Recommendation algorithms work best when they have data to work with. For known returning visitors with browsing or purchase history, Personyze can produce highly personalized recommendations. But what about new subscribers, anonymous traffic, or visitors whose interests don’t yet match any of your products with statistical significance? Without a fallback, those visitors would see empty or arbitrary recommendations — which looks broken.
Fallbacks solve this by defining a secondary algorithm that runs whenever the primary algorithm can’t produce meaningful results.
Setting a fallback
In the algorithm step of any recommendation wizard, look for the Fallbacks area next to the filters — there’s a small + icon. Clicking it opens an additional algorithm picker, identical to the main one.
Select any algorithm to act as the fallback. The most useful fallback choices are typically:
Best practices
- Always set a fallback for personalized algorithms. “Personalized for this visitor” only works if the visitor has data. Without a fallback, half your audience may see nothing.
- Match the fallback to the surface. A homepage hero deserves a high-quality fallback (Most Popular). A “Frequently Bought Together” widget on a product page should fall back to category-popular items, not random products.
- Avoid arbitrary fallbacks. “Any Products” with no filters can produce strange combinations (out-of-stock items, off-brand products, retired SKUs). Always layer in basic filters like in-stock and active SKUs.
- Test the fallback explicitly. Use the campaign’s QA step with a brand-new email address (no behavioral history). What renders is exactly what new visitors see — make sure it looks great.
When the fallback fires
The fallback activates whenever the primary algorithm can’t produce a statistically meaningful result. Common triggers:
- The visitor has no on-site behavioral history yet (e.g., new subscriber being shown an open-time email recommendation).
- Their behavior history exists but doesn’t intersect with your current catalog (visited a product that’s been removed).
- The algorithm requires a minimum confidence threshold and the available data falls below it.
You don’t need to manually detect these conditions — Personyze runs the primary, evaluates the result quality, and silently swaps in the fallback when needed.