Understand the algorithms of Sponsoreds Automation Software

Automation for your Sponsored Products campaigns is enabled if you’ve set up products in our Amazon PPC Management Software, and the requirements for the software were followed.

There are main algorithms where we use Big Data, which perform changes to follow your business goal (4 setup strategies). Here we covered all the changes in the Sponsoreds algorithm.  

⚠️ NOTE: every time we use “keyword” or “keywords” we mean both “keywords” and “Product Targetings” targetings on Amazon unless otherwise stated.

Thanks to that, the text easier to read and shorter.

What will change?

The algorithm now better recognizes situations where it is worth changing the bid for a keyword or Product Targeting. The algorithm takes into account the selected strategy for a product, as well as the "quality" of each of the keywords compared to the rest.

What is affected?

Thus, selecting a strategy that has a goal to make PPC profitable (PPC Profit Growth, Boost Overall Profit) will force the algorithm to always lower bid for keywords with ACoS higher than calculated Target ACoS. Also, the algorithm won’t increase bid for keywords if their ACoS is not much lower than Target ACoS.

On the other hand, selecting a strategy that is focused on product growth (Revenue Growth, Product Launch) will make the algorithm more aggressive. It will increase bids more aggressively for keywords with ACoS lower than Target ACoS. Also, the algorithm won’t decrease bids for keywords with ACoS higher than Target ACoS if they are producing high sales or conversions compared to other keywords.

What will change?

Previously our algorithm used something we call “sufficient clicks” to determine WHEN it should make a change to a keyword (whether to increase bid due to good performance, decrease bid or pause due to bad performance, etc.).

Sufficient clicks were calculated by using the product average conversion rate and “confidence level” parameter that determined how confident we want to be that we have enough data to make a change.

Even though “sufficient clicks” approach worked well for products with structured PPC, it fell short in other cases, like:

- start of product advertising, product launch. In this case, we either have no data at all or have a meager average conversion rate still. That meant that the algorithm required an unreasonably high amount of clicks to reach “sufficient clicks” due to the low conversion rate. So optimization changes were performed less often than it should have been;

- high CPC keywords that are not related to the product being advertised. Upon reaching “sufficient clicks” such keywords had already spent too much money due to high CPC, well beyond any reasonable margins based on product price, product margin, and strategy used for the optimization of PPC.

To improve the algorithm’s decisions in these and other cases, we introduce the next main changes:

What will change?

“Sufficient spend” metric to work alongside with “sufficient clicks”.

What is affected?

Similarly, it would determine whether keywords spent enough money to acquire enough data to make a change. “Sufficient spend” metric is calculated based on product price, product margin, average product order size, product conversion rate, selected optimization strategy, and confidence level.

Then keywords are checked against “sufficient spend”:

if their cost is higher than “sufficient spend” then algorithm checks if some action is needed.

What will change?

The algorithm now applies softer changes to bids for keywords. Also, a formula that is used to calculate bid multiplier now consider:

- match-type of keyword or Product Targeting type - “wider” type means a softer and smaller change in bid (the phrase is wider than exact, Category is wider than ASIN);

- “quality” of a keyword compared to other keywords (“quality” is determined as a conversion of the keyword compared to conversions of other keywords). Higher “quality” means a smaller change to a bid when ACoS is higher than Target ACoS;

- selected optimization strategy for a product.

Selecting a strategy that has a goal to make PPC profitable (PPC Profit Growth, Boost Overall Profit) will force the algorithm to perform stronger changes to bid when ACoS of a keyword is higher than Target ACoS, compared to a strategy that is focused on product growth (Revenue Growth, Product Launch).

What will change?

The algorithm now analyzes your current PPC campaigns for the selected product and pauses the duplicate keywords or Product Targetings (ASINs, Categories, Brands). The duplicates are defined only between the same match types (exact vs. exact, phrase vs. phrase, broad vs. broad) using next Amazon rules:

- the plural and singular form of a word is the same word;

- articles, prepositions, etc. are ignored.

What is affected?

The algorithm will find such keywords and pause them. If the algorithm finds two or more duplicated keywords, it will pause the keywords with the worst performance (e.g., only one with the best statistics will stay enabled).

The algorithm will also check for negative keywords when determining duplicate keywords. If negative keywords prevent competition between possible duplicates, then the duplicates remain untouched.

Product Targetings are being processed in a simpler way. Two or more fully identical Product Targetings (same ASIN, or same Category with additional settings) are marked as duplicates. Product Targetings with the words performance will be paused (e.g. only one with the best statistics will stay enabled).

The algorithm also checks customer search terms and pauses keywords of the same match type (except broad, since broad keywords often work for identical customer search terms) if they have worked on the same query.

What will change?

The algorithm now has stricter rules when identifying whether some keywords should be paused due to bad performance.

What is affected?

The algorithm will now more often make a decision to lower bid instead of pausing a keyword.

What will change?

The algorithm now optimizes Auto campaigns that do not have any negative keywords.

Previously, the algorithm ignored such campaigns because it treated them as our special “Auto CTA” campaigns from Zero To Hero, which should not be touched (bids should not be changed).

What is affected?

Now algorithm checks for (Auto CTA) tag in campaign name and ignores only those campaigns.

The algorithm now uses information obtained from Auto campaigns that do not have any negative keywords when searching for new customer search terms.

Obsolete auto campaigns that were created before Nov 14, 2018, are now also used when searching for new customer search terms.

What will change?

The average conversion rate for a product is replaced by “normalized CVR” - metric that is calculated separately for each match type or Product Targeting type (exact, phrase, broad, ASIN, Category, Brand and auto).

When no conversions data is available for the product (so no CVR is available), we use average numbers calculated based on 452 million data points. As PPC for a product starts to work and collects data, we smoothly transition into using the product’s specific numbers.

To calculate “normalized CVR” for different match types and Product Targeting types, we use formulas with weighted sums and known coefficients for conversion rates between different match types.

What is affected?

Since “normalized CVR” is used both by “sufficient clicks” and “sufficient spend” it means that both these metrics have slightly different numbers for each match type or Product Targeting type. This approach yields better results since there is a substantial difference in the way different match types and Product Targeting types do work.

What will change?

The algorithm now performs operations more often when the keyword has conversions (so ACoS can be calculated), and tries to be more confident when performing an action for a keyword without any conversions.

What is affected?

Adjusted confidence levels required for the algorithm to make a change for different cases.

For example, to pause a keyword due to bad performance, it should be more confident than in the case of a slightly increasing keyword’s bid due to good performance.