More people in the eCommerce world are becoming familiar with the tantalizing idea of “learning search.” Learning search promises to increase conversions by delivering results based on data “learned” over time. The sales pitch is that the technology works like a human brain – constantly watching, learning and improving. Sounds interesting, but does it deliver on this intriguing premise?
According to EasyAsk product experts, that answer is more “no” than “yes.” The first major hurdle is in understanding what “learning search” really does. Despite “learning” in the name, the search function is not learning and adapting based on the visitor’s past choices and behavior – which is a common misconception. It’s just pushing the most popular items to the top of the search.
At its essence, learning search is basically a popularity poll, ranking the most asked-for and clicked-on items at the top of the search results, regardless of nuance or relevancy to an individual searcher’s parameters. It’s really nothing more or less than an aggregator, putting the items purchased the most at the top of the search results, regardless of whether they are really what the customer is looking for.
Let’s say a customer is looking for a cold-weather winter coat. Using EasyAsk, a shopper could put in those exact search term, or any variation, such as “warm winter coat.” EasyAsk’s natural language technology actually understands the meaning of the words, and deduces the intent behind them. The learning search software will not understand any of these qualifiers to “coat” unless they part of the catalog product description, and instead simply deliver the most popular results for the word “coat.” If that happens to be a rain coat or other lighter outerwear, the results will clearly not be relevant to what the searcher is looking for.
Another issue with learning search is that is does not understand or account for typos or misspellings. If, for example, I hurriedly typed in “coal” instead of “coat” I would not get the results I was looking for. I would probably not get any results at all, unless the retailer surprisingly also sold coal in addition to coats.
If names of products are not English words, that can also be a serious issue.
SEO is another problematic area for so-called “learning search.” A fundamental flaw with most platforms which use this technology is the use of what are called sub-domains. This takes the visitor off of your website, degrading your SEO. For example, if I visit the mythical site www.store.com and enter in a search term, I will be directed to a subdomain, such as search.store.com. If I click on a product, I would then be routed back to the website, but this is not what you invested all that money, sweat and tears on SEO for. There are some exceptions to this rule of thumb, but they are generally costly to implement.
So despite the name, there are smarter technologies out there than learning search, including natural language search.