Set your Amazon PPC starting bid from your own economics, not from Amazon's suggestion box: revenue per click times target ACOS is the most you can pay for a click and still land on your target. Amazon's suggested bid is built to win the auction, not to protect your margin, so treat it as a reference and start below it. Then choose the dynamic bidding mode by the campaign's job: down only for discovery and unproven traffic, up and down only for proven exact match terms, fixed bids almost never. Every bid I set in the accounts I manage comes out of that one sentence. This guide works the formula on real numbers, explains what the suggested bid actually knows and does not know, and ends with the decision table I price bids from.
The starting-bid formula: revenue per click times target ACOS
Your maximum profitable bid is revenue per click multiplied by your target ACOS. Revenue per click is the keyword's own sales divided by its own clicks, or, when there is no history yet, product price times an estimated conversion rate. The result is the most a click can cost while the keyword still lands on target.
The algebra is short enough to keep in your head. ACOS is what a click costs divided by what a click earns: CPC over revenue per click. Rearrange it and CPC equals ACOS times revenue per click, which turns your target ACOS into a price tag for the click. Nothing about the auction, the suggested bid, or your competitor's aggression appears anywhere in that equation. The ceiling comes entirely from what your own traffic earns.
Run it on the dataset I grade across these guides: $111,058 in ad sales on 45,672 clicks makes each click worth $2.43 in revenue. At a 30% target ACOS, the fair price for that average click is $2.43 times 0.30, which is $0.73. The account actually pays an average CPC of $0.79. So the formula's verdict on this real account is that the average bid runs about six cents hot, roughly 8%, and that gap compounds across every one of those 45,672 clicks. Small mispricings at the bid level are how an account drifts from a 30% target to the 32.7% it actually posts.
Setting a bid for a brand-new keyword with no history
A new keyword has no revenue-per-click history, so you estimate it: product price times an estimated conversion rate, then times target ACOS. Then open below the result, not at it, because unproven traffic has not yet earned the full fair price. Let performance argue the bid upward from there.
Worked on the same dataset's averages: a $26.83 product at a 9.1% conversion rate earns an estimated $26.83 × 0.091, about $2.44 per click (a penny off the measured $2.43, which is the rounding talking). At a 30% target, the ceiling is $0.73, and a new keyword opens under that, not on it. For the conversion estimate, use your account's own average if the product has any track record, or the conversion floors for your strategic phase in my Amazon PPC benchmarks guide if it does not. Estimate conservatively: an optimistic conversion guess inflates the ceiling, and the ceiling is the one number in this system you do not want inflated.
Should you trust Amazon's suggested bid?
Treat Amazon's suggested bid as intelligence about the auction, never as a price for your click. It is derived from recent winning bids for ads like yours, so it estimates what winning currently costs. It knows nothing about your margin, your conversion rate, or your target ACOS, and it commonly sits above what those numbers justify.
Think about what the suggestion is for. Amazon's job is to fill ad slots, and a bid high enough to win the impression serves that job whether or not the click ever pays for itself on your side of the ledger. The suggested bid and its range answer one question, what does this auction cost to win, and stay silent on the only question that decides whether you should want to win it: what is a click worth to you. Those are different numbers computed from different inputs, and in the accounts I audit, the first commonly runs above the second. I frame that as a pattern I keep observing, not a law with a percentage attached, because the gap varies by category and by term.
The popular workaround is to open at some fraction of the suggestion, half or a bit more. I understand the instinct, but a discount off the wrong anchor is still the wrong anchor: half of a number that ignores your margin also ignores your margin. The disciplined sequence is the reverse. Compute your own ceiling first, from revenue per click times target ACOS, then look at the suggested range as market intelligence. If your ceiling sits inside the range, good, you can compete at a price that works. If your ceiling sits far below the bottom of the range, the auction is telling you something real: at your target ACOS you may not be able to afford this term, and holding the top of that auction is a deliberate investment decision, not a default. Either way, the bid starts at your number, and performance data, not the suggestion box, argues it up.
Dynamic bidding: down only vs up and down vs fixed
Amazon offers three bidding strategies per campaign. Dynamic bids, down only, lets Amazon lower your bid in real time when a conversion looks unlikely, never raise it. Dynamic bids, up and down, also lets Amazon raise it, by up to 100% at Top of Search and up to 50% elsewhere, when it predicts a conversion. Fixed bids spend exactly as written. Pick by the campaign's job.
| Strategy | What Amazon can do to your bid | The job I give it |
|---|---|---|
| Dynamic bids, down only | Reduce it in real time when a conversion looks unlikely; never raise it | Discovery and anything unproven: auto campaigns, broad match, new keywords |
| Dynamic bids, up and down | Raise it up to 100% at Top of Search and up to 50% at other placements, or reduce it | Proven exact match terms with real conversion history, nothing else |
| Fixed bids | Nothing; the bid spends exactly as written | Deliberate, short, budget-capped tests; almost never a standing setting |
Down only is my default, and it is the only mode I run on discovery traffic. An auto campaign, a broad match ad group, a freshly launched keyword: their job is to explore, and exploration means a large share of clicks that were never going to convert. Down only means Amazon takes a cut off your bid on exactly those clicks. That is a discount you collect for free, on the traffic most likely to waste money, which is why unproven bids start below the fair average and run down only.
Up and down is an authorization, and I word it that way on purpose: you are authorizing Amazon to double your bid at Top of Search whenever its model predicts a conversion. On a proven exact match keyword, one with a real order history at or under target ACOS, that authorization makes sense, because the prediction is playing with loaded dice and the extra visibility lands on a term you already know pays. On an unproven term it is a premium paid on a guess. My rule: a term earns up and down the same way it earns a place in the exact match performance campaign, with orders. Since the strategy is set per campaign, this is one more reason campaigns get drawn around jobs rather than around convenience, which is the argument of my campaign structure guide: discovery campaigns run down only, performance campaigns can run up and down, and the two never share a wall.
Fixed bids I use rarely, and only on purpose: a short test where the bid has to hold still so something else can be measured, capped by budget and ended by calendar. As a standing strategy, fixed refuses the one gift in this menu, the automatic discount on doomed clicks, and gains nothing in exchange that a down-only campaign does not already have.
How placement modifiers stack on top of your bid
Your bid entering a given auction is the base bid times one plus the placement modifier, and up-and-down bidding can then raise that result again, up to double at Top of Search. Price the base bid from your economics and let the modifier carry the placement premium separately. Never stack a modifier on Amazon's suggestion.
The arithmetic deserves a moment of respect. Take a $0.80 base bid, add a 75% Top of Search modifier, and it enters top-slot auctions at $1.40. Put the campaign on up and down, and Amazon may take that $1.40 as high as $2.80 on a predicted conversion. One console setting and one checkbox, and the click can legally cost three and a half times the number you typed. Now imagine the base was not $0.80 from your own formula but a $1.50 suggestion accepted at face value, and the stack prices clicks your margin never had a chance of carrying. The full placement math, including how I size the modifier from each placement's own revenue per click, lives in my Top of Search guide. The takeaway for bid setting is separation of duties: the base bid prices the average click from your economics, and the modifier prices the top slot's premium from its own placement data, each argued from its own numbers.
When to judge a bid, and when to change it
Give a bid one order's worth of clicks before judging it: one divided by your conversion rate, roughly eleven clicks at the 9.1% in my sample dataset. Then reprice on a cadence, weekly to biweekly on most accounts, using the same formula against fresh conversion data, with changes applied in bulk.
The fair-trial math matters because the most common bid mistake I see is not a wrong number, it is a right number abandoned too early. At 9.1%, an average order arrives every eleven clicks or so, which means a keyword with five clicks and no sales has told you nothing, and reacting to it is reacting to a coin that came up tails twice. Somewhere past double the fair trial with zero orders, the silence becomes a verdict; the full math on that threshold, and what to check before acting on it, is in my guide to clicks with no sales. Note that the trial length is personal: at a 3% conversion product, a fair trial is closer to 33 clicks, so compute yours before borrowing mine.
Repricing is the same formula on a schedule. Each pass, pull the keyword's own sales and clicks from the search term report, recompute revenue per click, multiply by target ACOS, and write the result down as the new bid: sometimes a trim, sometimes a hold, occasionally a raise. This is lever three of my five levers for lowering ACOS, and at any real account size it happens in the bulk file, where hundreds of bids move in one upload, each priced from its own row rather than by a blanket percentage. If you want to see what this looks like on real rows before running it on yours, the audit demo shows the same read on a real anonymized account.
The starting-bid decision, summarized
Four situations cover almost every bid I set: a new keyword with no data, a proven converter, a converter running above target ACOS, and a deliberate ranking push on a VIP term. Each gets its base bid from the same formula and its bidding mode from its job.
| Situation | Base bid | Bidding mode |
|---|---|---|
| New keyword, no history | (Price × estimated conversion rate) × target ACOS, opened below the result | Down only |
| Proven converter at or under target | Its own revenue per click × target ACOS | Up and down, in the performance campaign |
| Converter running above target ACOS | Repriced to revenue per click × target ACOS; never paused for cost alone | Down only until it re-earns the premium |
| VIP ranking term, deliberate push | Planned overpay with a budget cap, an ACOS you agreed to lose, and an end date | Up and down plus a Top of Search modifier, in its own campaign |
Two footnotes on that table. The high-ACOS row exists because expensive converters get repriced, not executed: pausing a keyword with orders surrenders the sales and the organic position they feed, when the actual problem was a price. And the ranking-push row is the one sanctioned exception to everything above, a controlled overpay whose rules, caps, and exit conditions are covered in the ranking-push section of the Top of Search guide. It is an exception precisely because it is planned. Most overpriced bids I find in audits are not plans, they are suggestions that were accepted in 2024 and never questioned since.
Frequently asked questions
Should I use Amazon's suggested bid?
As a reference, not a price. The suggested bid is derived from recent winning bids in auctions like yours, so it tells you what winning currently costs. It knows nothing about your margin, your conversion rate, or your target ACOS, and in the accounts I audit it commonly sits above what a seller's own economics justify. Compute your maximum profitable CPC (revenue per click times target ACOS), start at or below it, and let performance argue the bid up.
What is a good starting bid for a new Amazon keyword?
There is no universal number, because the right bid depends on your price and conversion rate. Estimate revenue per click as product price times an estimated conversion rate, multiply by your target ACOS, and open below the result, because unproven traffic has not yet earned the full fair price. A $26.83 product at an estimated 9.1% conversion earns about $2.44 per click; at a 30% target ACOS that caps the bid at $0.73, so I would open under that.
Dynamic bids down-only or up-and-down: which should I use?
Match the mode to the campaign's job. Down only for discovery and anything unproven: auto campaigns, broad match, new keywords. It lets Amazon cut the bid when a conversion looks unlikely but never raise it, which caps the cost of exploring. Up and down belongs only on proven exact match terms with real conversion history, because you are authorizing Amazon to raise the bid up to 100% at Top of Search, and only proven terms have earned that authority.
Why are my Amazon CPCs so high?
Usually because something multiplied the bid past what you think you set. Check three suspects in order: a starting bid anchored to Amazon's suggestion instead of your own economics, a placement modifier stacking a percentage on the base, and up-and-down bidding raising the result by up to another 100% at Top of Search. A $0.80 bid with a 75% Top of Search modifier under up-and-down bidding can legally cost $2.80 a click. Reprice from revenue per click times target ACOS, then audit the multipliers.
Should I ever use fixed bids on Amazon?
Rarely. Fixed bids spend exactly what you wrote, which forfeits the free discount down-only gives you when Amazon predicts a click is unlikely to convert. The legitimate use is a deliberate test that needs the bid held constant: measuring a placement, isolating one variable, or forcing delivery in a short, budget-capped experiment. As a standing strategy for normal campaigns I do not use it, because down only does everything fixed does minus overpaying for the worst clicks.
How often should I change my Amazon PPC bids?
On a cadence, not on a mood. A bid deserves judgment only after one order's worth of clicks: one divided by your conversion rate, roughly eleven clicks at 9.1%. Daily bid changes react to noise, not signal. I reprice on a regular pass, weekly to biweekly on most accounts, recomputing each keyword's revenue per click from fresh data and applying the same formula every time, with the changes made at scale through the bulk file.
Price your next bid pass from your own numbers
Everything in this guide runs on two numbers you already own: what your clicks earn and what ACOS you are aiming for. If you want the account-level version computed for you, upload a 60-day bulk file to the free Account Health Snapshot: it grades your CPC, conversion rate, and wasted spend against your phase in about 60 seconds, in your browser, no email, no account. The Audit Dashboard goes a level deeper and shows which targets are priced above what their own revenue per click supports. And if the gap between what you pay and what the formula says you should pay looks wide, bring the file to the free 30-minute diagnosis call and I will walk the repricing math with you on your own rows, whether or not you ever hire me.