Most stores show everyone the same thing - and sort by whatever the last collection edit left behind. Search, merchandising and recommendations belong together.
Synonyms, misspellings, weighted fields and filters that match your range. Plus the question of whether the built-in tools suffice or a dedicated search index is needed.
Sort logic per collection, promoting campaign and margin products, handling sold-out items and automated collections instead of manual curation.
Related and complementary products on product and cart pages, recommendations in the customer account, and segments for different treatment by market, customer group or purchase history.
Search, filters and recommendations work with what sits in titles, types, tags and metafields. If the data is thin, no software helps - it only makes the gap visible faster.
The default sort decides what thousands of visitors see first. It is the lever with the best effort-to-effect ratio - and in many stores it has been untouched for years.
Synonyms, field weighting, filters and complementary products can be set up at no extra cost. That carries surprisingly far for ranges up into the low five-figure product count.
An external search index pays off with large ranges, many attributes, multilingual catalogs or hard relevance requirements - not as the first response to a poor result list.
Collections keep showing unavailable products in the top rows. Automated collections and sort rules solve that, but it does not happen on its own.
Searches with no result reveal missing synonyms, wrong terms or genuine gaps in the range. Not evaluating them wastes the best source of improvements you have.
Filters built from inconsistently maintained attributes produce options with no matches. That is a data quality problem, not a front-end problem.
Recommendations showing what the customer just bought cost trust. Rules for exclusions and complements matter more than the algorithm behind them.
We look at top search terms, zero results, drop-off after search and the conversion rate of search users. In many stores, search users buy far more often than average.
Titles, product type, tags, attributes and metafields get checked for completeness and consistency. Without that step, any relevance tuning stays guesswork.
Configuring synonyms, field weighting, filter axes and complementary products. Then we review the result lists for the terms that actually generate revenue.
Defining the default sort per collection, factoring in availability, defining campaign slots. Rules rather than manual ordering, so it still holds next season.
Building related and complementary products in where they answer a question: product page, cart, and after purchase in the customer account.
Watching search-user conversion, click rate on recommendation areas and revenue per collection. Personalisation without measurement is just a claim.
First-time buyers need orientation and trust; repeat buyers need the shortest path to a reorder.
Range, availability and recommendations differ per market - the form of personalisation with the clearest business case.
Company accounts see their own catalogs and prices. Search and sorting should follow that split rather than fight it.
Consumables, spare parts and accessories are far better recommended from past orders than from behaviour within a session.
A noticeable share of your visitors starts at the search bar. The result list then decides your revenue, not the homepage.
What was sortable by hand at a hundred products is not at a thousand. From there, rules replace manual curation.
Unavailable products keep sitting in the top rows and crowd out exactly the items you can actually ship.
Company accounts, markets or repeat buyers need their own view of the same range - and currently all see the same order.
Titles, product types, tags and metafields are inconsistent or half empty. Search, filters and recommendations cannot build anything reliable from that.
Connect product data from PIM and ERPWithout a meaningful volume of searches there are no zero-result queries to evaluate and no basis for testing sort rules against each other.
Build visibility firstYour visitors land on the right product page and still leave. The problem then sits behind the result list, not in front of it.
Go to CRO optimisationThree formats - from evaluating your search data through to ongoing upkeep of relevance and sorting.
For stores that suspect search is costing revenue but have no numbers in front of them.
Ends with a ranked set of levers: what works first, what can wait, what you maintain yourself.
For stores where the priorities are settled and search, filters and sorting are to be genuinely rebuilt.
Ends with rules that still hold after the next delivery and into the next season.
For stores with a frequently changing range whose result lists would otherwise be stale again half a year later.
The result is a result list that grows with the range instead of trailing behind it.
For most ranges the built-in tools carry further than expected: synonyms, field weighting, filters and complementary products are configurable at no extra cost. An external search index pays off with very large catalogs, many technical attributes, multilingual content or when you need fine-grained relevance control.
Because both live off the same source: your product data. A recommendation is a filtered, sorted list - exactly like a search result. Splitting the topics means building the same data work twice and later wondering why different parts of the store disagree.
Shopify offers no built-in sort order for that. The usual route runs through automated collections with an inventory condition, plus rules in the theme or an app. What matters is that the rule applies automatically - a one-off manual sort is wrong again after the next delivery.
They work mainly where they answer an open question: matching accessories on the product page, complements in the cart, reorders in the customer account. As a decorative carousel at the bottom of a page they deliver little. Placement and exclusion rules matter more than the logic behind them.
Personalisation based on behaviour and profiles touches consent and data processing. Recommendations derived from the catalog itself - related products, complements, sort rules - need no personal profiles. That is why we build what works without profiling first. Anything built on profiles comes afterwards, and only with a properly wired consent state.
Through the conversion rate of search users against all users, the share of zero-result searches, the click rate on recommendation areas and revenue per collection before and after the change. Where traffic allows, we test sort and recommendation rules against each other instead of asserting them.
We review search terms, zero results and collection sorting and tell you which lever works first for your range.
Last updated:
Project start
Usually a response within 24 hours
Talk directly to a strategy or tech senior
No agency slide deck, just clear next steps
NICCOS
Die Seite konnte nicht geladen werden.
Bitte lade die Seite neu. Falls gerade ein Update live gegangen ist, wird damit die aktuelle Version geladen.