Search in e-commerce

Your shop's search decides more sales than you think

Someone typing in the search box has already decided to buy. They are only checking whether you have it. And yet it is the part of the shop that fewest people look at.

Updated August 2026 · 9 minute read

There is an important difference between someone browsing your categories and someone using the search. The first is exploring. The second already knows what they want and has decided that the fastest way to find it is to type it.

That difference in intent translates into money, and fairly brutally.

Read it the other way round, which is how it really sinks in: the traffic most likely to buy from you goes through a feature that, in most shops, nobody has reviewed since the day it was installed.

The three ways to lose that sale

A search engine does not fail in one way. It fails in three, and only one of them is obvious from the outside.

1. It finds nothing

Between 10% and 15% of searches in an average shop return not a single result. Each one of them is a person who wanted to buy something and ran into a wall.

The serious part is not that it happens —there will always be impossible searches— but what happens next. Close to half of shops offer no way out: no suggestions, no similar products, not even a link to a related category. Just a "no results found" and a dead end.

A page with no results is not the problem. The problem is that it is an end of the road instead of a fork.

And there is a nuance that gets overlooked: many of those "no results" searches are for products you do have. The customer typed "running shoe" and you have it catalogued as "shoes for running". Or they searched by a reference that only exists in an accessory's description. The product is in the warehouse; what fails is the bridge.

2. It takes too long

This is the least suspected one, because half a second seems like nothing. But in search, half a second is an eternity — and there is plenty of data:

• Amazon measured that every 100 milliseconds of latency cost it 1% of sales.
• Google observed that a 100 ms delay in the search response can reduce conversion by up to 7%.
• Half a second more to generate a results page cost Google 20% of its search traffic.

The reason is behavioural, not technological. A search engine that answers while you type invites you to experiment: you type three letters, you see something, you correct, you refine. A search engine that takes two seconds turns every attempt into a decision —"do I wait again?"— and most people only make it once.

3. It finds the wrong thing

The quietest of the three, because from the outside it looks like it works. There are results, there are products, there is no error. They are simply not the ones the person expected.

The textbook case: someone searches for "LED panel" and right at the top they get the driver for LED panel. It happens because the search engine adds up matches from every field, and the accessory repeats the term in its name, in its description and and in its category, while the main product says it once. The one that repeats it most wins, not the one that is it.

The other frequent case is references. Someone typing an exact code is looking for one specific product, and expects to see it first — not the five accessories that mention it on their page.

What makes a modern search engine different

It is not magic and it is not artificial intelligence. It is four technical decisions, each of which can be explained in a paragraph.

A separate index, not your database

The native search of most platforms queries the product tables directly with LIKE %text%. That cannot be fast, it does not tolerate typos, and it competes for resources with the rest of the shop precisely when most people are buying. A search engine keeps its own index, built for the purpose, and answers in milliseconds without touching your database.

Typo tolerance

"Shose", "lighbulb", "trosers". A search engine that demands perfect spelling is penalising the customer for typing fast on a phone. Tolerance has to be graded: plenty on long words, none on codes and references, where one different character is a different product.

Relevance per field, and adjustable

Not all fields are worth the same. A term appearing in the product's name says far more than if it appears in the description. Being able to raise the weight of the reference, or lower that of the categories, is what fixes the problem of the accessory overtaking the product. And the merchant, who knows their catalogue, should be able to do it without opening a ticket — in Dropfind that is a per-field control in the back office, precisely so that nobody has to be called.

Filters that do not choke the server

Category filters are a shop's other half-search, and they are usually worse than the search itself. Counting how many products are left under each option means going through the attribute tables again and again, and every combination of filters is a different address a crawler can visit. In large catalogues that combinatorics is the usual cause of a shop going down with no apparent explanation. If the counts come from the index and not from the database, the problem disappears. We go into it with numbers in why PrestaShop's filters bring shops down.

What to measure, and what to do with it

A search engine without analytics is a closed suggestion box. These four things are worth more than any general report:

1. Searches with no results, ordered by frequency. It is the shopping list your customers are dictating to you. Each line is fixed with a synonym, a priority rule or a product missing from your catalogue.
2. Heavily repeated searches with few clicks. There are results but they are not the ones people expect. This is where touching relevance pays off most.
3. What gets searched for most. It tells you what to feature on the home page and which categories are worth investing in.
4. From search to order. How much the search really brings in. Without this figure, any improvement is an opinion.

Where to start if you can only do one thing

Look at your no-result searches from the last week. You need no special tool to get value out of that list: half the lines will be products you have, written in other words. Adding those synonyms is the cheapest, highest-return intervention there is in a shop.

Then time your search. Type a word and count. If it takes longer than typing the next letter, you have a problem that no relevance improvement is going to fix.