequence with a real branch in it — and the branch is the interesting part, because it is what lets you talk differently to the people who said no. That has been standard in expensive funnel software for years. Here it is a sentence to your agent, on your own Stripe, on a checkout you did not have to lay out.
If you build one, keep it short. The window is thirty minutes and your buyer's goodwill is shorter.
