Mercent’s Dan Myers: Amazon Search Depends on Relevance

While content26 isn’t an SEO firm, we are focused on making ecommerce product content as successful as it can be. Part of that success hinges on Amazon search. dan-myers

In pursuit of more information about how Amazon site search works and what role keywords and other factors play, I spoke to Dan Myers, director of marketplace services at Mercent. 

Myers joined Mercent, which helps retailers optimize advertising content and merchandising offers to maximize visibility across different channels and platforms, after spending over a decade at Amazon.In this interview, he explains precisely how site search on Amazon works and gives tips on how vendors should approach keywords.


Product Rank Has a “King of the Hill Effect”

content26: What factors does Amazon take into account when they rank products in their search results?

Dan Myers: It’s a proprietary algorithm which they title relevance, with a capital R. The secret sauce changes a lot, and they don’t divulge it publicly. However, it’s largely based on sales and clicks. Typically sales, or conversions, are a result of people clicking and seeing an offer, usually with good shipping and a competitive price point. That begets more clicks as the product makes it way up into search. There’s a king of the hill effect for items that have a history of clicks and conversion–one would be hard pressed to knock them off the top of the list because they show up in search, so more customers gravitate toward them.

That all-product search page, which consumers see once they have entered terms in the search bar on the gateway page, is a heavily trafficked page. People often hang around on that first or second results page and make click decisions from there. The data suggests a smaller percentage of customers leverage the left-hand nav elements, but those are obviously critical for certain categories, such as Clothing and Shoes.

I call it the bag of words approach: brand name, title, and five search keyword fields get thrown into a bag of words that their algorithm will use to provide results for relevant products.

content26: So factors like how much time is spent on a page and unique content don’t matter that much?

Dan Myers: Not necessarily, because what gets served in search is a product of keywords, title, and brand. I call it the bag of words approach: brand name, title, and five search keyword fields (where you can in fact list multiple keywords per field) get thrown into a bag of words that their algorithm will use to provide results for relevant products.

If an item has no history of sales or clicks, it’s likely not going to show up in search and certainly not going to show up on that first page. However, as it gathers one or two clicks, or even as related products gather one or two clicks, it stands a better chance. What are not figured into the search algorithm are words in the bullet points and product description.

Now, having said all that, this is based on data I put together years ago. Amazon may decide to use bullet points in the bag of words approach. But as far as I know it’s brand, title, and search keyword fields.

content26: Could you say more about how keywords work?

Dan Myers: In every listing there’s metadata attached to the item, SKU, and ASIN. One of the metadata is the search keywords. There are five fields in Seller Central. You can add more than five separate terms by bunching terms together in a comma-separated list in one field. Those things get figured into the search algorithm.

These are space sensitive, so if it’s a two-word term in a list like “car toys, stereo, tape deck” the customer would have to type in “tape deck” (with the space) in order for that product to get served. But I think when it comes to that bag of words algorithm, brand and title go pretty far. I think that’s a result of the large catalog–not every ASIN has search keywords filled in, but brand and title are required fields, so Amazon relies more on those.

The changes you see are display of product search and maybe some of the ordering, but I think the fundamentals of what data they look at haven’t changed much.

content26: Do you think Amazon has changed the way it does site search since you worked there, or is likely to change it, or do you think this formula works well?

Dan Myers: I think the formula works really well for Amazon. The customer behavior I’ve noticed is not drastically different than it was years ago. What probably gets changed is the way they serve recommended products. They obviously have tons of community data regarding similar items, like items, or “Customers who purchased this also purchased this” type stuff.

So if an item is king of the hill but is out of stock or has a crazy price point, I could see them saying, “This is not going to be a good, relevant offer to put at the top of the search results,” and tweaking the algorithm. They’ll instead choose to show other items that have a better chance of getting converted, despite the fact that this item has been in the number one spot for a long time.

So the changes you see are display of product search and maybe some of the ordering, but I think the fundamentals of what data they look at haven’t changed much. Title, brand, keywords, and then, of course, historical relevance of products–does it get lots of clicks and lots of sales?

content26: It sounds like there’s no direct correlation between enhanced content and ranking, but if enhanced content leads to increased sales, that leads to better ranking, right?

Dan Myers: Yes. The most valuable real estate, far and away, is the detail page. Regardless of how the customer got there, the most critical piece of real estate and the one that’s most easily affected is that detail page where the buying decision is made. Enhanced content goes a long way there.

I don’t have any data to back this up, but when I was [at Amazon] we would see better conversion from pages that had the following: multiple offers, including third-party offers; multiple images or alternate product views–rich images that were zoom enabled; a coherent title that met the style guidelines for the category; bullet points; and a product description that was also coherent, not replete with high ASCII characters, malformed HTML, or phrases where clearly the seller thought the product description was feeding the search algorithm and pumped it full of keywords or something awkward.

The numbers vary, but we used to quote that 65-70 percent of sales were coming through search on the site.

All that being equal, the biggest driver of all is the landing price–product price plus shipping. Free shipping goes a long way, but that’s a tough ask for merchants. All things being equal, that’s when the conversion happens.

content26: Do other retail sites, such as Target, Walmart, or Toys R Us, work similarly to Amazon with their search?

Dan Myers: I tend to think they all have similar relevancy models where they serve up the products that historically customers have gravitated toward. The whole point is that while it’s great to have a wide selection, it does not make sense for your search algorithm to be random. You need to serve up the stuff that historically customers have gravitated toward and purchased. Otherwise, you’re just wasting time.

The numbers vary, but we used to quote that 65-70 percent of sales were coming through search on the site. It doesn’t make sense to do something vastly different.

The most critical piece of real estate and the one that’s most easily affected is that detail page where the buying decision is made.

content26: Vendors do have some control over keywords. How do you recommend they do keyword research for Amazon?

Dan Myers: There is a growing wave of self-serve operations for vendors. Amazon’s buyers, historically, would create the ASINs and detail pages and would populate the metadata, search keywords, title, and so forth. All of that has been getting pushed to vendors for a long time.

The advice we see in Seller Central, for merchants, is to include things like common misspellings or alternate spellings. Stay away from competitor names; that’s disingenuous and messes up the search experience, and they’ll take punitive action. Include simple keywords that may make sense in the context of a bulleted point or title, but are also useful in a search query, like “cotton shirt.” And, to your point, keywords that are successful on your own website or your AdWords program should definitely be imported into search keywords.

Daniel Myers

Dan Myers is the Director of Marketplace Services at Mercent Corporation, a Seattle-based SaaS firm which provides tools to accelerate overall retail revenue growth and profitability. Dan started at Mercent in 2011 after a 13-year stint at Amazon as a product manager in both the marketplace and retail businesses.