¶ Product and personalized recommendations in Hummerce
Product recommendations in Hummerce automatically pick items related to what a customer is viewing for them - based on what other customers who viewed the same product ultimately bought. Personalized recommendations go a step further and match the list of suggestions to a specific customer's browsing history. Together they shorten the path to purchase and raise cart value without manually picking related products.
As of Hummerce 2.35, updated 2026-10-07.
¶ Benefits and use cases
- "Customers who viewed this product also bought" recommendations on the item page - with no manual picking of related products.
- Personalized recommendations for every customer, logged in or not, based on their own browsing history.
- The recommendation list fills itself in - when there aren't enough suggestions for an item, category recommendations are pulled in, and as a last resort, store-wide recommendations, so the customer always sees a full list.
- Data comes from Google Analytics 4 - with no need to build a separate customer behavior tracking mechanism from scratch.
- Recommendations can be limited to the customer's current view (e.g. separately for a B2C store and a B2B wholesale channel).
- The AI-assisted configurator suggests the customer's next choices while building a set - see Configurator on the product page.
- The data source is Google Analytics 4 events (product views, purchases) - the store fetches them directly from GA4 or through the BigQuery service, depending on the chosen source.
- These build two datasets: what customers who viewed a given item ultimately bought, and what customers who viewed a given category ultimately bought - each with a count of unique purchases.
- The store fetches recommendations and assigns them to items: sorting from most relevant, excluding the item itself and entries that don't meet the configured criteria. When a recommendation points to an item variant, what the customer sees depends on the store's item list settings: if the store shows articles on lists, the customer sees the parent article; if it shows variants (since version 2.35), the variant itself is recommended.
- When there aren't enough direct recommendations for an item, the list is filled out with category recommendations, and as a last resort, store-wide recommendations.
- The same item does not repeat in a single recommendation list - even when several recommendations point to different variants of the same article (since version 2.35).
- Recommended products show up on the item page - in the Bootstrap 5 template too (since version 2.34.19), as a scrollable list of products. Personalized recommendations appear in the Bootstrap 5 template on the home page.
- It's worth also reporting recommendation impressions and clicks to GA4 - it lets you measure their effectiveness; this is set up during implementation.
- These are based on the item pages a customer has previously viewed, and require product recommendations to be configured - they don't work without them.
- They run asynchronously, so they don't slow down the store page's loading.
- The store first checks whether recommendations for a given customer are already saved in the cache - if so, it shows them right away.
- A customer with no earlier browsing history gets the store-wide global recommendations.
- A customer with history: the store fetches product recommendations for every viewed item, merges them into one list sorted by relevance, and saves the result in the cache.
- The customer sees as many recommended products as could be found - even if that's fewer than the standard limit.
- Product recommendations - an older platform feature (based on Google Analytics), updated to Google Analytics 4 and BigQuery in version 2.31.
- Personalized recommendations (based on the customer's browsing history) - since version 2.28.
- Recommendations in the Bootstrap 5 template (item page and home page) - since version 2.34.19.
- Choice between an article and its variants in recommendations and no repeats in the list - since version 2.35.
- Configurator choice recommendations (AI) - since version 2.35, see Configurator on the product page.
- Google Analytics 4 configured in the store - see Google Analytics 4.
- When choosing BigQuery as the source - the BigQuery service connected to GA4 and to the store (a service task), billed directly through Google Cloud.
- Personalized recommendations require working product recommendations - they can't be turned on separately.
- Requires implementation - covers choosing the data source (GA4 or BigQuery), setting the criteria for items covered by recommendations and for recommended items, and, if BigQuery is chosen, also connecting the service to GA4 and to the store.
- Minimum estimated time:
- product recommendations (item criteria, data source, package size) - from 6 h, not including the cost of the BigQuery service itself,
- personalized recommendations (require working product recommendations) - from 3 h,
- reporting recommendation impressions and clicks to GA4 - from 2 h.
- The time for a specific implementation depends on its scope - we quote it individually.
- What the time depends on:
- the number of views covered by recommendations (e.g. B2B and B2C separately),
- the complexity of the criteria for recommended items and for items covered by recommendations,
- whether Google Analytics 4 (and possibly BigQuery) are already configured in the store,
- the store's version - an older version needs an update first (quoted separately).
- When BigQuery is chosen as the source - the service is billed (by Google) and requires a payment card on file; cost can be limited with Google Cloud quotas, but it's a cost outside Hummerce.
- Without enough data in Google Analytics 4 (low traffic, a new store), recommendations may fall back mostly on store-wide suggestions.
- Personalized recommendations only work when product recommendations are configured - they can't be enabled independently.
- We do not verify the configuration for legal compliance (e.g. consent for processing customer behavior data, GDPR) - this is set by the store.
Are recommendations the same for every customer?
Not always. Personalized recommendations pick suggestions based on a specific customer's browsing history, regardless of whether they're logged in.
Where does the store get the data for recommendations?
From Google Analytics 4 - directly, or through the BigQuery service, depending on the chosen source.
What does a customer who hasn't viewed anything before see?
Store-wide global recommendations - the recommendation list is never empty.
Can recommendations differ between the B2B and B2C views?
Yes, once filtering recommendations by the customer's current view is enabled.
Is using BigQuery free?
No. It's a Google service billed separately from Hummerce - cost can be controlled with Google Cloud quotas.
How long does implementing product recommendations take?
From 6 h - covers choosing the data source, item criteria and the recommendation package size, not including the cost of the Google service itself.
Want to check how product recommendations could raise conversion in your store? Contact the Hummerce team (in Polish).
- Configuration: Settings -> Configuration -> Recommendations. Available parameters:
- Source for getting default recommendations - Google Analytics 4 or BigQuery,
- Time to collect recommendations (days) - how many days back the algorithm analyzes,
- Criteria of items to which recommendations are generated,
- Criteria for recommended items (e.g. availability, visibility, purchasability),
- Maximum size of the single data package retrieved from BigQuery,
- Number of recommendations assigned to item during BigQuery import,
- Number of recommendations displayed on item page,
- Enable filtering recommendation items by current customer view,
- Enable caching recommendations for user.
- Personalized recommendations don't have a separate settings section - they use the product recommendations configuration.
¶ REST API and MCP
- REST API: no dedicated operations for recommendations.
- MCP: general module configuration tools (
ec2_module_describe, ec2_config_search, ec2_config_get) - see AI assistant - MCP server.
The module has no message keys of its own - the recommendation list uses the same templates as the other product lists in the store template.
No additional item or customer parameters - the choice of recommendations relies solely on the item criteria set in the module's configuration.