Source: amazon.com/scout

Is Amazon on the right path to improved product discovery with Scout?

Jeff Bezos has said, “To invent you have to experiment and if you know in advance that it’s going to work, it’s not an experiment.” By that definition, Amazon Scout is an enormous experiment. 

Scout (amazon.com/scout) is a new visual search engine within Amazon. It’s aimed at categories where visual appeal is a big part of shoppers’ selection criteria, so furniture, women’s shoes and lighting are early subjects. Shoppers choose a product subcategory and then give a thumbs up or down to each image. It seems like a simple and intuitive method of culling preferred items from a list of possibilities, but without a rework, I believe it will be a failed experiment. 

The system utilizes machine learning to make replacements to the displayed assortment based on similarities or contrasts to the user-ranked item. The more rankings, the more tuned the selection until, in theory, the product choices displayed are largely purchase candidate items that meet the shopper’s desires.

The major breakpoint is that humans rarely make absolute judgements when it comes to shopping, especially when there is an emotional component to the goods being considered. Scout forces the normal expanse of gray area decision-making into one of black or white, which may result in unwanted results, time/effort wasted, disappointment, frustration and lost discovery opportunities. 

amazon scout your likes
Source: amazon.com/scout

I see Scout further hindered by shortcomings including: 

  • An inability to ascertain the reason for a like or dislike;  
  • Potentially insufficient/incorrect metadata to effectively support the replacement process; 
  • Seemingly no input from prior purchase history; 
  • Inconvenient access to supplemental product selection attributes, including ratings and reviews; 
  • A UI design fail that expects users to shop one way for visually oriented items and another for other items. 

Scout does allow users to see a “Quick View” pop-up of product attributes and within that “See More Details,” which opens the standard product description in a new tab. They can also rank a sub-selection of related products. Surprisingly, ranking new products does not change the main results page. It does populate “Your Journey” history, allowing deselected items to be either viewed in a new window or removed from disliked status. Whether it resets the main results or not, depends on where the product view originated. 

If all of that sounds fairly complex and cumbersome, it is. If Scout is a washout, not all is lost. In fact, Mr. Bezos has said, “If you want to be inventive, you have to be willing to fail.”

BrainTrust

"Visual search will become far more valuable and significantly faster than keyword search as the technology and image libraries become more refined."

Adrian Weidmann

Managing Director, StoreStream Metrics, LLC


Discussion Questions

DISCUSSION QUESTIONS: Do you think Amazon is on the right path to improved product discovery with Scout? Does it matter if Amazon fails with Scout if it’s a learning experience?

Poll

16 Comments
Oldest
Newest Most Voted
Inline Feedbacks
View all comments
Bob Amster
Trusted Member
5 years ago

In the case of Amazon, the answer is simple. If Amazon Scout misses the mark with consumers, Amazon will either throw money at fixing it if it is fixable, or will scrap it in favor of some future equivalent. They have the money, they can spend a small part of it on testing.

Chris Petersen, PhD.
Member
5 years ago

Thumbs up or down sounds a lot like swiping left or right. Might work for some very visual categories, but users will quickly tire of the novelty if it doesn’t add relevance to their search. I agree with Ken that this initial form factor does not sound too promising. However, the initial launch does not always forecast the final iteration. If there is one thing that Amazon is good at it is “iteration” and learning from data. Imagine the amount of data that can be collected from Scout and stored by individual customer.

Art Suriano
Member
5 years ago

It’s nice to have unlimited funds and be able to spend money on anything you want. Sometimes I wonder if Jeff Bezos does that more for the publicity rather than the result. I don’t see Scout as a big win but instead another wasted attempt for an idea driven by expensive and unproven technology. The article used the keyword when pointing out the pitfalls of this concept: Inconvenience! I don’t know when retailers across the board are going to wake up when it comes to technology and realize what is right for them may not be good for the customer. I don’t see this as successful, and I’d rather see Amazon invest their money in something more practical.

Adrian Weidmann
Member
5 years ago

While Amazon’s Scout may have its challenges, I believe that through experimentation, future iterations will make this visual search engine successful. There have been numerous times where this capability would have saved me hours trying to find replacement parts and allowed me to narrow my focus based on my aesthetics and preferred styles. Visual search will become far more valuable and significantly faster than keyword search as the technology and image libraries become more refined.

Ananda Chakravarty
Active Member
5 years ago

Shoppers are not looking for pictures. Why would a customer want another step in the buying process to click on thumbs when they can scroll through these on cat pages. Sounds like a glorified recommendation engine for images. The actual selection of products and purchase of products would have stronger influence than a thumbs up/down on the images if the end goal is product sales. There are so many systems that can enable the “machine learning” that they’re attempting to implement.

Ken Lonyai makes some great points as to why it won’t succeed. Where it has a value is in culling poor quality, removing errors such as the wrong picture for the product, or too much duplication through crowdsourcing.

What they are doing is interesting though, because the next stage of images is where it gets juicy — images within context, like a snapshot of your favorite celebrity wearing jewelry or a watch and video — think Home Shopping Network.

As for now, Google does visual search better.

Ryan Mathews
Trusted Member
5 years ago

I think the same criticism — to some degree or another — can be leveled at all of Amazon’s ML functions, so I’m not sure if Scout is any more or less guilty. I have been an Amazon customer almost from Day One and they still don’t get my book recommendations right. As Ken points out, there is a problem in not nuancing selection criteria and metadata. For example, on the traditional Amazon side, I bought a new book on William Burroughs. The algorithm read him as a gay author and so flooded my recommendations page with LGBT literature suggestions. At the top of the list? The Gay Science by Friedrich Nietzsche, which of course has nothing to do with being gay or William Burroughs for that matter. Now over the course of the years, I have purchased all kinds of books on and by the Beat Generation, but Amazon can’t seem to figure out that that is where the Burroughs book should be classified. All that said, they are doing pretty well with failed instruments. Maybe I should read more Nietzsche after all.

Doug Garnett
Active Member
5 years ago

Amazon has found the wall they can’t get past: While it’s great to buy things online, online sites are quite weak for shopping or browsing. These are fundamental limits due to technology.

I recommend the folks at Amazon read Edward Tufte’s work. Digital devices are all incredibly limited when it comes to discovery as he documents — and mere replacement of words with photos isn’t going to change that. (Scout appears to be a minor step forward with claims to do far more. It won’t.)

Ryan Mathews
Trusted Member
Reply to  Doug Garnett
5 years ago

Doug — I actually disagree, at least over the long run. Written words, I’m sad to say, are being replaced. Look at voice activation of everything from your television remote and doorbell to your thermostat and car. Now, look at texting with its automatic suggestion for symbolic replacements for words. Type in birthday and you get a prompt to substitute a picture of a cake, etc., etc. And then there are those damn Emojis.

So, technologists are training us to communicate more and more in non-alphabetic symbols — words, after all being symbols in their own right. So, while I am a huge fan of Tufte’s work, and conceding he still may be right today, the day is coming soon when pure visuals — say a three dimensional “mini movie” that uploads images from your home and allows you to test out various chairs in digital situ may be preferred over words. We aren’t there yet, and we won’t be there tomorrow, but we will be there sooner than most of us are prepared to admit.

Doug Garnett
Active Member
Reply to  Ryan Mathews
5 years ago

Appreciate all that. The limitations I see are more serious than “words vs pictures”. When we see furniture, for example, online, there really can never be the perspective needed to sense whether it’s right for our world. Even at Ikea they use their displays to shift the appearance of furniture to make it appear more substantive than it is.

But browsing in a store gives me all my sense and the ability to walk around something while touching it, to put something next to it to compare, to get accurate color and texture (subtleties online photo’s can never carry), etc.

The part of Tufte I was referring to is his critique of the communication limitation of digital displays. Color is never correct. There is a dot limit that can never get things. It’s a 2d display — so just like in paintings, any sense of a 3rd dimension is an optical illusion.

Things probably can be better than I expect. But Amazon has had a long time to work on it (say the last 10 years) and their website is no better at solving the browse problem than it was 10 years ago.

Appreciate the thoughts…

Shep Hyken
Active Member
5 years ago

AI is getting better and better. Even just a few months of development can make a difference. If anyone can make Scout work, Amazon has a shot at it. And Amazon is quick to walk away when things aren’t working. Bezos and team will know when it’s time to move on.

Carl Van Ostrand
Carl Van Ostrand
5 years ago

I might be in the minority here, but I feel like this has some legs. I played around with it (searching for a patio bench) and it did fairly well at creating a set of benches that I liked. I am not sure if it was actually a more efficient way to search than a more standard filtering process, but I didn’t hate it, in terms of a shopping experience.

Then I saw, “women’s shoes” as a category, and I thought, “oh my, my wife would love this.” In our heads, in some categories more so than others, I feel this is akin to our process as we walk down an isle in a store — it’s a yes/no/maybe decision set.

Cate Trotter
Member
5 years ago

Hmmm. I could see this idea working as some sort of first-pass whittling down process, e.g., if I searched for “red coats” I could quickly go through the results giving a thumbs up to those I wanted to know more about and a thumbs down to those I didn’t like, and then be left with just the ones I had thumbed up which would give me a smaller pool of options to review in more detail before making a decision.

The fact is that online shopping gives rise to a huge amount of choice and we know that too much choice can be off-putting. Customers get paralyzed and they don’t purchase. We’re also becoming much more visually driven because of the like of Instagram, Snapchat, other social media etc. People do want to see good photos of anything they consider buying online. In its current iteration Scout probably isn’t going to change the world, but I wouldn’t bet against Amazon refining it and making it something that could.

Liz Adamson
5 years ago

A visual search feature like Scout is due to be tested. And Amazon has the capabilities and resources to test and iterate this concept. While the technology may still be rudimentary, if there is a way to fine tune the AI to respond to customer visual preferences, it will be a powerful search tool. Relying on keyword search will only get you so far; it’s easy to search for “blue pillow” but harder to describe the shape, style, design, materials, etc. all in a keyword search query.

Cynthia Holcomb
Member
5 years ago

Sorry Amazon. Scout does not evaluate the 3-dimensional, individual HUMAN SENSORY and emotional preferences of fit, look and feel. Humans purchase and keep a product when the product’s attributes in combination match the buyer’s individual sensory preferences. Only then is there an emotional home run for the buyer. Here is a link to my comments on the Scout announcement earlier this week. Read if interested in a deeper context on why the Scout solution does not solve the sensory gap of digital shopping.

Ken Morris
Trusted Member
5 years ago

Again, another example of Amazon testing a new concept — an intriguing use case for AI. Conceptually, I think it is a great way to help consumers narrow down choices for products that are visual and include lot of options. It is also a form of entertainment, as it is fun to see what pops up to replace the item you give a “thumbs down.”

Unfortunately, I think Amazon may have launched this feature a little prematurely as they appear to be missing some key product categories and it seems like the products within the categories are a little limited. For example, in the furniture category, they don’t have dining tables or dressers to search. For a more accurate test, I think Amazon will need to make the options more robust. However, kudos for Amazon to jump-start this feature and we will likely see other retailers testing it soon.