What am I doing wrong? Upload three reports and find out.
Amazon shows you what happened. It will not tell you which part of it was a mistake, which product cannot afford the ads pointed at it, or what to change first. You get the answers as a file you upload straight back into Amazon.
No login and no Amazon account to connect. Your data stays yours: we do not sell it, share it, compare it against anybody else’s, or train anything on it — and nobody here goes looking through it.
See it on a real account
A whole account, already analysed — every bid with the reasoning behind it, the terms to negate, and the ones that convert and must never be.
Three files Amazon already gives you. Back comes the sheet below, filled in: bid changes, a harvest list, negatives, budget calls and a diagnosis of what moved and why.
No files, no account, nothing to download. Type in one product's numbers and find out what it can afford to spend on advertising — and what a click is worth.
Or keep scrolling — the file writes itself
| Record | Campaign | Keyword or target | Match | Bid | New bid | Operation | Why |
|---|
This is the sheet Seller Central accepts.
Empty, it is a thousand cells nobody has time to fill on a Sunday. Keep scrolling and the engine fills it, one decision at a time, with the reason beside each.
Every bid, priced against a real margin.
One formula, everywhere: target ACOS × conversion rate × order value, and no single move larger than 15%. Not a target you typed in and hoped about.
Harvest with the other half attached. Negate on clicks, never on ACOS.
A term that earns its place is promoted to exact and negated where it came from. A term that converts but loses money is repriced, never thrown away.
And the rows it would not write.
Under ten clicks it will not price from a conversion rate. A conversion problem never gets a bid change. It says so in the file, rather than filling the gap with something plausible.
1,412 rows. One upload. Yours takes three reports.
That was an example account. Everything it does, and every number that went into the file, is one page on.
What is this click actually worth?
Not what you are bidding, and not what the category charges. What this keyword can earn on this product, at this product's margin.
Its conversion rate, your order value, your target.
The target is not typed in. It comes down from this product's breakeven of 38.0% — what is left after Amazon's cut, the FBA fee and what the unit costs you.
Target ACOS × conversion rate × order value.
One line of arithmetic, the same on every keyword in the account. 24% × 11% × $36.50 = $0.96.
So you are paying 46% over the target.
At $1.40 this keyword runs at 35% ACOS — under the 38.0% breakeven, so it is not losing money. It is making about 13¢ a click where it could be making 57¢, on every click, for as long as nobody works it out.
$1.19, not $0.96.
No bid moves more than 15% at once. A bigger cut would make the result unreadable — you would not know whether the keyword changed or you did. It steps, and checks again.
Every recommendation arrives with the line it was measured against.
This is everything Amazon hands you.
Impressions, clicks, spend, sales, ACOS. Five columns of what happened, on every search term you paid for.
35%, 42%, 17%. Which one is the problem?
There is no answer in this table. The same figure is money well spent on one product and a loss you are paying to repeat on another, and nothing here says which.
What the unit costs you.
It is the only figure no Amazon report contains, and it is the one that decides every other number on the row. Two columns — the ASIN, and what one unit costs you landed. That is the whole of what we ask for.
One cost. Three different answers.
Nothing above changed. Two of these are being overpaid for and one is being starved — and the table only says which once it knows what the unit costs you.
Amazon gives you the numbers. This tells you which of them is a problem.
Every figure below comes out of reports you already have, worked out in plain arithmetic you can audit. Nothing is applied to your account for you, and nothing here is a projection dressed up as a fact.
Money
what a product can actually affordBreakeven ACOS, per product
Price, less landed cost, Amazon's referral fee, fulfilment and anything else variable — as a share of the price. That share is the ceiling: spend more than it advertising a unit and the unit loses money. Every bid in the product is derived from this one number.
A product below zero is flagged loudly. It loses money on every sale before a penny of advertising, and no bid change fixes that.
Where each figure came from
The price you were actually paid, fees from your Fee Preview, landed cost from your own sheet. Each figure is labelled exact, derived or estimated, so you know which numbers to lean on and which to go and check.
Fees are category-correct across all 34 Amazon categories, including the tiered ones that stack rates across bands rather than applying one rate.
A profit and loss that includes your own costs
Storage, a VA, software, freight outside Amazon. None of it appears in any Amazon report, and without it a net profit figure flatters you. A line we were not given is never treated as a zero.
A forecast that admits what it is
Trailing trend and day-of-week shape, shown as a range built from your own week-to-week spread — never a single number. Under eight weeks of data it says the forecast is unreliable and shows it anyway, with that warning attached.
This is what the report shows.
ACOS, day by day, for one keyword. A line that wanders between thirty and forty-five percent. Good or bad? The report cannot say.
Then breakeven lands on it.
38.0%, from your own costs. Every day the line is above it is a day this keyword lost money on every order it won. Most of them.
And the target sits inside the margin.
24% for a Scale campaign — breakeven less the margin you asked to keep. That is what the bid is priced to hit.
Same keyword, repriced.
What the last thirty days look like at 96¢ a click instead of $1.40, at this conversion rate. Not a projection dressed up as a fact — the same arithmetic, run forward.
Bids
a number you can type into Seller CentralOne formula, applied everywhere
ACOS is cost per click divided by conversion rate times order value. Rearranged, the most a click can be worth is your target ACOS times conversion times order value. That is the whole of it, and every recommendation shows its working.
Guardrails that do not get tired
No bid derived from under ten clicks. No single move larger than 15%. Never a bid change and a placement change on the same campaign in the same week, because then you cannot tell which one worked.
It knows Amazon charges more than your bid
Dynamic bidding and placement adjustments can take a click well above what you set. The gap between your bid and what you were actually charged is measured per keyword and built into the number it gives you back.
Sponsored Products, Brands and Display
All three read from one upload. A Brands campaign on a lifetime budget is not paced day by day, and a Display campaign charged per thousand views is not given a cost-per-click bid — both are refused out loud rather than answered with a made-up number.
Search terms
what to promote, what to kill, what to leave aloneHarvest, with the other half attached
A term that has proven itself is promoted to exact at a bid derived from its own conversion rate — and negated in the campaign it came from, in the same file. Promote without negating and your two campaigns bid against each other.
It can also build the whole campaign for a harvested term: campaign, ad group, product ad and keyword, ready to upload.
Negation on clicks, never on ACOS
A term is negated when it has taken at least twice the clicks that product normally needs for a sale and made none. A term that converts but costs too much is never negated — it gets its bid cut. Negation throws a real buying signal away for good; a bid cut just reprices it.
The losing word, not just the losing term
Fifty terms wasting three dollars each look like noise one at a time. Split into their component words and rolled up, they usually share a root — free, cheap, a competitor's name — and one negative phrase does the work of fifty negative exacts.
Where you are bidding against yourself
The same keyword live in two campaigns means Amazon enters one of them and you push your own price up, while the evidence splits in two. It names the copy that converts best and what to do with the others — never "delete", because that throws the history away.
| Search term | Clicks | Orders | Needs | ACOS | Verdict |
|---|
Negate on clicks. Never on ACOS.
"Needs" is twice the clicks this product usually takes to make one sale — eighteen here. A term that spends that with nothing to show is negated. Watch the fourth row.
It converts. It just costs too much.
Seven orders at 44% ACOS is a real buying signal priced wrong. Negating it throws that signal away for good. A bid cut just reprices it — so that is the row it writes.
And the ones it leaves alone.
Fourteen clicks and no orders is bad luck until it is eighteen. It says "watch", and checks again next week, rather than guessing early.
Campaigns
four jobs, four different metricsLaunch, Harvest, Scale and Defend do different jobs. Holding all four to one ACOS target strangles launches before they gain rank and overspends on brand defence. Each campaign is classified from its name, its targets and its structure, told what it should and should not be judged on, and the classification can be overridden — which recalculates everything downstream.
One ACOS target across everything is the commonest mistake there is.
It strangles a launch before it gains rank and overspends on brand defence. Each campaign is read for the job it is doing.
Misread a campaign? Change it, and everything recalculates.
The classification is shown, not hidden. A campaign named "KTC-2 Sept" tells nobody anything — you say what it is, once, and it remembers.
Budget, rank and events
the parts nobody has time to do by handBudget pacing that refuses to guess
Spend and days come from the same dated file. With no dates it says so rather than dividing by a number it invented, and it will not report a utilisation Amazon's monthly averaging makes impossible.
Share of search
Brand Analytics is the only free Amazon report that says where you stand against everyone else selling into the same search. It is what lets a launch be judged on rank moving rather than on an ACOS it was never meant to hit.
An event plan for Prime Day
Warm-up, event and cool-down budgets per campaign, derived from breakeven. Scale campaigns are raised only where they are already inside target, launches get a ceiling and a stop condition, and harvest campaigns are held — terms "proven" during an event are proven against a week that will not happen again for a year.
Ninety days, then eighteen months
Amazon keeps about ninety days of search term history — and only sixty on Sponsored Brands and Display — then deletes the rest. Connect your Amazon Advertising account and Vantwyn pulls your reports itself, every day, read-only — nothing to upload — and you hold eighteen months, the longest Amazon’s own rules allow anyone to keep it. It is yours: never sold, never shared, never pooled with anybody else’s. A day already held is written once, so nothing ever doubles.
Why it moved
the question a report cannot answerACOS can only rise for three reasons: clicks got dearer, fewer of them converted, or orders got smaller. The movement is split between the three and each given its share — per product and per campaign, not just across the account, where the answer is usually "a mix".
If conversion fell while the cost per click held, the problem is your listing and a bid cut makes it worse. That distinction is the whole reason the decomposition exists.
What you get back
three files- An Amazon bulk upload file — bid changes, new keywords and negatives, formatted to go straight back into Bulk Operations, and checked before you get it. This is the one that turns a report into an afternoon's work done.
- A spreadsheet per table — bids, harvest, negatives, budgets, products. For people who want to sort and filter it themselves.
- A PDF of the whole analysis — the profit picture, what needs attention and why, for handing to somebody who is not going to log in. Put your own logo and name on it and ours comes off, which is the version an agency wants.
What it will never do
on purposeFour things this refuses
It will never change your account for you. Every recommendation is yours to approve, ignore or argue with. Nobody hands a black box unsupervised control of real money.
It will never send your raw report rows to a language model. The arithmetic is plain Python on our own machine. The model only ever sees a computed summary, with product identifiers stripped out.
It will never invent a number. Where a figure cannot be worked out honestly, the screen says so and says what is missing — rather than filling the gap with something plausible.
Signed out, it will never keep anything. Your files are read in memory, the report is built, and nothing about your business is written down.
A whole account, already analysed — every bid with the reasoning behind it, the terms to negate, and the ones that convert and must never be.
Upload what Amazon already gives you. Back comes the sheet, filled in — bid changes, a harvest list, negatives, budget calls and a diagnosis of what moved and why.
Type in one product's numbers and find out what it can afford to spend on advertising — and what a click is worth.
Your account
Everything worked out from your reports, kept so it is still here next week — and kept for you alone. We do not sell it, share it, or pool it with anybody else’s.
Your data is your own. We do not sell it, share it, pool it with other sellers’ data, or train anything on it, and no one here reads through your account looking at what you sell. It is used to produce your analysis and nothing else.
On this free tool specifically: your files are read to build the report and the server forgets them — close the tab and it is gone. The one thing kept is a random marker in your browser so the free AI answers cannot be farmed; it carries no name, no email and nothing about what you sell. With an account we do keep your history — that is the whole point of one — and you can delete it whenever you like.
No files needed. Enter one product's numbers and Vantwyn works out what it can afford to spend on advertising — the breakeven ACOS, and the target that leaves your margin intact. The referral fee is looked up from the product's category.
What this means
Account health
How the account is built, before any bid is touched: a score out of 100, and for each check what was found and the one thing to do.
| Check | What we found | Do |
|---|
The anchor
Every bid below traces back to one number per product: the advertising cost at which that product stops making money. Open what if on any row to drag price and cost around and watch the margin move.
Bid recommendations
Every move is capped at 15% of the current bid. A bigger jump makes the result unattributable and forces Amazon's delivery algorithm to relearn the keyword. Click any row for the working.
| Keyword | Job | Bid now | Recommend | Change | ACOS | Target | Breakeven | Data |
|---|
Every keyword and search term
Search terms
Converting but unprofitable is a bid cut, never a negation
Negating a term that converts destroys a real demand signal permanently — you stop seeing it, and if your price or costs improve later you never find out it became profitable. A bid cut only reprices it, and you can cut again next cycle. It is an easy distinction to get backwards, and getting it backwards cannot be undone.
Negatives blocking your sales
A negative keyword stops anything it matches from showing in that campaign. These block your own keywords, or searches that brought you orders before the negative was added -- the sales shown are from before it went in; with it in place they cannot sell there. Archive or narrow them in Campaign Manager: removing a negative cannot be undone, so it is not in the upload file.
| Negative | Where | What it blocks | Orders | Sold before the block | Do |
|---|
Stock and bids
From your FBA inventory report. A product that will run out before more arrives has its Scale and Harvest bids lowered 15% to stretch what is left, instead of paying for clicks you cannot ship. One that is out has every bid change held: its ads have stopped by themselves.
| Product | Stock | Days left | Sells a day | What it did to bids |
|---|
Keyword research
Every keyword in the research file you uploaded, checked against your own account: already running, already proven, your brand, a competitor's, worth bidding on, worth a test, or not for you -- with the match type, the campaign it goes in and a starting bid from your own margin and conversion rate.
Fit it to your budget
Not by search volume: by what each keyword costs to learn, whether your margin can win it, and how many sellers advertise on it.
Every keyword
| Keyword | Verdict | Match | Goes in | Starting bid | Their volume | Who ranks |
|---|
Harvest
Each promotion comes with its negation. Adding the keyword without negating it at source puts two of your own campaigns into the same auction, inflating your CPC and splitting the data across both.
| Search term | From | To | Clicks | Orders | ACOS | New bid | Then |
|---|
Word roots
Fifty terms each wasting three dollars look like noise one at a time. Rolled up by word they usually share one losing root, and a single negative phrase does the work of fifty negative exacts.
| Root | Terms | Clicks | Orders | Spend | Verdict | Seen in |
|---|
Campaign jobs
Campaigns do four different jobs and must be judged on four different metrics. Applying one ACOS target across an account strangles launches and overspends on brand defence.
| Campaign | Job | Judge it on | Spend | ACOS | Target |
|---|
Budget pacing
| Campaign | Budget | Avg spend | Used | ACOS | Verdict |
|---|
Why the numbers moved
What this run could see
Do these first
Ask about this account
Pick a question. Answers come from the numbers on this page and
nothing else — every figure quoted is checked against them
before you see it.
Instant answers are free and unlimited. Questions
that need AI are limited to three a day.