EliteAI™ Ultra already picks great products to recommend out of the box, with no setup needed. But if you want tighter control over exactly which products it can suggest and how they rank, you can create your own fine-tuned versions of it.
This is an advanced, optional feature. It's marked with a Beta badge, and you never have to use it - the two built-in versions of EliteAI™ Ultra work perfectly on their own. Fine-tuning is for merchants who want to shape recommendations around their own rules.
Every fine-tuned version starts from one of two built-in engines:
EliteAI™ Ultra Original - recommends products customers most often buy together with the item in the cart. Best for accessories, refills and natural add-ons.
EliteAI™ Ultra CrossCategoryBoost - recommends complementary products from product types other than the cart item. Best for broadening the basket and product discovery.
Info: Because CrossCategoryBoost only ever pairs different product types, "same product type" rules aren't available when you build on it. If you need same-type rules, start from Original instead.
A few common reasons merchants create their own version:
Only ever recommend products under a certain price.
Prefer cheaper add-ons (or, for upselling, prefer pricier ones).
Keep recommendations inside a specific collection.
Recommend premium products only when the cart item is itself premium.
If none of these apply to you, the built-in versions are a great choice and you can stop here.
In the left sidebar, click Upsells.
Open the AI upsells tab.
Find the Fine-tune EliteAI™ Ultra card (marked Beta).
Under Pre-trained versions, you'll see the two engines. Click Create fine-tuned version on the one you want to build from.
On the editor page, give it a name in the Strategy name (internal) field (for example, "Premium add-ons under $50"). This name is internal only - your shoppers never see it.
Add your rules. You can shape recommendations two ways:
Filters decide which products are even eligible.
Boosts nudge eligible products up or down the ranking.
Click Save.
For the full rundown of filters and boosts, see the companion article below.
Tip: Filters and Boosts for Fine-Tuned Engines walks through every rule type with examples.
When you save, EliteCart needs to teach your engine the new rules. The Training status badge at the top of the editor shows where that's at:
Preparing → Analysing orders → Analysing products → Finishing up → Trained
Training typically takes 5 to 60 minutes and runs in the background. Until it reads Trained, your version keeps serving its previous recommendations (or, for a brand-new version, the fallback set below).
Creating a version doesn't change anything for your shoppers on its own. You have to assign it as the recommendation source somewhere.
On your storefront:
On the AI upsells tab, scroll to Customize by display location.
On the location you want (Cart, Two-step cart, or Product page on plans that include it), turn on Show EliteAI™ recommendations when no manual flow matches.
Open the Choose your engine dropdown. Your fine-tuned versions appear at the bottom, labelled "Your name (custom)".
Pick yours and click Save.
At checkout:
Open a checkout module and go to its Upsell setup tab.
In the Strategy picker, choose EliteAI™ Ultra or Hybrid: Flows + AI so the AI recommendations appear.
In the Recommendation source dropdown, pick your version (also shown as "Your name (custom)").
Click Save.
Each location and each checkout module is independent, so you can run one version on the cart and a built-in engine everywhere else.
You can have up to 3 fine-tuned versions live at once. To make room for a new one, retire an old one:
Open the version you want to retire.
Click Archive strategy and confirm.
Info: A version that's still assigned to a location or checkout module can't be archived. Point that location at a different engine first, then archive.
