Loomi Search+ FAQs
This guide answers frequently asked questions about Loomi Search+.
Overall functionality
- If I apply Loomi Search+, how does it impact other Discovery features?
This table summarises the impact on other Discovery features:
Feature | Expected Impact |
---|---|
Automatic Query Filtering | Automated Query Filtering will be applied to the recall generated by Loomi Search+. |
Content Search | No impact. |
Dashboard Synonyms | Synonyms will continue to bring products in recall from Keyword search. |
Dynamic Categories | No impact. |
Dynamic Grouping | No impact. |
Facet Recall | It will be computed from the recall generated by Loomi Search+ (Keyword +Vector). |
Insights | No impact. |
Lookups | No impact. |
Merchandising operations | Merchandising operations like boost, bury, add to recall, etc., will apply to the Loomi Search+ recall. |
Partial Part Number Search | No impact; it will work as it does today. |
Pathways | Loomi Search+ will power the Keyword-based widget recall. |
Query Relaxation | No impact; it will work as it does today. |
Ranking Diagnostics | The products added to the recall via the Vector search component, as well as Keyword match, will be visually indicated in the UI. |
Redirects | No impact. |
Relevance-by Segment | No impact. If Vector search brings products into recall, we will compute the scores of performance signals for these products as well. |
Real-Time Segments | No impact. If Vector search brings products into recall, we will compute the scores of performance signals for these products as well. |
Search Recall Precision modes | Search Recall Precision modes are designed to remove noise from Keyword Search, so these will not have any impact on recall coming from Vector search. We offer vector temperature settings for removing noise that came via Vector search. |
SEO | No impact |
SKU Searchability | Vector search component only retrieves results using product-level data. |
SKU Select | No impact. |
Spell Correction | No impact. |
.
- Can I AB test the new recall experience with Loomi Search+?
Yes, you can A/B test the Loomi Search+ experience using the available API parameters. Visit the guide to see the AB testing process. Note that these are self-managed AB tests since you can manage the tests via a third-party platform. The parameters are controllable through the front-end API. AB testing via dashboard is also available. Visit the AB testing guide. - Does this feature support both Search and Category?
Loomi Search+ is only supported at the Search level. - Is Loomi Search+ available for non-English catalogs?
Loomi Search+ only powers English catalogs in this feature iteration. We will support non-English catalogs in the upcoming feature iterations. - Will it be configurable for customers to decide whether to stay on Keyword or Loomi Search+?
Yes, this is configurable via the dashboard Recall Studio. - Is Loomi Search+ applicable only to the entire site, or can I configure Keyword vs. Loomi Search+ at the query level?
Loomi Search+ can be configured at both levels. - Is this Vector temperature configuration a global setting, or can it also be set on a query/category level?
Vector temperature can be set at the global and query level. Vector search is applicable only for query search. Category search is already relevant and precise. - At which level(s) are the temperature settings available when seeking to reduce noise?
Standard and High. High is more precise than Standard. Standard is the default. - Can Loomi Search+ handle special queries like “shirts under $50”, “sleep aid without melatonin", and "sleep aid with melatonin"?
No, in this iteration, Loomi Search+ isn’t designed to solve these specific use cases, but it can give you close to expected results in some cases. - Is there a way to understand which are the additional products that were generated by the Vector recall?
Yes, Ranking Diagnostics has labels indicating whether the product was brought into the recall using Vector, Keyword, or both.
- Will the recall set of Vector search be added to the ranking rule's Visual Editor? If so, will I be able to identify the products from the Vector search?
Ranking diagnostics will show which PIDs are recalled via Vector search. - Should customers who are set to alternative precision modes switch to Text Match to test Loomi Search+ results?
No, precision modes will continue to work as is, i.e., they will only remove irrelevant products coming up via Keyword search. To remove irrelevant products showing up via Vector search, customers can try the high vector temperature settings. - How does Loomi Search+ handle null results?
Read the detailed article on how Loomi Search+ improves null results with vector temperature relaxation.
Data Handling
- Which ML model powers Loomi Search+?
We use vector embeddings from VertexAI. - Will Vector search also consider external signals like product popularity on Google or social media popularity?
Vector search supports attributes that are part of the catalog data and are marked searchable in the Catalog management. - How long do we expect vector embedding to take in the context of catalog size? Does that change based on catalog size? For example, a big enterprise with multiple 100K products.
Upto 48 hours for generating vector embeddings for the first time. Every 24 hours after that. We update embeddings once a day, regardless of the size of the catalog and frequency of delta/full updates. - How would we know embeddings are generated?
You can see the status of embeddings via the dashboard. - Can Bloomreach accept embeddings in the product feed?
No.
Technical Implementation
- Are there any changes in the Search response format?
The format is unchanged. There are only additional fields that show vector-powered results. - Does the Loomi Search+ generate embeddings at the Catalog level or Site level?
The embeddings are generated at the Catalog level. - Does the feature require changes in feed or pixel?
No additional integration effort is required on your end. - Do we know the impact on search response time for Loomi Search+?
Yes. We have a huge cache to reduce the latency of computing vectors.
Updated about 1 month ago