I run PPC automation on my own Amazon account — last month it proposed 32 changes, I approved 27 and rejected 5. My real rule stack, why the rejections matter more than the approvals, and why the review queue is the whole point.
I run this automation on my own Amazon account. Rowdy Rooster Woodworks sells wood-finishing products — cutting board oils, hardwax oil, furniture waxes — and last month my rules proposed 32 changes across my campaigns. I approved 27 of them and rejected 5. The five I rejected are the reason I'll never hand my bids to a system that changes them without asking me first.
So if your instinct when you hear "automate your Amazon bids" is to recoil, that's a healthy instinct. The horror stories are real — sellers who flipped automation on and woke up to blown budgets and jacked-up bids. But the problem in those stories is never automation itself. It's the wrong kind of automation: running without guardrails, and without a human in the loop.
Black-box AI vs rule-based automation
Black-box AI takes control of your bids and changes them based on internal logic you can't see. You hand it a target ACoS and it does whatever it decides to hit that number. Sometimes that works. Sometimes it bids $4.00 on a keyword you'd never have paid that much for, burns your daily budget by noon, and reports it as "optimization" — and you can't see why it made the call, because the reasoning is hidden.
Rule-based automation works the other way around. You define the conditions and the actions. The system checks your rules on a schedule, proposes the changes that match, and then waits. You review them, approve or reject, and only then does anything happen.
The difference is accountability. With rule-based automation I can explain every change and exactly why it was made. With a black box, you usually can't.
The rules I actually run on my own account
The starter version is five rules, and it's genuinely where everyone should begin: reduce bids on keywords running hot, raise bids on strong converters, negate search terms that get clicks but never orders, pause the keywords Amazon barely shows, and harvest converting search terms into their own campaigns. That's the infographic at the bottom of this post — and before you turn on a single one, it's worth a quick audit of what you're already running so you automate the right things. If you're just getting started, those five are enough. They're the foundation, and you shouldn't add a sixth until they're running clean. My own account has evolved far beyond them — and here's what those five turn into after months of tuning.
Mine aren't five switches — they're a tiered system that treats a proven winner, a promising-but-unproven keyword, and a money-loser completely differently:
-
Scale what's proven. A keyword under 27% ACoS, converting well, with at least 5 orders behind it, gets its bid raised 20%. The ones with only 3 or 4 orders — promising, not yet proven — get 10%. Winners get room; maybes get a nudge.
-
Rein in what's hot but proven. A keyword running between 27% and 60% ACoS with real orders behind it doesn't get killed. It gets pulled back toward my 27% target, recalculated from what it's actually earning.
-
Reset what's hot and unproven. Same high ACoS, but only a click or two of history — the bid drops back to what a click is genuinely worth, instead of me guessing high on a keyword that hasn't earned it.
-
Cut the bleeders on a rising scale. Ten dollars of spend with zero orders and the bid gets throttled to 28 cents. Sixteen dollars with still nothing to show and it gets paused outright. The more money a keyword wastes, the harder the rule comes down.
Every automated bid is capped at $2.50. The AI can recommend changes, but it's on a leash — and I'm holding it. No rule, no matter what it thinks it sees, gets to bid me into a war I can't win on the math.
One caveat before you copy any of this: these numbers fit my business because I know my margins. The 27% target, the $2.50 ceiling, the $16 kill line — those are mine. Yours will almost certainly be different. The structure transfers; the exact numbers don't.
The lookback window is about volume, not the calendar
New sellers agonize over the right ACoS threshold — 30%? 35%? 40%? — and overlook the setting that matters more: how far back the rule looks. And the lookback isn't a number you copy off a blog. It's a question of scale and volume. The longer you look back, the more data piles up, and the more likely any condition becomes true — so a loose rule on a long window will fire constantly on noise.
What you're really doing is waiting until enough has happened to trust the signal. A high-traffic keyword builds that evidence fast, so I scale those on a 7-day window. A slow one needs longer before I touch it. Amazon's own 7-to-14-day attribution lag is why you never act on three days of data — a click today might not convert until next week.
But some signals don't need a month. If a keyword has spent $16 and returned zero orders, that's a problem whether you saw it over 14 days or 30. Once real money has gone in with nothing coming out, the spend is the signal — the window just has to be long enough to reach it. Match the lookback to how much volume the keyword actually throws off, not to a number someone told you was safe.
Placement is a rule too
A placement multiplier is just another bid. If you're already automating bids, there's no reason you can't automate your Top of Search adjustments too — and almost nobody does. The same review-and-approve loop tunes placement on my account: when one of my cutting-board campaigns converts under 30% ACoS, the system nudges its Top of Search share up a couple of points; when a campaign runs hot, it eases that share back down. Small, constant, reversible adjustments I'd never keep up with by hand. I go deeper in how I bid for Top of Search — the point here is that placement belongs in the same rules-and-approval system as everything else.
The approval step — and the five changes I rejected
The most important part of any automation system isn't the rules. It's the queue where you approve or reject what they propose. Here's the clearest example I have of why.
On the first of the month, my rules flagged two search terms to negate for running over my ACoS ceiling: "oil for cutting board" at 51% ACoS on 3 orders, and "cutting board conditioner" at 45.7% on 7. A system that executes rules without review might have negated both overnight, and I'd never have known. I rejected both.
Because those aren't wasted spend — they're the exact words that describe what I sell. A high ACoS on a core, category-defining term isn't a leak; it's the cost of defending the search that is your product. It's the same reason you'd never let a rule negate your own brand name. Any rule that cuts everything above a threshold will, sooner or later, come for the keywords you most need to own — and only a person looking at the term knows the difference.
That's the whole case for the review step. The rule is right most of the time — I approved 27 of last month's 32 changes. It's the handful of exceptions that would quietly wreck an account, and no threshold catches those. A five-minute review does.
Tools that skip that step — that change bids and log it after the fact — are the ones that produce the horror stories. If you can't review before execution, you don't have automation. You have a bot with your credit card.
The goal of automation isn't to remove you from the loop. It's to make the loop faster.
RedHen Labs runs your rules on your schedule and proposes changes to a review queue, so no bid moves without your approval unless you deliberately switch that automation to run unattended. Every change is logged with a one-click undo. Your rules, your thresholds, your approval — at $129/mo flat, not a percentage of your spend. Start your free 14-day trial.
