How Alien Road Uses Amazon Ads AI Tools in Client Campaigns

Amazon has added automation and generative features to its advertising console over several years. The ones we use most often in client work fall into four groups.

• Bidding: dynamic bids (down only, up and down), placement adjustments, and rule-based bidding for Sponsored Products.
• Targeting: automatic targeting with close match, loose match, substitutes and complements, plus theme-based targeting for Sponsored Brands.
• Creative: the AI image generator and video generator in the console, which create lifestyle scenes and short videos from product images.
• Measurement and planning: Amazon Marketing Cloud (AMC) for custom audience and path analysis, and recommendation features inside campaign manager.

Availability differs by marketplace and account type, and features change, so we check each client’s console rather than assuming a feature is there.

Automatic campaigns let Amazon match ads to searches and products based on the listing. They are a discovery tool, not a final structure. We run them with modest budgets and treat their search term reports as research.

Every week, the team exports the search term report and sorts terms into three groups. Converting terms with acceptable advertising cost of sale (ACOS) move into manual exact and phrase campaigns. Irrelevant or loss-making terms become negative keywords in the auto campaign. Terms with clicks but too little data stay for another cycle.

This “harvest and negate” loop is not new, but AI-driven matching makes it more important: the algorithm finds more distant matches than before, and some of them are useful while others are simply expensive.

Dynamic bids up and down raise bids by up to 100% for top-of-search placement and up to 50% for other placements when Amazon predicts a sale is more likely. That suits campaigns with a proven ACOS and enough margin. For new products or tight margins we start with “down only”, which lowers bids when conversion looks unlikely.

A simple rule set

In a typical account the team writes the bidding rules down before launch so changes are deliberate:

• New product launch: dynamic down only, fixed placement adjustments of zero, review after 14 days.
• Proven hero product: dynamic up and down, top-of-search adjustment based on placement report data.
• Defensive brand terms: fixed bids, since the goal is coverage rather than efficiency.
• Clearance stock: rule-based bids that lower spend as inventory drops.

These are starting points. The placement report decides adjustments after a few weeks of data.

Small brands often have only white-background product photos, which are required for the main listing image but weak for Sponsored Brands and Sponsored Display. The AI image generator places the product in a generated scene, such as a kitchen counter or outdoor setting, from a short prompt.

Our designers use it for quick variants, then review each image against Amazon’s creative policies and the brand’s style. The product itself must not be altered, and generated scenes must not imply features the product does not have. Where a brand has strong photography, we use real photos and keep the generator for testing new concepts.

Video follows the same logic. Sponsored Brands video tends to stand out in search results, and the video generator gives brands without a production budget a first version to test.

AMC is a clean-room environment where advertisers run SQL queries on event-level ad data, with privacy rules that only return aggregated results. It answers questions the standard reports cannot: how many customers saw a DSP ad before searching the brand, how long the path to purchase is, or which campaigns bring new-to-brand customers who buy again.

For eligible clients we use AMC to build audiences, for example people who viewed a product detail page but did not purchase, and to check whether upper-funnel spend is helping lower-funnel results. AMC needs analyst time and a minimum amount of data to return results, so it is not the right tool for every account.

Recommendations in the console are written by the platform that sells the ads. Many are sensible, such as raising budgets on campaigns that run out early. Some simply increase spend. We evaluate each against the agreed targets instead of accepting them in bulk.

Generated creative needs review, both for policy and for honesty. And automatic targeting needs constant negation; left alone for a month, it can drift into broad, low-intent traffic.