Best Amazon PPC tool: the shortlist matters less than how you test it

Why almost every tool trial produces no information

Comparison tables in this category run twenty columns wide and settle nothing, because nearly every tool does bid adjustment, keyword harvesting, negatives and dayparting. The differences that decide your outcome are not on the feature axis, and the only way to find them is a test. Then the test goes wrong in one of four predictable ways.

• Three things changed at once. The tool went in, somebody rewrote two listings, and a promotion ran. Whatever happened next has three candidate causes and no way to separate them.
• The metric was chosen afterwards. ACOS improved, so ACOS became the metric. ACOS improves reliably whenever you spend less, which is not the same as making more money.
• The window was two weeks. Long enough to measure the disruption of switching and nothing else.
• The season arrived. A test that spans a tentpole event measures the event.

The rest of this page is the version that works. It costs nothing but discipline, and it produces an answer about your own catalogue, which beats every roundup on this topic — including this one.

One piece of context to carry into it. Across the 47 brands we manage, Amazon delivered clicks on 3,086,624 distinct search terms between 1 May 2026 and now — tens of thousands per account, none of them chosen by anybody. That is not a list a person reviews thoughtfully once a week. Whatever you buy has to cope with that shape of problem, and your test has to be capable of detecting whether it did.

Step one: pre-register the metric, and do not make it ACOS

Write the metric down before the trial starts, in a document with a date on it. This single act removes most of the ways an evaluation goes wrong, because it stops you from selecting the number that flatters the decision you already made.

The metric we would use, and the most portable idea in our own practice, is marginal ACOS — the ACOS of the money you added, rather than the account's blended rate. The arithmetic is one line: divide the change in spend by the change in attributed sales over the same window. Blended ACOS moves slowly and hides the decision you actually made; marginal ACOS asks a single answerable question, which is whether the extra dollars came back.

Across our managed accounts that is how we judge every spend change, and it regularly points the opposite way from the blended number. On one account an added $930 came back at a 22.35% marginal ACOS while the blended rate sat higher — meaning the correct move was to add more, where the blended figure would have said stop.

A single illustration of how large the gap can get, offered as a story rather than a benchmark, because it is one account on one occasion. On an eight-figure sports nutrition brand, restoring $588 of weekly defensive spend recovered about $8,400 of weekly revenue at a 7.01% marginal ACOS — the campaigns went from $3,403 to $11,791 of weekly revenue on a spend change small enough that nobody had flagged the original cut. Nobody should expect that ratio. The point is only that a blended average would never have surfaced the opportunity, and a trial measured on blended ACOS would have scored the tool that found it as neutral.

If you would rather not use marginal ACOS, use contribution after ad spend. Use anything except a ratio that improves when you do less.

Step two: hold back a control group, and know what it cannot control for

Pick a set of comparable ASINs — similar price, similar category position, similar review count, similar seasonality — and leave them entirely alone for the duration. No bid changes, no new targets, no budget moves. That is your control.

Then be honest about what a control group on Amazon can and cannot tell you, because this is where confident test results go wrong. Your products compete with each other in the same auctions. Push the test group harder and some of the volume it wins may come out of the control group rather than out of a competitor, which flatters the test twice — once by lifting the treated ASINs and once by depressing the baseline. Halo runs the other way too: a lift on a hero product can pull traffic to its siblings in the control set and mask a real gain.

Three practical mitigations, none of them perfect. Choose control ASINs from a different subcategory or a different parent where you can, so they are not bidding against the test group. Track total account revenue alongside the group-level numbers, because a test that improves the treated group while the account total sits still has told you something important. And record both groups' impression share, not just their sales, so you can see whether volume moved between them.

If your catalogue is too narrow to split — a handful of ASINs, one parent — say so and run a before-and-after with a longer window instead, and treat the result as weaker evidence rather than pretending the design was clean.

Step three: the window, which is longer than the free trial

Our working rule across managed accounts is blunt: sponsored ads are not judged on seven- to fourteen-day data at all. The decision window is four to eight weeks, and an account takes three to four weeks to settle after a meaningful spend change before its numbers mean anything.

The reason is a feedback loop rather than a preference. A short window reads ordinary variance as signal. The signal produces a cut. The cut produces a real decline. The decline reads as confirmation that the cut was correct, and the account ratchets downward through a series of individually reasonable decisions. The same trap works in reverse: a brand that restores spend and checks after ten days concludes the restoration failed, right before it would have worked.

Two consequences for your evaluation. First, sixty days is the minimum honest trial, which means most free trials in this category are too short to answer the question and you will be paying for part of the test. Budget for that rather than being surprised by it. Second, if the tool's first act is a significant spend change, your clock starts when the account settles, not when you switched it on — so the first three weeks are setup, not data.

And do not run the window through a tentpole event. If your test would straddle one, either finish before it or start after, and say which in your notes.

Step four: subtract the whole fee, on the basis the vendor actually publishes

A tool that improves performance by less than it costs has taken your money and a quarter of your year. So the last arithmetic is the tool's own price, in full — and this is where more evaluations go wrong than anywhere else, because pricing pages have tabs.

Check which billing tab you are reading, on any vendor including us. Nearly every page here shows a monthly figure and an annual figure, and the annual one is usually expressed per month. Read the wrong one and you will build a budget on a number nobody is charging you. Every figure below was read on 21 August 2026 with its basis named.

• Ad Badger publishes six bands set by monthly ad spend, with both bases side by side: $275/month billed monthly or $2,550 a year up to $5,000 of spend, then $440 or $4,080 up to $25,000, $660 or $6,120 up to $75,000, $920 or $8,500 up to $225,000, $1,375 or $12,750 up to $750,000, and $1,830 or $17,000 up to $1.5M. No percentage of ad spend, no quote, no call. That transparency is rarer than it should be.
• Teikametrics lists Essentials at $179/month billed monthly or $149/month on annual billing, up to $10,000 of monthly spend, with Advanced and Enterprise quoted individually and described as plus 3% of ad spend over $10K.
• Helium 10 lists Platinum at $129/month billed monthly or $99/month on annual billing and Diamond at $359 or $279 on the same two bases, with a 2% management fee on PPC spend managed through Helium 10 Ads on Diamond. That fee overtakes the Diamond subscription itself at around $18,000 of monthly managed spend on the monthly basis, or around $14,000 on the annual one — which is to say the crossover you were told about depends on a tab.
• Scale Insights prices by automated ASIN, from $78/month billed monthly for five to $688 for a hundred, with a separate plan charging 1% of ad spend for unlimited ASINs. Thirty-day trial, no card.
• Perpetua lists Essentials at $695/month and Growth at $695/month plus a percentage of ad spend that the page does not quantify — the figure to get in writing before the trial, not after. No currency is named anywhere on that page.
• sellerboard prices on monthly orders rather than spend, from $19/month billed monthly or $179 a year up to $79/month or $759 a year, with a one-month free trial.

Take the annual total, divide by twelve, and see whether it matches the monthly figure. If it does not, you are looking at two different bases and must say which is which. Do it on our number too.

What the result licenses you to conclude, and what it does not

You now have a number. Be careful what you do with it.

A clean sixty-day test on your own catalogue tells you what that tool did on that catalogue in that period. It does not tell you what it will do next year, on a wider catalogue, or after Amazon changes a placement type. It is one account, which is the same limitation we apply to every account-level figure on our own pages. Write down the caveats alongside the result while you still remember them.

There are also three outcomes people misread. A flat result is a result — if a tool changed nothing measurable, it is not worth its fee, and that is a decision rather than a failed test. A large improvement in the first fortnight is usually repair, not the tool: switching platforms forces somebody to look at the account properly for the first time in a year, and a person did that. Re-run the comparison from week three. And a worse result is not automatically the tool's fault — check what else moved, because your own changes are the variable you actually controlled.

Then match the answer to a spend band, which is where the shortlist finally comes in. Under roughly $2,000 a month, none of these are worth buying; use Amazon's own console, a profitability tool, and a weekly search-term review by hand. Between $2,000 and $25,000, Ad Badger for a forecastable flat cost or Teikametrics if you would rather set a goal than write rules, and it is the pick if Walmart is a real channel. Above $25,000, every serious platform bids competently and the constraint stops being the tool.

When the test says the tool was never the problem

Two failure modes survive a well-run trial, and both mean the answer is not on the shelf.

The first is that nobody opens it. If your honest answer to who runs this on a Tuesday is that nobody does, the trial will show a flat result for every option, and the correct reading is that you are shopping in the wrong category. That is the gap Dr. PPC was built for. It reads the whole ad account, writes a strategy for each product against that brand's real unit economics, and proposes each change with the evidence behind it, a measurement plan and a rollback trigger attached — so the discipline this page describes is the product rather than something you have to impose on yourself. You set the autonomy level, from every change waiting on a click through to fully autonomous inside guardrails you agreed. Inventory risk, pricing, launches and creative always come back to a person. $300/month plus 3% of ad spend, capped, month to month, first 30 days free, with Orbit — analytics, search terms, campaign and ASIN profitability, keyword and traffic tracking, inventory, finance and the BSR, buy box, price and fee trackers — included at no extra cost. Dr. PPC and Orbit are products of Full Circle, a full-service Amazon management company with $500M+ in managed revenue across 100+ brands.

If you would rather hold the controls yourself, buy Ad Badger or Teikametrics instead and we mean that. Paying for judgement you intend to overrule is the worst value in this whole market.

The second failure mode is that the account is already efficient and growth stalled anyway. When sponsored placements already win the terms worth winning and the efficiency work has been done twice, more bidding sophistication returns very little, and the honest next question is reach rather than another point of ACOS — Dr. DSP for that as a product, or reMKTR to buy it as a service. And if the trial keeps being ruined because campaigns go dark when units run out, that is a supply problem wearing an advertising costume: Dr. Stock is the right purchase, and no tool on this page will ever surface it for you.