One of the benefits of running a PPC agency is that I get access to awesome data. See, as an agency, we have access to the Amazon Advertising API which provides amazing insights into how Amazon’s algorithm works (albeit in the form of weird files that are super hard to open and then when you do super hard to convert to a spreadsheet….but I digress).
After about five days of coffee-induced nerding out over dozens of excel and notepad++ files, I’d like to share what I have found. A peak “under the hood” so to speak of Amazon’s machine learning algorithm.
I believe understanding how it works helps us to make informed decisions on how to launch and grow our products on the platform. So I’ll dive right in, but first I’d like to give a shout out to Tarik Berrada and Mina Elias for their insights (ya’ll dont’ think I figure all this stuff out on my own, do you?)
Amazon Suggested Keywords; An Insight Into Amazon’s Machine Learning
The initial question was, where does Amazon get suggested keywords from for a brand new listing?
We can speculate that suggestions come from category experience, listing activity, browse paths, and a million different other places for established products, but what if it doesn’t have any activity or past?
I analyzed three brand new products of our own first. These products were just launched and we immediately turned on PPC for them using our AI + Managed service.
The first suggested keywords coming through the API for two of the ASINs had only eleven suggestions. The third ASIN got about twenty-four. These numbers remained the same or decreased over two weeks.
That was interesting.
Then, exactly two weeks later all three ASINs saw a surge in suggested keywords. The first two jumped up to over thirty and the last ASIN went to over forty. Then, suggestions remained constant or dwindled.
After conferring with our PPC managers, it looked like this matched their experience; that in the first few weeks the largest keyword dumps happened.
So it would appear that the way the machine learning algorithm works is to grab as many keywords as possible and then refine from there (this is likely how and why the honeymoon period happens).
So the next question was, what caused that surge two weeks later? Was that just a normal “honeymoon period” action, or was it triggered by something else?
It turns out that for all three ASINs about a week prior the ACoS target was increased…and then about two days prior spend dramatically increased. This would indicate that spend had a direct impact on the amount of keyword suggestions provided by Amazon.
This rabbit trail led me to wonder how suggested keywords fluctuated for older, more established products as well.
Many, ranging in suggestions from forty to over one hundred, didn’t change over time. I looked at reports a week old, a month old and two months old.
However there were slight changes. One ASIN in particular caught my eye. For some reason, it only had a single keyword suggestion. Then, inexplicably, about a week or so later, it had seven.
What caused that?
After consulting with my PPC expert friends, I learned that it appears a relevance boost, or “mini honeymoon” occurs at the beginning of any campaign.
This means that the largest keyword suggestion dump happens in the early stages of a listing, during which time the algorithm is learning all it can and casting a wide net.
However, after it has become established, refinements are made by lowering relevance for many keywords and adding new ones only when an event triggers it.
But the biggest ‘AHA’ about all this is what event triggers it; Adspend.
You can impact keyword suggestions by increasing your budget. And this makes logical sense, considering that more budget means more exploration, learning and impressions, which inform the algorithm.
This also falls in line with anecdotes about people launching successfully with ONLY PPC and no other promotions. By being aggressive, they forced the algorithm to find keywords, or otherwise cast a larger net, to spend the entire budget. In doing so, more impressions and therefore higher relevance for more keywords was the byproduct.
**Ok, so you’re saying that we should turn on PPC immediately for new products to help gather data? This isn’t something we haven’t heard. Tell us something we DON’T know…
It’s All About the Money 🤑🤑
In case you haven’t caught onto the implications here, I think this is pretty huge.
Sure, turn on PPC for new products. That’s nothing new. However, what we see here is EVIDENCE that you can sway the outcome of your listing dramatically with PPC.
Basically, you can BUY relevance!
Let’s look at what we know:
- We know impressions (at least at first) can boost relevance
- We know more budget will often times get more impressions
- Now we know that this impacts initial keyword suggestions
E.g. If you launch with an aggressive PPC budget, you’ll force Amazon to find more keywords during the honeymoon period to serve impressions for, and thus will become relevant for more keywords (expanding your visibility, and rank).
💥BOOM!
Sure plenty of experts in our space have been telling us to turn on our PPC immediately. But now we can see the evidence of why it is important. What exactly is going on. No more vagaries about “so you can give Amazon more data.”
Now we understand this might be a time sensitive issue. So, it is definitely important to come out of the gate with a new product employing aggressive PPC.
However, you’ll want to do so intelligently (can’t just throw a boat load of money at the wall to see what sticks). I’ll go over a strategy for this later.
Amazon’s Latest Algorithm Update
In the midst of diving into the inner workings of the elusive “algorithm” I was also led down another rabbit trail of data (shout out to Ted Charon for this one). It was brought to my attention a section in Category Listing Reports called “Product Feed Type.”
This is essentially the template type, or subcategory, of the product. What was brought to my attention, though, was that a lot of these feed types seemed new, and there weren’t any inventory templates for them.
Whaaaa?
Amazon was making a recent change to the way it categorizes products (very important for relevance, indexing and ranking) and inventory files didn’t even exist yet for these changes.
So what are these new product feed types?
They appear to be refinements, or otherwise more narrowly defined categories. For example the wider product feed type Sportinggoods used to encompass all products in the category Sports and Outdoors.
However, refinements have been added, such as Recreational Balls, Golf Equipment, Orthopedic Braces, etc. These provide much more fine tuned relevance to products, as they dictate what keywords a product can be relevant for.
My own client’s products have found a home in a seemingly new product feed: outdoorrecreationproduct
This is an awesome insight into how Amazon is defining keyword relevance for products. However, if the inventory templates don’t exist for these products, how can flat files be used anymore to correctly upload and categorize new products?
It appears Amazon has finally solved the issue.
Before, if you Googled “inventory file template amazon” you’d be brought to a page with each primary category listed and where you could find the inventory template and browse-tree guide.
However NOW all links and pages redirect to a new page that directs you to generate the file yourself.
Investigating further, now I see that from the “Add a product via Upload” page in seller central, you can generate your own template based on the subcategory or item-type-keyword.
Ok, but back to the refinements…
So what does this mean?
It almost seems as if Amazon is taking the control over how they categorize our products on its platform out of sellers’ hands.
Not entirely, since we still categorize the item at first, but between Amazon changing categories without informing sellers and refining subcategories to this degree, it seems like the direction they are going.
Looking past the negative aspects of losing control over factors that impact visibility for our listings, this can be a good thing. With Amazon’s machine learning algorithm becoming “smarter” by learning and training with more data, we can rest assured that most of the time relevance should be accurate.
Connecting the Dots
- Ok, Amazon is refining how it categorizes products.
- This therefore impacts how keyword relevance is issued to listings.
- Amazon also provides keyword suggestions.
- That suggestion list can be impacted by ad spend.
What can we do with this knowledge?
Let’s start with the obvious.
✅ First, it means you need to have an extremely optimized listing from the beginning. I know everyone says this, but a lot of sellers take it for granted as something they can always fix later.
Don’t underestimate the importance of a well optimized listing right from the get-go. This affects keywords suggestions in the very beginning, not to mention also impacting whether you will stay in your chosen subcategory (because if Amazon determines another category is more relevant for your product, it will move it).
Also, do your research and make sure you categorize your product properly initially. Research your browse-tree guide and see what subcategories exist for your type of product.
✅ Next, launch with aggressive PPC from the start. Now, I don’t suggest you turn your budgets up to the max and set-and-forget your auto campaigns. That is probably not efficient use of precious startup funds.
Personally (but I am biased) I think the better strategy is to employ AI to refine and scale bids rapidly based on data (which is how our agency at Signalytics handles it).
But whatever your strategy, it is important to “buy” as much relevance as possible early on, to ensure your product has the greatest chance of success through a wide net of keyword visibility.
✅ Lastly, don’t leave it all up to PPC. By employing promotions and off-Amazon traffic, you give the algorithm even more conversions and data to work with, especially in the first few weeks of the life of a listing.
This will reduce the amount of time necessary to gain relevance for more keywords. It will also likely decrease PPC costs due to conversion improvement.
Ultimately the name of the game still is visibility, which for good products, will equal sales. However, understanding how the algorithm does what it does, and a little speculation as to why, will inform your launch strategy.
As it stands, the formula for success appears to be:
(Optimization + PPC + Inside/Outside Traffic) * Honeymoon/Discovery = Relevance = Keyword Visibility
I hope this equation serves you well. 😇