How people discover content on e-tailer sites

Three main mechanisms allow users to go from one page to the next.

• On-site search: the search bar that 99% of sites have for searching product items.
• The recommendation modules such that often are at the bottom of product pages (“people who looked at this product have also considered these products”)
• Category browsing: some people still use only their mouse to drill down from categories into subcategories until they find what they need.

Putting an exact figure on the respective share of each three of these mechanisms is extremely difficult. Of course, the split varies tremendously between sites, as they will all put different levels on emphasis on different types of navigation.

Jumpshot has made some measurements for Amazon.com specifically and found out that 90% of product page views came from on-site searches.

Whether it’s on-site search, recommendation engine or category browsing, it all comes down to lists of products. Each site has a different method to come up with these lists, but whatever they are, it always comes down to rules in a search algorithm.

E-Commerce search algorithms

Search algorithms are mathematical formulas that create an ordered list of results for each search terms (or search tokens) that they receive. Each search algorithm is defined by the factors that it takes into account to pick and rank the results.  If we can understand which settings an algorithm is looking at, then we can figure out some tactics to improve the ranking of our products.

Algorithms come in all sorts of shapes and colors. An efficient way to look at them is to split them into three groups:

• Text-based algorithms: they rely only on text resemblance between the search terms and product page texts.
• Business-based algorithms: they add a layer of business intelligence to rank products based on the expected business outcome.
• User-based algorithms: they rank results based on known user parameters and inferred preference.

Text-based algorithms

Creating and optimizing computer programs to recognize matching strings of text in large databases in a mathematical field of its own. We do not need to delve into such details though. What text-based algorithms do, as far as e-commerce is concerned, is surface product pages that include the inputted text. They feature different levels of advancement:

• some will only find exact matches, while others support typos, plurals or the same words in any order,
• some just look up the words in the product titles, while others index the descriptions, specs, categories, etc.

Business-based algorithms

These algorithms introduce some layers of intelligence in the way that they rank the results. They pick results to display to the customers based on some business factors.

As an e-tailer, your goal is to make sure that as many searches as possible result in a sale. The way to do that is to display relevant products, that sell better than others. It means that your algorithm will usually select all relevant results, then rank them. The algorithm is “fed” some additional variables to come up with a score. These may include:

• Sales velocity
• Average review score
• Number of pictures on the product page
• Average search result CTR (how often is this product clicked on when displayed)
• Current stock status (show out-of-stock products last)

Advanced business-based search algorithms like Amazon’s factor in dozens of variables into search rankings.

User-based algorithms

These are the algorithms that are capable of tailoring search results based on what they know from the customer. A straightforward example might be:

• a shopper searches for “shoes” on an apparel site,
• the shopper only clicks on male shoes in the search results,
• the shopper then searches for “trousers,”
• the algorithm displays all male trousers at the top of the results.

In that case, the algorithm figured that the shopper was probably more interested in male clothes. It uses this knowledge to increase the relevance of results for this shopper. Other user-based mechanisms may include segmenting based on browsing behaviors, the device used, past purchases, etc.