Our own product · 2026
UpAfter AI
A post-purchase upsell app that uses an AI model to choose the offer, write the copy and explain its own reasoning.
The problem
Every upsell app on the App Store makes the merchant build the rules: if the customer bought X, offer Y at Z percent off. Merchants set these up once, never revisit them, and the offers slowly stop matching the catalogue. The work that makes upsells effective is exactly the work merchants don't have time for.
The approach
Move the decision to the model
On the order webhook, the app sends catalogue and order context to Claude, which picks the product, writes the offer copy, sets a discount and returns its reasoning. The merchant configures nothing.
Compute early, render instantly
Recommendations are generated at webhook time and cached, so the thank-you page reads from the database. An AI call in the page render would have put a visible delay in front of the offer.
Fall back rather than fail
When model confidence is low, a rule-based engine takes over. An upsell that's merely adequate beats an empty slot.
Four surfaces from one engine
Thank-you page, product page, cart page and cart drawer all draw on the same recommendation logic through separate extension surfaces.
What shipped
- Four upsell surfaces built on a shared recommendation engine
- Billing API wired with four plan tiers and usage metering
- Deployed to Railway, currently in private beta ahead of App Store submission
Got something similar?
If any of this looks like the problem in front of you, send us a paragraph about it.
Email us