Most Amazon advertising accounts are not underperforming because of bad bids. They are underperforming because the structure underneath the bids makes good decisions impossible.
When branded and non-branded terms share a campaign, the branded conversions subsidise the wasteful ones and the reported ACOS looks fine. When one budget covers thirty keywords, the two that convert get starved by the twenty-eight that do not. No amount of bid tuning fixes either problem. The account has to be able to tell you where the money is going before it can tell you where it should go.
One campaign, one question
The organising principle is that every campaign should answer a single question. If you cannot say in one sentence what a campaign exists to find out, it is doing two jobs and the data will be unreadable.
Auto campaigns are instruments, not channels
Auto campaigns exist to surface search terms you did not think of. That is their entire job. They get a tight budget, low bids, and their search-term report worked weekly.
Treating an auto campaign as a revenue channel, scaling budget because its ACOS looks good, is a common and expensive mistake. Its ACOS looks good because it is picking up branded and highly specific long-tail traffic that would have converted anyway.
Broad and phrase expand, under supervision
Broad match with tight budgets, aggressive negation, and the expectation that most of what it finds will be negated. Phrase match narrows around themes that broad confirmed.
Both are cost centres that produce information. Budget them accordingly.
Exact match is where the money goes
Terms that have proven they convert graduate into exact-match campaigns, where bids can be pushed because the risk is understood. This is where the majority of spend should sit in a mature account.
Branded defence is isolated, always
Branded terms convert at rates non-branded terms never will. Leaving them mixed into acquisition campaigns makes the whole account’s efficiency look better than the acquisition work actually was.
Separate campaign, separate line in the report. You want to know what it costs to acquire a new customer, and that number is invisible while branded conversions are averaged into it.
Segmenting by product, not just by keyword
Campaigns also need to separate products with different economics.
Two products with different contribution margins have different break-even ACOS, so they need different bid ceilings. The break-even calculation covers why. Putting them in one campaign with one budget means the account cannot express that difference. The higher-margin product gets underbid and the lower-margin one gets overbid, simultaneously.
The same applies to stock position. A product with three weeks of cover and a product with four months should not share a scaling decision, for reasons the stockout piece goes into.
Search terms, weekly
Search-term reports get worked every week, not every quarter.
Converting terms graduate into exact match with a bid set from their own data.
Terms with clicks and no conversions get negated at the level that fits the problem. Negate at ad-group level if the term is bad in that context but fine elsewhere; negate at campaign level if it is bad everywhere. Getting this wrong (blanket-negating a term that converts well on a different product) quietly removes profitable volume.
Terms with impressions and no clicks are usually a relevance or main-image problem rather than a bid problem. They rarely need negating; they need the listing looked at.
This work is unglamorous and repetitive, and it is where most of the recoverable waste in an account lives.
Placement multipliers come from your data
Top-of-search converts differently from rest-of-search and from product pages, and the same keyword deserves a different bid in each.
Set multipliers from the account’s own placement report, not from a rule of thumb. In categories with high impression volume, small placement errors compound quickly. A 300% top-of-search multiplier applied where the data does not support it can consume a budget before lunch.
What good structure feels like
You know the structure is right when a question has an obvious place to look. “Which non-branded terms are acquiring new customers profitably?” should be answerable in one report, not reconstructed from a spreadsheet. “What are we spending to defend the brand?” should be one line.
If answering a basic question about the account requires exporting and pivoting, the campaigns are structured for the tool rather than for the business.
When to rebuild versus optimise
Both, in that order of assessment.
If the existing structure separates intent cleanly and has usable historical data, optimise inside it. History is valuable and rebuilding resets it.
If keywords are mixed across match types, campaigns share budgets they should not, and branded is folded into acquisition, a rebuild costs less than fighting the structure for a year. The lost history is real, and it is worth less than a structure that can answer questions.