After ten years in Amazon e-commerce and a desk that usually carries 15 to 20 accounts at once, we see the same automation mistake more than any other: the flat percentage bid rule. One percentage on every bid looks tidy, but it cuts small bids too hard and big bids too gently, so the account never settles. The heavy cuts blow past the profitable bid and need a rebound the next cycle. The light cuts never reach the profitable bid at all. You get an account that oscillates instead of converging. We replaced the single percentage with a matrix that sizes every move to the bid's own magnitude, and this post prints that table in full.
Here is how the table runs. Bids around $3 step down $0.30, bids around $2 step down $0.20, and bids around $1 step down $0.10. Below a dollar the step is $0.07. Between $0.30 and $0.10 the step is $0.05, and between $0.10 and $0.07 the step is $0.02. Bids already sitting at $0.02 or $0.03 stay put, because they have nothing left to give and cutting them further only adds noise to the change log.
Direction comes from one line, the account's ACOS target. On the account this table was built for that line sits at 25 percent. Every target running below the line walks up the table one step at a time, and every target running above it walks down. One override ignores the line entirely: any enabled target with spend and zero sales walks down regardless, because a target with no sales has no ACOS to measure, and spend without sales is the one signal that never needs more context.
Proportional steps work because the size of a sensible move scales with the bid. A $0.30 move on a $3 bid shifts auction position enough to print a readable result by the next cycle, while that same $0.30 step applied to a bid near $0.30 would erase the bid, which is why that band steps down $0.05 instead. When every step matches what the bid can absorb, the account finds its profitable bids within a handful of cycles and then holds there, instead of sawing back and forth across the target forever.
Two caps keep the matrix straight
Cap one is the account's CPC max bid limit, set from margin rather than ambition. One of our accounts caps every bid at $1.56 and another caps at $3.00, because the two catalogs earn different margins and a bid ceiling is a margin statement. No rule in the matrix may walk a bid above that number, so even a target on a hot streak cannot climb past what the unit economics can repay.
Cap two fires when a bid sits above the target's own realized CPC while sales are zero, and the action is simple: set the bid to the CPC. Your bid is a ceiling and the CPC is what the auction actually charges, so a bid well above its realized CPC with no orders is just an inflated ceiling on a keyword that has not earned one. Snapping the bid down to the realized CPC keeps the target live at the price the auction has proven it will charge, and it removes the headroom that one expensive day could otherwise spend.
One more input runs on a schedule: Amazon's suggested bid. On standard campaigns, refactored keyword campaigns, and search term harvest campaigns we periodically reset bids to the suggested bid minus 20 to 25 percent, because those campaigns exist to explore and the suggestion is a decent map of the current auction once you discount it for Amazon's optimism. We do not apply that reset to single keyword campaigns, because a SKAG bid defends rank rather than a number, and rank defense stays manual.
This is plain work by design. A matrix of steps, one ACOS line, two caps, and a scheduled reset keep bids converging across 15 to 20 accounts at once, and the flagship accounts on this system hold TACoS under 15 percent. We have managed over $10M in ad spend on the belief that bidding is not the place for cleverness, so every rule above is boring on purpose, because boring rules behave the same way at three dollars as they do at three cents.
← Back to all posts