Google recently launched a new search algorithm that understands the meaning of texts and can answer complex questions. Neural networks and machine learning have already become our real thing and are working for us with success. For example, we have already taught the catalogues of online stores to adjust to the preferences of customers. It's time to teach search engines to analyze a variety of queries.
We understand why it is important to pay attention to the optimization and configuration of search, and how it affects the success of the online store.
Qualitative search - what is it?
According to statistics, visitors using the search, account for about 30% of the traffic of the entire online store. Ideal search should be optimized for the specifics of a particular retailer, be able to analyze and structure requests for more accurate implementation of features: for example, creating a dictionary of synonyms based on user requests.
There are three types of search queries: navigation (iPhone), refinement (iPhone 5s) and point (iPhone 7 128 Gb red).
*Based on Diginetica's analysis of search queries from online multi-category retailers and consumer electronics stores. Moreover, you can get this data from your website with the help of eCommerce reporting.
At the same time, we do not forget that the buyer does not have to speak "in the language" of the retailer, may make a misprint or even speak English badly. Therefore, a good search engine should be able to understand the buyer in the literal sense of the word.
Processing of any search queries can be called quality enough if a number of important criteria are met.
What prevents the search to be qualitative?
Let's return to the imperfection of customer requests. Approximately 20% of requests from online shoppers contain errors. Basically, it is typos, grammatical errors and the request in the wrong layout, and 18% of retailers search can not correct them.
At the same time, automatic error correction can be called one of the most important components of a quality search, because the reformulation of the request by the user himself worsens consumer experience. Moreover, users may simply not know how to correct the request.
Because of the poorly performing search, some retailers' products are "hidden" in the catalog. Many potential buyers use the search not to choose something, but to find a particular product. If the search for the specified model (say, iPhone 7 128Gb red) leads to an empty output, the buyer logically assumes that the product on the site is not.
In fact, there is, but the search engine works only with the name of the product (iPhone), not taking into account other attributes. If the model of the product is specified in the attribute, but not in the name, such a search will return zero results.
How to improve it?
Many retailers have integrated one of the most popular engines, like Solr, Elastic and Sphinx, and do not intend to change it. What are the reasons for that: high cost, IT resources load or the fact that international developer communities support open source technologies? Moreover, there are modules for these searches that allow solving problems of natural language processing (for example, morphological analysis and error correction). But the quality leaves much to be desired.
A good quality search must be "learned" from the user. The best approach to date is considered to be the use of machine learning methods.
What qualitative search gives to the retailer?
Updated qualitative search is able to reduce the number of actions of the buyer on the site, increase conversion and generally improve user experience. Dependence is simple to genius: the better the search, the more people use it, finds what they want and buys.
Image source : searchenginejournal.com