Personyze Wiki Personyze Wiki docs
Open Personyze
Recommendations

Managed Upselling and Cross-selling

Upload your own curated upsell and cross-sell relationships (which products go with which) so Personyze can recommend them in widgets and emails — instead of relying solely on aggregate purchase…

Updated 4 days ago 3 min read
A
by Admin

Personyze’s standard recommendation algorithms learn cross-sell relationships from aggregate purchase data — “people who bought X also bought Y.” That works beautifully for established catalogs with rich purchase history, but it has two limits: brand-new products without sales data, and curated bundles where you (the merchant) know better than the data does.

Managed Upselling and Cross-selling is the answer: you upload your own curated product-to-product relationships, and Personyze uses those instead of (or in addition to) the auto-learned ones.

When managed cross-sell is the right fit

  • New product launches — no aggregate data exists yet, but you know what should pair with the new SKU.
  • Strategic bundling — you want to push specific accessory pairings (“iPhone 15 + this case + this charger”) regardless of what the data alone would surface.
  • Margin optimization — your highest-margin accessories should go with high-volume products, even if other items are technically more popular.
  • Editorial picks — your merchandising team knows the lifestyle pairings that aggregate data misses (camera + lens + bag combos).
  • Fallback for sparse-data products — use managed cross-sell as the fallback algorithm when crowd-based recommendations come back empty.

Data format

The relationships are uploaded as a simple two-column structure:

Subject (main product ID) Linked (cross-sell product ID)
iphone-15 case-clear-15
iphone-15 charger-magsafe
iphone-15 headphones-airpods
iphone-15-pro case-clear-15
iphone-15-pro charger-magsafe

The same partner ID can be linked to multiple subjects — a charger or a case can be a cross-sell for many phones. There’s no limit on how many cross-sells a single subject product can have, but in widgets you’ll typically display 3–5.

Three ways to upload

📄 File uploadDrop a CSV with subject_id and linked_id columns. Best for one-time setup or infrequent updates. Re-upload to refresh.
🔄 Synced feedProvide a URL to a feed that Personyze polls on a schedule (hourly, daily, weekly). Best when your cross-sell relationships are managed in another system that exports periodically.
🔌 API pushUse Personyze’s API to push relationship updates programmatically. Best for real-time updates or when relationships are computed live by your backend.

Configure all three under Settings → Products Catalog → Cross-sell / Upsell Data. After upload, Personyze shows a preview so you can verify the data parsed correctly.

Using managed data in widgets

Once the relationships are uploaded, the Managed Cross-Sell and Managed Upsell options appear in any recommendation widget’s algorithm picker. Select either as the primary algorithm — Personyze will surface the linked products for whatever the trigger product is.

Common usage patterns:

  • Product page widget — “Frequently paired with this” → managed cross-sell with the current product as trigger.
  • Cart widget — “Complete your setup” → managed cross-sell triggered by whatever’s in cart.
  • Email recommendation — order confirmation email shows accessories for the products just purchased.
  • Fallback role — set “managed cross-sell” as the fallback for a “frequently bought together” widget so new products without aggregate data still get strong recommendations.

Best practices

  • Mix curated with auto-learned. Use managed data for hero products and strategic pairings, let auto-learned cross-sells handle the long tail.
  • Refresh regularly. Old SKUs come back to haunt you when they show up as cross-sells for current products. Sync at least monthly.
  • Audit emptys. If a subject has no linked products in your data, the widget falls back. Check the preview after each upload to spot SKUs that lost their pairings.
  • A/B test against aggregate. Run managed cross-sell against auto-learned cross-sell for the same trigger products — see which produces better conversion before committing to one approach catalog-wide.