Every change should have a reason you can read.

Start with the paper trail

The search-term report is the closest thing an Amazon advertiser has to a receipt. It shows what shoppers typed, what they clicked and what they bought. Read properly, it answers three questions at once:

1. Where is money wasted? Terms with meaningful spend and no orders. Adding them up gives a number every seller should know.
2. Where is the listing letting us down? A relevant term with many clicks and no orders usually points at the page, not the term.
3. Where is the opportunity? Terms that convert well and are not yet getting the attention they deserve.

We sort terms by buyer intent first and volume second. Grouping by intent shows families of shoppers with different needs, and each family tells us what creative and structure to build.

Tie bids to what a click is worth

A bid is a price you pay for a click. The right price depends on what that click is worth, which comes from conversion data: how often it becomes an order, and what an order earns. Bids set by habit or by Amazon's suggestions are guesses. We set them from the numbers, and we use the target ACoS or TACoS agreed with you as the ceiling.

TACoS, ad spend as a share of total sales, matters more than ACoS alone, because advertising should also lift organic sales. A campaign with a poor ACoS can still be worth running if it builds rank that lowers TACoS overall. We track both and say which one a decision is based on.

Rules that keep decisions honest

• Sample size before action. We do not change a bid on one click, or negate a term that has one or two orders at a high conversion rate. That is a small sample, not a loser.
• Patience for attribution. Sales can be credited to an ad for up to 14 days, so very recent data is incomplete.
• Negate in the right place. Off-target terms are negated in broad, phrase and auto campaigns. Terms we deliberately rank for are left alone.
• Know the data before slicing it. Bidding by hour of day can lift efficiency, but only when the pattern is real. In one reported account example, ACoS swung from about 20% to 54% depending on the hour. Acting without understanding your data can do damage.
• One variable at a time. If we change the image and the bids in the same week, we learn nothing about either.

Use other people's data, too

Brand Analytics holds information most sellers never read. The Market Basket report shows what shoppers buy alongside your product. If the co-purchased item is a competitor's, it is a ready-made ad target. It can also point to the next bundle to launch. Search Query Performance shows your click-through and conversion rates against the market for the same query, which tells us whether the image or the page is the weaker link.

The role of AI and automation

We use AI tools to move faster through large reports, spot patterns, draft variants and run more tests in the time a manual process would run one. We treat the output as a draft that a person checks. The principle is that context matters more than clever prompts: a tool given your real data and your real goals does better than one given a long instruction. Speed is the advantage. Judgment stays with people.

What we do

• Review the search-term report weekly, and more often in launch or peak periods.
• Keep a written change log: what we changed, why, and what we expect.
• Measure results against agreed metrics: click-through rate, conversion rate, cost per click, ACoS, TACoS, rank for chosen terms.
• Report in plain language, with the numbers beside the explanation.

Common mistakes

• Negating after two bad days.
• Setting bids from suggestions only.
• Judging every campaign by ACoS, regardless of its job.
• Reporting that lists metrics without saying what to do.
• Trusting a tool's output without checking it against the account.