Pairings that learn what sells together.
AI-written pairings cover every product from day one. Real orders take over as they come in, and shopper behavior keeps them sharp.
Where every pairing comes from.
Seven sources, in priority order. What you set always comes first.
While a store has few orders, AI pairings lead. As orders build up, what’s really bought together takes over. And “goes well with” never suggests another product of the same type.
- 1Your pinsSet in Valence. They survive every rebuild.
- 2Search & Discovery pinsImported from Shopify’s app automatically.
- 3Shopify’s recommendationsShopify’s own suggestions, as a starting point.
- 4Bought togetherPaid orders from the last 90 days.
- 5AI pairingsUp to eight per product; low-confidence pairs are dropped.
- 6InheritedNew products borrow from similar ones.
- 7Similar itemsFrom product details and what shoppers browse together.
Watch it build.
Rebuild any time and watch each step. The first build takes a few minutes; after that, changes flow in on their own.
- Applying your pins
- Reading product details
- Finding similar products
- Finding products that go together
- Learning from recent orders
How it learns.
Every day, what shoppers did re-weighs the pairs they touched.
- Clicks, add-to-carts and purchases each make a pair stronger.
- Views and items carted together count too, as weaker signals.
- A dismissed suggestion gets weaker, and so does a pair shown often without engagement.
- A pair that grows too weak is dropped, and stays out of AI pairings for 30 days. A weekly audit reviews the weakest.
Shopper activity comes from a Shopify web pixel that runs only with analytics consent and sends no customer identifiers.
- AI pairing written
- Shopper clicked itstronger
- Added to cart togetherstronger
- Bought togetherstronger
- Shown often, never clickedweaker
How it stays current.
Your store changes all day. Shopify tells Valence as it happens, and only what changed is redone.
An order is paid
What was bought together gets stronger.
A product’s details change
It’s re-read and re-paired. Price and stock edits don’t trigger a rewrite.
A product sells out
It drops out of suggestions until it’s back.
A new product arrives
It borrows pairings from similar products, then earns its own.
Every week
A full refresh of the graph.
Everywhere a shopper decides.
On product pages
A theme block: your heading, complementary or related, two to eight products. Shoppers can dismiss one.
In the chat
When a shopper asks what goes with something, and in the complete-the-look skill.
Right after add to cart
The assistant follows an add with what goes with it.
From triggers
Recommend pairings on the pages and moments you choose.
- AddedMerino trail socks$18
You stay in control.
Every pairing is yours to keep, change or leave out.
Pin
Pin a suggestion and it survives every rebuild. Pins set in Search & Discovery show as pinned too.
Exclude
Leave out a product, a whole product type like gift cards, or a vendor.
Sold out, hidden
Sold-out products drop out of suggestions on their own.
See every pairing
A searchable table of every product, with what goes well with it and what’s similar.
Rebuild
Rebuild any time and watch each step. The first build takes a few minutes.
Clear alerts
Told plainly when AI pairings pause, order data is missing, or shopper activity isn’t being recorded.
Frequently asked questions.
My store is new. Will it work?
Yes. AI-written pairings cover every product from the start. As paid orders come in, bought-together data takes over.
Can I choose pairings myself?
Pin any suggestion and it stays through every rebuild. Exclude products, product types or vendors you never want shown.
Does it use Shopify Search & Discovery?
Yes. Complementary and related pins you set there are imported automatically and shown as pinned in Valence.
How often do pairings update?
Continuously. Product and order changes update the graph as they happen, shopper behavior adjusts it, and a full refresh runs every week.
Does it track my shoppers?
A Shopify web pixel records views and add-to-carts, only with analytics consent, and sends no customer identifiers.