Product recommendations
Pick the right products from your product catalog for each recipient and show them in campaigns, journeys and emails. Recommendations are computed daily from your event data. Decision record: ADR-0063. A product catalog must be uploaded first…
Pick the right products from your product catalog for each recipient and show them in campaigns, journeys and emails. Recommendations are computed daily from your event data.
Decision record: docs/adr/0063-urun-oneri-motoru.md. A product catalog must be uploaded first.
Setup
Dashboard → Product Catalog → Recommendations:
| Field | Description | Default |
|---|---|---|
| View events | Events counted as product views, at most 10 | Empty |
| Purchase events | Events counted as purchases, at most 10 | Empty |
| Product ID property | The event property that carries the ID | product_id |
| Window | Number of recent days to count, 7–365 | 90 |
Enter at least one view or purchase event. Saving starts the first computation; after that it refreshes every day at 04:50 UTC.
Your event can carry the product ID as a string, number or array:
{ "name": "purchase", "properties": { "product_id": ["A1", "B2"] } }IDs that are not in the catalog and events from users without a known identity are not counted.
Strategies
| Strategy | What it shows | If there aren't enough results |
|---|---|---|
personalized | Products similar to the user's products | popular, new_arrivals |
popular | Products with the most interactions in the window (purchases count 5×) | new_arrivals |
trending | Products rising in the last 7 days compared to the previous 7 | popular, new_arrivals |
similar | Products similar to a given product | popular, new_arrivals |
category | The most popular products in a given category | None |
recently_viewed | Products the user viewed but didn't buy | None |
new_arrivals | Products most recently added to the catalog | None |
Out-of-stock products and products the user already bought are excluded. Similarity is derived from users who engaged with both products; a pair needs at least 2 shared users. When data is sparse, the list is filled from the fallback source.
In templates
{% recommendations 'urunler' strategy: 'personalized' limit: 4 as liste %}
{% for urun in liste %}
{{ urun.title }} {{ urun.price_text }}
{% endfor %}{% recommendations 'urunler' strategy: 'similar' item: custom_data.product_id as benzerler %}
{% recommendations 'urunler' strategy: 'category' category: 'Ayakkabı' limit: 3 as enIyiler %}limitis a constant number from 1 to 20; the default is 4.strategymust be a quoted constant string.itemandcategorytake a variable path or a quoted constant string; filters cannot be used.- Product fields are the same as in the product catalog.
- Works for push, email, SMS, WhatsApp and RCS sends. In in-app messages the list returns empty.
- The preview shows sample products.
Email block
In the email designer, drag in the Recommendations block and choose the catalog, strategy and number of products (1–6). If the list is empty, the block, including its heading, is hidden in the email, and the email is still sent.
API
curl "https://api.pushizo.com/v1/apps/$APP_ID/catalogs/$CATALOG_ID/recommendations?strategy=personalized&external_id=musteri-1&limit=4" \
-H "Authorization: Bearer $PUSHIZO_API_KEY"| Endpoint | Scope | Description |
|---|---|---|
GET .../recommendations | analytics:read | strategy, limit, user_id, external_id, item_id, category |
GET .../recommendations/config | analytics:read | Model and computation status |
PUT .../recommendations/config | admin | Configures the model and starts a computation if enabled |
POST .../recommendations/compute | admin | Queues a computation (202) |
source in the response is the source of the first product in the list; if a fallback was used, it differs from the strategy value. These endpoints require an API key and must not be called directly from a browser or
app.
Limits and privacy
- Recommendations are daily; same-day behavior is reflected in the next computation.
- Similarity looks only at behavior; titles and categories are not used. A product with no interactions appears only in
new_arrivalsandcategory. - A computation takes the most recent 200,000 users and the last 20 products per user.
- When a user is deleted, their personal list is deleted too. Access to the feature depends on the Dynamic Content plan.