Optimizing Vexture Search Feature with Tuning
Once you have gone through the Vexture feature set up with professional services you will see the new AI settings and Vexture Search Indexes tab options added to your Admin.
Path: Settings > Store Settings > More > Vexture Search Indexes
Path: Settings > Store Settings > More > AI Settings
Professional Services assisted you in configuring Vexture into your Miva admin and created your Vexture Search Index. The next step is to tune your Vexture Search Results. Every merchant’s product catalogue is unique, Vexture Tuning allows you to make real time adjustments to ensure search results are relevant to your customers, all in the Miva Admin.
What is Vexture?
Vexture is Miva’s native AI-powered search and merchandising solution natively integrated into the Miva platform. Unlike third-party systems, it runs locally, ensuring speed and reliability.
Instead of only using keyword matches like traditional relational search features (where “visor” won’t return “visar” if misspelled), it uses vector-based semantic search combined with traditional search to provide the best possible results:
How it works:
- Every product’s text data (name, description, custom fields) is converted into a vector embedding — a long list of numbers that capture meaning.
- When a shopper types a query, it is also converted into a vector.
- The most semantically similar products are returned as results.
Key advantages:
- Understands intent beyond exact keywords.
- Handles typos, synonyms, and natural language queries.
- Native to the Miva platform → no fragile integrations, faster than third-party search.
- Native integration: Works directly with facets, filters, Miva APIs, and flex components like carousels and product lists.
- Results can be tuned (specific, broad, sparse) depending on store preference
Tuning Vexture Search Results
Path: Settings > Store Settings > More > Vexture Search Indexes > Select Index (check box) > Tune Results > AI Search Rules (left side menu)
Vexture AI provides a real-time tuning interface so administrators can dial in search results to include the best possible and closest related products. AI tuning delivers the closest option to what the shopper is searching for, this super intelligent non-zero search result capability drives sales by ending dead end searches of the past. Exact matches (traditional results) are included when available and prioritized as 100% relevant.
The AI side of Vexture returns results in one of three ways based on the relevancy score of the first result for a search. AI will display products from most similar results to least similar results cutting off based on the tuning settings created in the AI Search Tuning admin.
Specific Results = where the first result is highly relevant
Broad Results = where the first result is moderately relevant
Sparse Results = where the first result doesn’t hit the required thresholds for specific or broad.
1. Specific Search Results
This classification of result allows tuning for when a search term finds very good results. The logic is that if there are very good results, we don’t want to dilute them with mediocre results. The primary reason is that your shopper may re-order results, e.g. by Price which would hide the great results among the not-so-great results
- Triggered when the first result is equal to, or better than the configured threshold, in this example 82%
- A specific search needs a “cutoff” configured. In the example above the cutoff is 77.
- For example,
- It is recommended to tinker with these percentage thresholds to find what is best for each store catalog.
2. Broad Search Results
Useful when shoppers might type vague queries and you still want to guide them toward relevant categories.
If a shopper’s search term does not trigger a Specific search, but the first result is the equal to or better than the Broad Search setting, then a Broad search is triggered, and the results will be all products with a relevance greater than or equal to this configured number.
- Triggered when the best result is lower than the Specific Results threshold, but higher than the Broad Search Threshold. In this case, that’s where the first result is between 82 and 68%.
- It is recommended to tinker with these threshold percentages to find what is best for each store catalog. Run some searches with meaningful, but not specific products, and look at the returned results. Scroll down to the point where results are being excluded (in this case at 68%). Are the products at the bottom of the results relevant? If not, this cutoff might be too low. If products below this cut off are still relevant, then this number might be too high. Run several searches and try and find a setting that works best most of the time.
3. Sparse Search Results
- Triggered when the first result has a relevancy below the Broad Search setting.
- Don’t think of this result set as a search fail! It is very powerful and will result in sales. Think of it this way. These results are the closest thing to what the customer is searching for, and the customer is likely searching for something you sell. This is where Vexture search shines compared to traditional search which will very often return no results.
- The configuration for this results set is simply how many products should be returned.
- Helps avoid “dead ends” in the shopping experience.
The Exact Match Layer - Blended Traditional Search Rules
The Vexture feature also supports blending traditional search with semantic search, because, while AI search is powerful for intent, another common shopper’s search habit is to type what is effectively a series of keywords, and even sometimes a single word – which is a part number. This is where blending comes in. It combines the power of AI search to understand when shoppers type semantically meaningful searches with keyword exact match searches.
You can activate the Exact Match Layer by toggling the Blended Traditional Search Rules located below the Sparse Results sections of the Tune Vexture Search Results page. Here you can add Exact Search Fields for products like SKU, product codes, product names, etc. AI will check this field first for exact matches then search results based off the relevancy scores defined above.
The Exact Match configuration is very similar to Miva’s traditional search configuration; however, you should think differently when configuring these settings. In traditional Miva Search you are casting a wide net. You are including fields with somewhat lose rules to try and include as many products as possible. With Exact Match you can rely on the AI Search to include products based on context relevancy. What you want to do with the Exact Match fields is dial things back to specifically try and layer in results for exact match scenarios.
Enable the Exact Match Layer
Path: Settings > Store Settings > More > Vexture Search Indexes > Select Index (check box) > Tune Results > Scroll down the left side menu to Blended Relational Search Rules
- Toggle Blended Relational Search Rules
- Click Make Selection
- Toggle Enable fields that contain data likely to be exact matches to shopper searches. Example: If your SKU field contains part numbers, enable for SKU, so shoppers searching by part number can find the product easily.
- Select Search Type from drop down (see below)
- Click Update
Quick guide to the 4 Search Type options in Exact Match Fields selections:
- Exact Match – The entire search string must exist in the search fields, exactly as searched, ex: “Blue Jersey”.
Example: You publish a catalog that includes part numbers, and those part numbers are the Miva SKU field. - Contains - The entire search string as entered is searched for and can be a substring of another term.
Example 1: You have a custom field containing a comma delimited list of alternative part numbers. This would match if the shopper searched for just one of them.
Example 2: Shoppers search for partial part numbers, ABC-. You want them to see results for products ABC-1, ABC-2, ABC-3 etc. - Contains (Term) - This is the default setting. The search string is split into individual words, terms, all the terms must appear in the fields being searched, in any order including as a substring. This is good configuration for your product NAME field. It will work when a customer enters several specific keywords that are all contained in the product name. This COULD be a good setting for the product description field, however, warning: this can often lead to too many non-relevant results.
- Full Text – A natural language comparison search is performed. Not recommended for Vexture AI Settings
Tuning Tips
Test a Variety of Queries
- Exact product names, SKU, or keywords to test Exact Matching
- Broad categories, to tune Specific, and Broad AI search results.
- Common misspellings, synonyms, even long dialogue searches to see how Sparse searches – the catchall – are of very valuable .
Leverage Facets Alongside Tuning
If your site has a large catalog, and supports Miva Facets, tune Broad Search to return more results rather than less. It's OK for some less relevant products to be included in the results. Shoppers will zero in on what they are looking for using facets.
If you have a large catalog and are not using facets, consider if facets could improve your shoppers experience.