Overengineering Our Amazon Advertising Optimization Process

Many interested parties have asked how we’ve achieved overwhelming success with our clients’ Amazon Advertising campaigns, despite the increasing level of competition for placement, and increasing bids over time across the board.

In a nutshell: our algorithmic, self-healing processes which, as of now are a combination of statistical algorithms and human analysis, contains roughly 500 rules (and growing). Some parts of the engine are even self-adjusting. Cue in theme music for angry AI gone wild!

When we originally developed our engines in August 2018, there were roughly 50 decisions which make up the straightforward routine optimization that should be universally executed on any Amazon Advertising account. Any campaigns, from the poorly structured to the optimal, from the underperforming to the highly profitable ones, benefit from having these fundamental optimizations applied to them on a recurring basis, ideally weekly. However, since then, the decision set has grown 10x due to several factors, chiefly due to real-world needs: where campaigns are very poorly structured yet need to be sustained for the foreseeable future due to their large role in generating revenue, where real-world budget control at the SKU level & allocation is needed, where bids need to be tested & retested (in an automated fashion), and so on. All this needs to happen while sometimes (and optionally) minimally disrupting existing campaign structure, especially with clients whose requirements are very sensitive to change. Though competing 3rd-party Amazon Advertising automation tools exist, their lack of depth and the disastrous disruption to campaign structure where these tools force new customers to structure campaigns based on *their* own proprietary campaign paradigms (instead of maintaining a campaign structure preferred by Amazon) may put new customers at a huge disadvantage over the customers’ competition on Amazon. In plain English, the stats history built up by old (but potentially promising) campaigns that’s evaluated by the Amazon Advertising ad auction is irreparably killed off when they’re archived, and new ones are created in their place. Additionally, there are 20 customized parameters (and growing) which we’ve defined in the backend which control the lifecycle of the Amazon Advertising campaigns. Nearly all of these parameters have default values based on historical best practices, but our team does make adjustments based on specific needs.

Like the decision set, parameters are likely to change over time as best practices improve, as Amazon algorithm optimization changes are discovered, and new features are introduced. For example, since October 2018, Amazon has been introducing dramatic new changes in the data schema, with new ad products, new targeting options and bid behavior definitions. Rarely does a week goes by since October where engineering needs to update Coral8’s processes & engines to accommodate the new changes (or workaround them even though they’re not actively being implemented yet, as internal testing is still in-progress prior to wide deployment). When we take into consideration these changes, it’s easy to see how our current set of roughly 500 rules will continue to dramatically increase to help clients maximize these new options. Although presenting meaningful dashboards to clients has been on our roadmap, we would’ve liked to deliver them by now if it weren’t for the changes being introduced by Amazon, but we prioritized by addressing more urgent real-world needs. Our priorities are to focus on maximizing existing needs & processes first before succumbing to the shiny object syndrome. Once the dust settles and bandwidth becomes available, rest assured we’ll push the envelope on innovating other aspects of our business to maximize your success, moving forward.

We still have an outstanding feature wishlist even if Amazon didn’t roll out new changes. Many of them consisted of various bid evaluation scenarios, some which may be covered by Amazon’s new bidding behavior definitions unveiled today, but not entirely.

The beauty of what we’ve built is simply this: the heavy lifting is done by our engine & processes, guided by human analysts. Despite its complexity & intensity, it hums along in the background without significant human intervention. One can imagine that if we created a SaaS based on a version of the engine in the foreseeable future, we could easily anticipate customer interaction only being necessary when creating their account for the first time. Beyond that, all actions could feasibly occur behind the scenes without the customer modifying or reconfiguring anything in their account ever again. Customers wouldn’t need to define parameters. Existing campaigns wouldn’t need to be archived for new campaigns to take their place. The engine will handle everything optimally. In fact, the engine will even dynamically re-adjust (which it already does for certain decisions). It’ll be a crude version of some of the more popular, automated features that Google Adwords provides. I get giddy just envisioning this.

Ultimately, Coral8 is a name indicative of our data-driven obsession. It’s in our DNA. Amazon’s data sets can be disjointed and incoherent and different from other search engines, but Coral8’s ultimate motivation is to correlate and tie behavior, trends, theories & rumors together as cohesively as possible to paint a rational story for us to understand. Meanwhile, here’s a list of configurable parameters so far (with default values as best practices) for campaign optimization. We may publish a deeper dive as to what each parameter represents in future posts. BEACoS

















Howard Lee

Howard Lee

Principal at Coral8, an Amazon-Accredited Brand Marketing Platform | I help Amazon sellers increase margins & sales

Book Appointment

Book Appointment

Amazon Ads Management

Amazon Consulting

Amazon Product Listing

Amazon Product RESEARCH