Your search engine has been running for years. It knows your catalog, supports your filters, and fits into an eCommerce platform your team has spent considerable time building. Someone has worked through the product-code quirks, the inconsistent brand names, and that one category that never quite behaved like the others. You may have already invested in relevance tuning and seen a worthwhile improvement but now you want more from it.
We’ve had a company return after a successful tuning project with exactly that request: they wanted to “take search to the next level.” And while that’s a reasonable ambition it can also be seen as an open question. Better results? More control for merchandising? Some of the AI capabilities everyone keeps talking about?
For a retailer already using Elasticsearch, OpenSearch, or Solr, taking search to the next level doesn’t have to mean replacing the engine. Before you decide whether to stay or go, it helps to understand what your current search can support and where it might be holding you back.
You might want search to bring in more sales while your merchandisers just want to move a few products without opening a ticket. Meanwhile, the developer handling those tickets has other work piling up. Better search needs to help the people running it too.
Consider a search for “patio chairs” that returns 430 products. The engine has matched the words, but the first page is dominated by chair covers and replacement cushions. The actual chairs are buried further down, and the filters don’t separate furniture from accessories. There’s no single fix hiding in this example. Solutions may need to address query handling so core products outrank accessories, cleaner product attributes for filtering, and merchandising tools that let a business user feature a seasonal patio assortment.
Better search would put the chairs where shoppers expect to find them and make it easy to narrow down the choices. Your merchandising team could feature the seasonal range without waiting for a developer and see how shoppers respond. Analytics and insights about the shopper’s journey would be easily available and helpful in making decisions on next steps.
For example, your team might test “soft-close drawer dampers” because that’s what the catalog calls them. A shopper might type “stop drawers slamming.” The first search works beautifully while the second may turn up nothing. Searches with abbreviations, older product names, and everyday descriptions of what a product does may not be giving desirable results.
Looking beyond the results to what shoppers do next can help you understand whether or not they’re finding what they need. Searches that return nothing obviously should get attention, but so do searches that leave shoppers digging through hundreds of products. Repeated reformulations, little product engagement, or shoppers leaving after seeing results can point to things worth investigating (though price, availability, and delivery options can play a part too).
Look at the team’s work as well. Which changes require development? What gets maintained in spreadsheets? Are there rules nobody wants to touch because their original author left three years ago? You may discover that relevance is reasonably strong, while the business has very little visibility or control.
If products do begin showing up where they should, the team might start asking more questions. Why do products that shoppers rarely choose keep appearing above products they engage with? Can we feature a seasonal assortment for particular searches? How do we find queries that have become less effective as the catalog changes?
When a retailer introduces a new product line there’s little behavioral history to guide placement. A ranking approach heavily influenced by past engagement may give it limited visibility so merchandisers may want to promote it in selected contexts, examine how shoppers respond, and adjust a month later. This kind of process needs controls, useful reporting, and the ability to review and modify the outcome.
It’s frustrating to see when search teams are expected to improve search performance with little more than a conversion rate chart and a search box. They need to understand what’s happening between the search and the sale and have a way to act on it. FindTuner gives teams using Elasticsearch, OpenSearch, or Solr a way to see how shoppers use search and make improvements without needing a developer for every change.
Someone searching for “stop drawers slamming” may never think to type “soft-close drawer dampers.” Keyword search could miss those products unless you’ve accounted for that wording. Semantic search has a better chance of finding them because it can connect the description with what the product does.
Another example is “water filter for a Samsung RF28R7351 refrigerator.” Finding refrigerator filters is only part of the job. They have to fit that model. Semantic search may bring back filters for other refrigerators because the descriptions are similar. Keyword search can pick out RF28R7351, but the extra words in the query can get in the way if they’re all treated as required matches.
You could handle this search by recognizing the model number and using it to filter for compatible products, then using semantic search to help find the water filters within that selection. Your product data still needs to list which refrigerator models each filter fits so search can return compatible products.
With FindTuner, a strategy can apply that filter and control how keyword and semantic results work together. You can choose where hybrid search applies, how many products semantic search contributes, and how much those products are boosted. That lets you use it broadly or focus on searches that need help, while preserving the exact-match behavior that already works well.
You can evaluate the changes against your current results using both difficult queries and everyday searches. The goal is to help shoppers who weren’t finding what they needed without making familiar searches less reliable.
Your search engine may support all kinds of adjustments but if business users need a developer to make them, even a simple change adds friction and is a lost opportunity for both the business team and the development team. There’s a ticket, some back-and-forth about what’s needed, and a review of the results all while the campaign deadline gets closer. With the right merchandising tools, business users should be able to handle everyday changes themselves, whether that’s promoting products for a search, reordering a category, scheduling a campaign, or choosing which filters shoppers see.
With all the talk about AI, shouldn’t it be able to handle this for you? It absolutely does, and it can automate a lot, but business users still need a say in the results. A supplier promotion may need to go live on Monday regardless of what past clicks and sales would put first. Merchandising controls let them make that happen without waiting for shopper behavior to catch up.
Business users also need to know where their changes will apply and how they’ll interact with rules already in place. Promoting a product for one search shouldn’t unexpectedly change its position across the site and a preview should help catch those surprises before shoppers see them. Afterward, reporting should show whether shoppers clicked on the promoted products and went on to buy them. Otherwise, it’s easy to leave a change running long after anyone remembers why it was made or whether it helped.
On a custom eCommerce platform, search often does more than return products. It supports filters, category pages, and rules about which products shoppers can see. Improvements need to preserve those behaviors while still offering opportunities to improve the shopper experience.
Your developers will need to connect FindTuner to the existing search process and make sure everything still works as expected.
FindTuner follows this approach by adding search and discovery optimization capabilities to an existing Solr, Elasticsearch, or OpenSearch implementation. Depending on the requirements, those capabilities can include merchandising controls, analytics, behavioral optimization, and hybrid search.
You may end up combining a few of these options. Adding merchandising tools, for example, doesn’t mean you’re finished tuning relevance or improving product data. When comparing costs, look beyond getting it launched. What will your developers and merchandisers need to do to keep it working well? Looking at the next three years can give you a more useful comparison than putting a subscription price next to a one-time development estimate.
Ready to see what’s possible with the search engine you already have? See FindTuner in action with your own catalog. We’ll show you what you can improve, how your team can manage it, and what it takes to get started.
We’ve had a company return after a successful tuning project with exactly that request: they wanted to “take search to the next level.” And while that’s a reasonable ambition it can also be seen as an open question. Better results? More control for merchandising? Some of the AI capabilities everyone keeps talking about?
For a retailer already using Elasticsearch, OpenSearch, or Solr, taking search to the next level doesn’t have to mean replacing the engine. Before you decide whether to stay or go, it helps to understand what your current search can support and where it might be holding you back.
What would better eCommerce search actually look like?
Better eCommerce search combines accurate product matching (aka relevancy), an engaging experience around the results, and practical tools for the team maintaining it. Where you start depends on what’s frustrating your shoppers or making extra work for your team.You might want search to bring in more sales while your merchandisers just want to move a few products without opening a ticket. Meanwhile, the developer handling those tickets has other work piling up. Better search needs to help the people running it too.
Consider a search for “patio chairs” that returns 430 products. The engine has matched the words, but the first page is dominated by chair covers and replacement cushions. The actual chairs are buried further down, and the filters don’t separate furniture from accessories. There’s no single fix hiding in this example. Solutions may need to address query handling so core products outrank accessories, cleaner product attributes for filtering, and merchandising tools that let a business user feature a seasonal patio assortment.
Better search would put the chairs where shoppers expect to find them and make it easy to narrow down the choices. Your merchandising team could feature the seasonal range without waiting for a developer and see how shoppers respond. Analytics and insights about the shopper’s journey would be easily available and helpful in making decisions on next steps.
How do you know where your current search falls short?
There tends to be a general intuition about when a search engine is not performing as desired. Queries that seem like obvious paths to performing products may show inconsistent or even incorrect results. Queries that are intuitive may fail to produce results that are just good common sense. Shopper paths may show engagement that seems promising but is not resulting in conversions. Things are just…off.For example, your team might test “soft-close drawer dampers” because that’s what the catalog calls them. A shopper might type “stop drawers slamming.” The first search works beautifully while the second may turn up nothing. Searches with abbreviations, older product names, and everyday descriptions of what a product does may not be giving desirable results.
Looking beyond the results to what shoppers do next can help you understand whether or not they’re finding what they need. Searches that return nothing obviously should get attention, but so do searches that leave shoppers digging through hundreds of products. Repeated reformulations, little product engagement, or shoppers leaving after seeing results can point to things worth investigating (though price, availability, and delivery options can play a part too).
Look at the team’s work as well. Which changes require development? What gets maintained in spreadsheets? Are there rules nobody wants to touch because their original author left three years ago? You may discover that relevance is reasonably strong, while the business has very little visibility or control.
What comes after relevance tuning?
If the team has completed a round of search relevancy tuning and core relevance is working well the next opportunity often involves merchandising control, better analytics, and ongoing optimization using shopper behavior. Further tuning may still be part of the work, but usually at this stage we’re looking for ways to begin elevating the experience further.If products do begin showing up where they should, the team might start asking more questions. Why do products that shoppers rarely choose keep appearing above products they engage with? Can we feature a seasonal assortment for particular searches? How do we find queries that have become less effective as the catalog changes?
When a retailer introduces a new product line there’s little behavioral history to guide placement. A ranking approach heavily influenced by past engagement may give it limited visibility so merchandisers may want to promote it in selected contexts, examine how shoppers respond, and adjust a month later. This kind of process needs controls, useful reporting, and the ability to review and modify the outcome.
It’s frustrating to see when search teams are expected to improve search performance with little more than a conversion rate chart and a search box. They need to understand what’s happening between the search and the sale and have a way to act on it. FindTuner gives teams using Elasticsearch, OpenSearch, or Solr a way to see how shoppers use search and make improvements without needing a developer for every change.
Do you need keyword search, semantic search, or both?
You’ll probably want both keyword and semantic search. Keyword search works great when shoppers use the words you have in your catalog data. Semantic search helps when they describe what they need in words your catalog doesn’t use.Someone searching for “stop drawers slamming” may never think to type “soft-close drawer dampers.” Keyword search could miss those products unless you’ve accounted for that wording. Semantic search has a better chance of finding them because it can connect the description with what the product does.
Another example is “water filter for a Samsung RF28R7351 refrigerator.” Finding refrigerator filters is only part of the job. They have to fit that model. Semantic search may bring back filters for other refrigerators because the descriptions are similar. Keyword search can pick out RF28R7351, but the extra words in the query can get in the way if they’re all treated as required matches.
You could handle this search by recognizing the model number and using it to filter for compatible products, then using semantic search to help find the water filters within that selection. Your product data still needs to list which refrigerator models each filter fits so search can return compatible products.
With FindTuner, a strategy can apply that filter and control how keyword and semantic results work together. You can choose where hybrid search applies, how many products semantic search contributes, and how much those products are boosted. That lets you use it broadly or focus on searches that need help, while preserving the exact-match behavior that already works well.
You can evaluate the changes against your current results using both difficult queries and everyday searches. The goal is to help shoppers who weren’t finding what they needed without making familiar searches less reliable.
How do you give the business team control while keeping search consistent?
Your business team needs to be able to make changes, choose where they apply, and see what the results will look like before putting them live. They should be able to recognize and respond to immediate trends, handle shifting seasonal promotions, and make tailored and highly-specific adjustments on the fly. They should also be able to tell when an existing rule might interfere with what they’re trying to do.Your search engine may support all kinds of adjustments but if business users need a developer to make them, even a simple change adds friction and is a lost opportunity for both the business team and the development team. There’s a ticket, some back-and-forth about what’s needed, and a review of the results all while the campaign deadline gets closer. With the right merchandising tools, business users should be able to handle everyday changes themselves, whether that’s promoting products for a search, reordering a category, scheduling a campaign, or choosing which filters shoppers see.
With all the talk about AI, shouldn’t it be able to handle this for you? It absolutely does, and it can automate a lot, but business users still need a say in the results. A supplier promotion may need to go live on Monday regardless of what past clicks and sales would put first. Merchandising controls let them make that happen without waiting for shopper behavior to catch up.
Business users also need to know where their changes will apply and how they’ll interact with rules already in place. Promoting a product for one search shouldn’t unexpectedly change its position across the site and a preview should help catch those surprises before shoppers see them. Afterward, reporting should show whether shoppers clicked on the promoted products and went on to buy them. Otherwise, it’s easy to leave a change running long after anyone remembers why it was made or whether it helped.
Can you build on the search engine you already have?
Yes, you can keep your existing engine and add business user controls, search analytics, and an easier way to put AI search to work. Elasticsearch, OpenSearch, and Solr support many AI search capabilities. FindTuner helps you adopt them more quickly and build on work your team has already done.On a custom eCommerce platform, search often does more than return products. It supports filters, category pages, and rules about which products shoppers can see. Improvements need to preserve those behaviors while still offering opportunities to improve the shopper experience.
Your developers will need to connect FindTuner to the existing search process and make sure everything still works as expected.
FindTuner follows this approach by adding search and discovery optimization capabilities to an existing Solr, Elasticsearch, or OpenSearch implementation. Depending on the requirements, those capabilities can include merchandising controls, analytics, behavioral optimization, and hybrid search.
How should you choose an improvement path?
You choose an improvement path by looking at what needs to work better and what your team can realistically take on. That helps you decide whether further tuning will do the job, you need additional tools, or it’s time to consider replacing the engine.| Approach | The Task | When to Implement |
|---|---|---|
| Further relevance tuning | Improving matching and ranking through field weighting, query handling, synonyms, and other adjustments to how the engine uses your catalog data. | When relevant products are missing or poorly ranked, whether search hasn’t been tuned before or results have slipped as the catalog changes. |
| Custom development | Specialized requirements that your team has the expertise and capacity to build. | When a specific business requirement isn’t covered by existing tools and your developers can build and maintain it. |
| An optimization platform | Adding merchandising, analytics, and ongoing optimization while keeping your existing search engine. | When the engine works, search teams need more control, better analytics, or an easier way to manage ongoing improvements. |
| Search replacement | Addressing limitations in capabilities, performance, support, or architecture that warrant a different foundation. | When the current engine can’t meet your needs, or improving and maintaining it would take more work than replacing it. |
How do you justify the investment?
You can start with a few things you probably already track:- Search-related revenue and margin. How much business flows through search, and where could improvements plausibly affect outcomes? Avoid applying an assumed uplift to all online revenue.
- Evidence of friction. Identify important queries with weak engagement, repeated reformulations, or other signs that warrant investigation.
- Staff time. Estimate how much effort goes into routine ranking requests, manual reporting, and merchandising workarounds.
What’s your next step?
Taking search to the next level starts with deciding what you want to do better. That might mean understanding which searches disappoint shoppers, giving merchandisers more control, or helping customers find products without knowing the catalog terminology. We can help you work through those questions, look at what your current search is doing, and decide what’s worth adding or changing.Ready to see what’s possible with the search engine you already have? See FindTuner in action with your own catalog. We’ll show you what you can improve, how your team can manage it, and what it takes to get started.