Amazon Advertising Optimization & Automation

Amazon advertising was a material growth lever for a $50M+ e-commerce business. I analyzed and supported strategy for more than $5 million in annual spend.

The commercial question was not “how do we spend the budget.” It was whether that spend was producing profitable net sales, and whether the team could make that judgment every week without burning the week on exports.

The account had grown past 5,000 campaigns. At that count, two people can look at the same week and reach different conclusions, because nobody can review the long tail with the same rules.

Structure also hid performance. A campaign-level ACOS could look acceptable while one placement absorbed spend that did not earn its keep. Consolidation was an operating decision, not a cleanup task.

ACOS, spend, and attributed sales were the starting point, not the decision. I defined the review around:

• ACOS, with an explicit before-and-after read (above 16% down to 14.5%)
• Net sales, so efficiency was not celebrated if the business got smaller
• Placement differences, including top of search, rest of search, and product pages
• A profitability lens using cost and margin context where the data supported it

Same-SKU and other-SKU sales were separated when the export allowed it, because attributed sales are not the same thing as incremental profit. Where incrementality could not be measured cleanly, I said the recommendation was efficiency and mix, not a proven lift study.

Each cycle started with where the money and the variance were. Campaigns and placements were ranked by spend and by how far they sat from the efficiency and sales outcome we wanted.

That kept the conversation on a short list of actions: bid and placement changes, budget concentration, and structure cleanup. The rest of the account was monitored with the same rules so it did not depend on memory.

The Python system encoded the first layer of that review. It pulled performance, calculated the metrics we had agreed to use, compared current modifiers with recommended ones, and wrote the result to a review file and a database history.

Automation did not replace the commercial call. It made the same logic show up every time, and it cut the manual optimization work by approximately 60–80%. People spent the remaining time on exceptions, launches, and decisions that needed context the model did not have.

In parallel, I helped migrate the structure from more than 5,000 campaigns to fewer than 1,000. Fewer campaigns made the rules easier to apply and made performance easier to explain in a business review.

Consolidation followed the measurement approach. We did not collapse structure in a way that destroyed the ability to see a placement or a product group that was spending without returning net sales.

ACOS came down from more than 16% to 14.5%. Net sales increased over the same period. Manual optimization effort fell by about 60–80%, and the account became something a small team could actually run.

I supported the strategy and the operating system around the spend. I do not claim sole control of the advertising budget or of every adjacent sales driver.

Advertising optimization is a commercial problem that happens to require software. The useful system is the one that makes a profitability-aware decision easier to repeat, and that can be explained to someone who will never open the script.

The campaign count was a management problem. Until it came down, better metrics could not be applied consistently.