Mike Cassidy

Mar 14, 2017

Artificial intelligence in eCommerce: How AI provides better search results

As great as humans are, they are no match for the accelerating needs of a modern commerce site search engine.

Consider the vast web and the incalculable number of ways those who use the web describe what it is they are looking for. Pants, knickers, slacks, jeans, trousers, corduroys, dungarees, khakis. And that’s just products.

Think of all the adjectives the world uses to describe products. And then think about trying to keep up with it all. Why, you’d have to be superhuman.

"Unfortunately, your website visitors will use search terms that are different from the product descriptions in your catalog"

RealDecoy’s Richard Isaac said during a December webinar on improving site search hosted by BloomReach. “This is one of the reasons that you can’t just rely on search technologies that have great keyword search.”

You need site search that relies on artificial intelligence and natural language processing to constantly learn from user behavior — because users’ behavior constantly changes.

You need an engine that can automatically match a billion synonyms or more based on context and user behavior. The days are past when site search teams can take it upon themselves to write rules that will deliver the customer searching for “dungarees” to the page of blue jeans that he or she longs for.


Humans can’t keep up with digital consumers’ increasing demands

Some argue that legacy, rules-based site search systems, like Oracle’s Endeca, give marketers more control over site search results. But in fact, the dizzying number of combinations of product descriptions and phrases that customers use to describe products means it is not humanly possible to write enough rules fast enough to ensure that consumers are presented with relevant and personalized results through on-site search.

And irrelevant results mean unhappy customers and unhappy customers mean lost sales and lost loyalty.

A true learning system doesn’t simply “hand the keys to the robots and say, ‘You drive,’” as Forrester’s Mark Grannan recently put it during a webinar. Instead, it allows human intervention when desired, but in the meantime, it automatically offers relevant and personalized results to the customers that an enterprise has worked so hard to attract.

That frees up marketers, merchandisers and others to work on more creative and higher-value challenges and projects. In fact, Staples, the No. 5 retailer on the Internet Retailer 500, moved away from a rules-based system to BloomReach Commerce Search. Staples executives said their digital team was able to handle its manual site-search tuning in a fifth of the time it once took.

Oracle’s own research shows that businesses are ready to embrace the cloud and machine learning to take advantage of the agility and rapid iteration the technologies provide. It’s a reality that has even Oracle’s top executives are urging business leaders to build their strategies around artificial intelligence and machine learning.

To that end, Oracle, which sells Endeca, is focusing on a cloud-based future distinct from Endeca, which was born in an earlier internet era.

Endeca, which was founded in 1999, comes from a heritage of faceted navigation. Its search engine does not have natural language capabilities. It lacks the ability to develop a semantic understanding of the words that customers use when shopping and that enterprises use when they describe what it is they sell.

Why is that important?

Consider a consumer looking for an “iPhone.” If he or she types the phrase into an onsite search engine without semantic understanding, he or she is likely to get a set of results that includes iPhone cases, iPhone chargers, iPhone screen protectors, iPhone adapters and any number of products that are not at all what he or she is looking for.

But a system built around natural language processing and machine learning knows that an iPhone is, in fact, a phone and the search results will reflect that.


I want an iPhone, not an iPhone case

The difference between returning a phone in search results and showing a phone charging cord could be the difference between prosperity and ruin. RealDecoy in its report, “Endeca vs. BloomReach: taking site search to a new level,” cities a 2015 Forrester study that found that 90 percent of site searchers do not read past the first page of results — and that searchers will often just give up if they are frustrated by poor results.

“Usually website visitors are on your site for a specific reason,” RealDecoy’s Isaac said during the webinar.

“Generally speaking, if your search is not good, they stop using, not just your search box, but they leave your site and don’t come back to your site, period.”

During the webinar, Isaac laid out what happens when search goes bad. Take a million new annual visitors, he said. Thirty percent of them will use the search box the first time they visit a site. Eighty percent of those will bounce because of a bad search experience. That means you’re losing 240,000 visitors because of bad search.

Now, multiply that by an average order value of $158 and an average conversion rate of 5 percent. By Isaac’s math, you’re out nearly $2 million and that doesn’t count whatever you spent to attract those lost customers in the first place.

Photo of blackboard by Mr Hicks46 published under Creative Commons license.


2018 Site Search Report

Download the Report
2018 Site Search Report