NEW: Fine-Tune EliteAI™ Ultra
EliteAI™ Ultra already picks great products to recommend out of the box. Now you can build your own fine-tuned versions of it to control exactly which products it suggests and how they rank.
The Update
Until now, EliteAI™ Ultra gave you two pre-trained engines and recommended whatever the model judged best. That works well for most stores. But some merchants want tighter control - only recommend products under a certain price, keep suggestions inside one collection, or push pricier complements for upselling. Fine-tuning lets you shape recommendations around your own rules without leaving the EliteAI™ engine behind. It is an advanced, optional feature, and you never have to use it.
What's New
Start from a pre-trained engine: Every fine-tuned version builds on one of the two base engines - EliteAI™ Ultra Original (products customers most often buy together, best for accessories and add-ons) or EliteAI™ Ultra CrossCategoryBoost (complements from other product types, best for broadening the basket).
Filters decide what is eligible: Hard rules that remove products that do not qualify. Filter by price, tag, vendor, product type, collection, or title - for example, only recommend products under $50 from your Accessories collection.
Boosts nudge the ranking: Soft rules that raise or lower a product's rank without removing it, set on a simple five-way scale from Strongly dampen to Strongly boost. Prefer your bestsellers, same-vendor items, or cheaper add-ons.
If/Then conditions: Apply a filter only when the cart item matches - for example, only recommend premium products when the product already in the cart is itself tagged premium.
Assign per location: Each saved version becomes a selectable recommendation source. Run one version on the cart and a different engine on the product page or inside a checkout module.
Impact on Your Store
Recommendations stop being one-size-fits-all. You can keep margins healthy by capping recommended prices, protect brand fit by staying inside chosen collections, and tailor the strategy per surface. You can have up to three live versions at once.
Configuration
Upsells → AI upsells → Fine-tune EliteAI™ Ultra. Pick a base engine, click Create fine-tuned version, add your filters and boosts, and save. Each save queues a training run that typically takes 5 to 60 minutes. Creating a version does not change anything for shoppers until you assign it as the recommendation source on a surface or checkout module.
Why This Matters
The smartest recommendation engine still works best when it knows your business rules. Fine-tuning hands you that control while keeping all the learning EliteAI™ Ultra does from your orders and catalog.
For detailed setup, see our Fine-Tuning EliteAI™ Ultra help article, and the Filters and Boosts reference for every rule type with examples.