Amazon PPC is a system, not a bid sheet.
Most accounts get managed by reflex — sales drop, so bids move. I trace every change through traffic, conversion, stock, price, the Featured Offer and the account's own edit history before touching anything. Then I fix the highest-leverage cause first, and scale only once the data has earned it.
This page is the whole method, published. No decision rule is held back for the call.
Most PPC problems are not PPC problems
Advertising owns two links in this chain. It buys visibility and it buys the click. Everything after that is the listing, the offer and your stock position — and a break anywhere downstream still shows up in the ad report, which is why so many conversion problems get misdiagnosed as bidding problems.
scroll to see the full diagram →
Where each number actually comes from
Sessions and conversion rate do not live in the same system as clicks and spend. Mixing them is the most common analytical error in Amazon reporting, and it produces confident wrong answers — a conversion rate calculated against ad clicks instead of sessions can be out by a factor of three.
Business Reports
Seller CentralSessions · CVR (units ÷ sessions) · Featured Offer % · OPS · ASP · units
What happened on the listing — every visit, paid or organic, and whether it converted.
Advertising Console
Campaign Manager / Ads APIImpressions · clicks · spend · CPC · ad orders (7d) · ad sales (7d) · top-of-search IS
What happened in the auction — only the traffic you paid for, attributed on a 7-day window.
What the numbers are telling me
These are the patterns I read first. Each one points somewhere different, and getting the direction right is most of the job — the wrong first move on a falling account usually costs more than doing nothing.
scroll to see the full diagram →
A traffic problem, and the listing is cleared. Fewer shoppers arrived; the ones who did bought at the same rate or better.
Investigate reach — recent bid and budget edits, lost impression share, stock, and whether either week contained an event.
A conversion problem. The traffic still came and the page stopped closing it.
Audit the offer: price, images, reviews, delivery promise, Featured Offer share, and whether the query actually matches the product.
The ad account is not the main culprit. Something upstream of advertising — listing, organic rank, price, demand — moved as well.
Apportion before you act. Comparing the two deltas tells you how much of the decline the ad account can actually explain.
Competitiveness was removed faster than demand changed. A cut from full funding to a fraction of it is not a controlled test.
Rebuild in steps rather than restoring in one move, so cause and effect stay separable.
An offer problem, not an ads problem. Losing the Featured Offer suppresses conversion and ad eligibility at the same time.
Repair eligibility first — price, competing offers, delivery speed, seller health. More spend cannot fix a product shoppers can't reliably buy.
Brand defence is subsidising the number. Acquisition is losing money underneath a total that looks fine.
Split branded and non-branded permanently. Brand CVR comes from people already looking for you — it never predicts cold-audience performance.
A budget constraint, not a bid constraint. It is winning when it runs; it just isn't running often enough.
Raise the budget in steps. Raising the bid first pays more for traffic you were already winning.
Possibly not an auction you're losing. Practitioners report AI-assisted results taking positions above sponsored placements on mobile, which displaces the slot rather than outbidding you. Amazon hasn't confirmed this, so treat it as a check rather than a conclusion.
Read the placement report before touching the bid. If the position is being displaced rather than lost, paying more buys nothing — the route back is organic eligibility, which is a listing job.
Proven waste, not a bidding problem. It has had its chance to convert and didn't.
Negate it at the right level — campaign, ad group or account — rather than bidding it down forever.
A promising signal, not proof. The sample is too thin to carry a budget decision.
Fund it as a controlled test on its own line, and scale only once click and order counts support it.
Not a decline. Event days and baseline days are different demand curves and shouldn't be averaged together.
Strip the event days, recompute the base, and compare like with like before calling anything a drop.
“My ACoS is too high” — six causes, one fix each
High ACoS is a symptom, not a diagnosis. It has at least six distinct root causes and they need different fixes — only the last one is solved by lowering a bid. Reaching for the bid first is why so many accounts get cheaper and smaller at the same time.
scroll to see the full diagram →
Each branch, and how you tell them apart
The listing isn't converting
CVR is below what the category and the product's own branded terms achieve.
A listing problem wearing an ads costume. Fix the page; a lower bid just buys less of the same failure.
The spend mix is discovery-heavy
A large share of spend sits in Broad and Auto rather than proven Exact and Phrase.
Structural. Harvest what converted into Exact, negate what didn't, and let discovery run at a size that matches its job.
Nothing has been negated
The search term report shows spend against queries the product does not serve.
Negate at the correct level — campaign for irrelevant-here, ad group for internal cannibalisation, account for never-converts-anywhere.
Winning campaigns are budget-capped
The campaigns clearing target are hitting their ceiling daily while weaker ones run free.
Not an ACoS problem at all — a distribution problem. Move budget toward what already clears.
The auction is genuinely expensive
CPC is high across the board and CVR is fine; the category is simply competitive.
Decide whether the term is worth owning at that clearing price. Sometimes the right answer is to stop competing for it.
You're over-bidding
Everything above has been ruled out and CPC still sits well above what a click is worth.
The one case where lowering the bid is the answer. Reprice off CPC actually paid, and step toward the target.
Seven steps, in this order
The order matters more than any individual step. Fixing bids before fixing structure just optimises a leaky funnel, and skipping straight to step six is what produces accounts that have been optimised for a year and still lose money on acquisition.
Set the business target
What does this account have to do, in the business's own numbers?
- Margin, landed cost and break-even ACoS — the target comes from the product's economics, never from an industry average
- Budget ceiling, stock position, days out of stock and inbound units
- A dated change log: promotions, price changes, stockouts and every account edit
A written target the business can actually fund — plus the context needed to tell an event week apart from a real decline.
Check the window before you trust the delta
Is this comparison even fair?
- Did either week contain a retail event, a promotion or a price change?
- Was there a deliberate bid or budget edit — and is it in the change log?
- Any days out of stock, or a Featured Offer loss, inside the window?
Strip the contamination and recompute the base. A raw week-over-week number that contains an event is not a result, it's an artefact — and acting on it is how good accounts get broken.
Find the real cause
Where in the sales chain did the number actually break?
- Walk the chain in order: spend → impressions → clicks → sessions → CVR → sales
- Split ad sales from non-ad sales and compare both deltas — it apportions the blame
- Read TW against LW at parent level to find the symptom, then drill to the child that moved
- Rule alternatives out explicitly: if stock held all week, inventory is eliminated, and say so
A cause supported by several connected signals, with the alternatives named and eliminated. One metric moving is a symptom; three moving together is a diagnosis.
Check retail readiness
Can this page convert traffic before I buy more of it?
- Title, images, bullets, A+ Content, reviews, price, delivery promise, stock, Featured Offer share
- Every advertised child ASIN, not just the parent listing
- Query-to-product match: a 'sugar-free' search must not land on a product containing sugar. Same for flavour, size, format, pack count
Fix, pause or block. A mismatch between the search and the product is not a bidding problem, and no bid is low enough to make it one.
Separate the account until the cause shows
What is a blended number hiding?
- Branded versus non-branded — defence versus acquisition, always reported apart
- Match type: Exact and Phrase hold proven intent, Broad and Auto are discovery
- Target versus customer search term — you bid on one, shoppers type the other
- Placement, product targeting and display kept on their own lines
Separate budgets and separate reporting for each job, so a strong brand number can never mask weak acquisition.
Make the smallest useful change
What is the least I can change and still read the result?
- Stop proven waste first — negate and pause before touching anything that works
- Reprice bids off the CPC actually paid, never the bid you set or the suggested range
- Harvest and negate are different jobs with different criteria — don't conflate them
One change at a time, each with a review date attached. Aggressive account-wide edits destroy your ability to attribute what happened next.
Prove it before scaling
Has this earned more money, or has it just had a good week?
- Return held across enough clicks, orders and days — not one good day
- Retail readiness confirmed before discovery restarts
- Budget treated as a ceiling, with scale money released conditionally
Release the next tranche, hold, or reverse. Every action ends with a metric, a checkpoint and a rule for what happens next.
How a 30% drop stops being a 30% drop
This is the sequence in practice — each read narrowing what the last one left open. Notice that the headline number survives about eight seconds.
Illustrative figures — the sequence is the point, not the numbers
The number that starts the conversation, and on its own it is worth nothing. Nothing gets actioned from here.
The comparison is contaminated. Strip the event days and recompute against baseline days only — the underlying week may be close to flat.
The listing is cleared. Conversion improved, so fewer people arrived rather than more people refusing to buy. The investigation turns outward.
Advertising made it worse but did not cause it. Most of the decline sits outside the ad account, which changes what gets fixed first.
The parent-level number was a symptom. Now find what is true of that one variation and not the others.
Inventory is ruled out explicitly. Featured Offer share is the isolated cause — and it suppresses conversion and ad eligibility simultaneously.
Sequenced, gated, and with the pause trigger stated before recovery starts rather than after something breaks.
The numbers I work to — and why
These are starting points, not laws. Every one moves with your margins and your data volume — an account doing 40 orders a day can read a change far faster than one doing four. What doesn't move is that each number has a reason attached, so you can argue with it.
Target CPC, worked backwards
CPC × ROAS ÷ target ROASRepricing off the bid you set tells you nothing — you rarely pay it. This starts from the CPC actually paid and the return actually earned, so the new target is anchored to what the keyword really does. Then step toward it rather than jumping.
Data before a verdict on a keyword
10+ clicksBelow that, a zero-order keyword is noise, not evidence. The real floor scales with the product's conversion rate — if it converts at one in ten, ten clicks is the minimum before absence of a sale means anything at all.
Bid and budget moves
10–20% per stepBig enough to move the auction, small enough that the result is still readable. A cut from full funding to a fraction of it removes competitiveness so fast you learn nothing about demand.
Hold between steps
48–72 hoursAttribution and delivery both lag. Reading a change before the window closes means reacting to noise. Higher-volume accounts can move faster — the rule is enough data, not a fixed clock.
Before pulling reports on a new campaign
2 weeksNew campaigns deliver unevenly while they find their footing. Judging week one produces a decision about the ramp, not about the keyword.
Opening bid on a new target
50–100% above suggestedCounterintuitive, and it saves money. Amazon's suggested bid is calibrated to get an ad showing, not to make you profitable — it knows nothing about your margins. Opening above it buys the placement and the impressions needed to learn something, and the auction then settles toward the real clearing price. Bidding the suggestion often buys nothing at all, which is the most expensive outcome there is: spend with no data.
Know your true bid at top of search
base × modifier × dynamicThese stack multiplicatively and it catches people out. A $1.00 bid with a 200% top-of-search modifier is $3.00 there — and with dynamic bidding set to up-and-down, Amazon can raise that by up to another 100%. Most accounts running both settings were never told they were bidding six dollars.
Don't read the last few days
7-day attributionAn order counts against a click for seven days after it, and reporting lags on top of that. Yesterday's ACoS always looks worse than it will turn out to be, so cutting bids on a 48-hour panic is optimising against incomplete data. It's why weekly beats daily.
Loss limit before a pause
≈ 1 unit of margin, no orderWaste is defined against the product's economics, not a flat dollar figure. Once a term has spent roughly what a sale would earn and produced nothing, it has answered the question.
Featured Offer before budget
95%+ shareBelow that, a share of your paid clicks land on a page that can't complete the sale. Funding a campaign in that state buys someone else's conversion.
Conditional scale pool
released weekly, or not at allBudget is a ceiling, not a quota. Money only moves to targets already clearing the agreed return; if nothing qualifies, it stays unspent. Unspent budget is a result, not a failure.
The bid you set is not the bid you make
Five things most accounts do that quietly cost money
None of these are stupid. They're all reasonable-sounding advice that stops being true once you know how the mechanics underneath actually behave.
Bid what Amazon suggests
The suggested bid is calibrated to get an ad serving. It has no idea what your product costs to make, what Amazon takes, or what margin you need — so it optimises for Amazon's fill rate, not your profit.
Derive the ceiling from what a click is worth: revenue per click against your target return. That number is yours; the suggestion is theirs.
Top of search is always the best placement
It averages the highest click-through and conversion across the whole market, which is not the same as being right for your query. Consideration-stage searches often convert better from product-page placement at a materially lower cost.
Read the placement report per campaign. Let each query type earn its own placement rather than paying the premium everywhere.
Optimise every day
Attribution runs seven days behind and reporting lags on top. The last few days always look worse than they'll settle at, so daily changes are reactions to data that hasn't finished arriving.
Change on a weekly rhythm against a window long enough to have settled. Move enough to matter, then let it read.
A high ACoS campaign is a bad campaign
ACoS only counts revenue Amazon attributes to the click. It cannot see the organic sales the ad drove through ranking, or the value of a first-time buyer versus a repeat one.
Judge campaign efficiency on ACoS against break-even, and business health on TACoS. A campaign can fail the first test and pass the second — that's a launch, not a leak.
Add more keywords for better coverage
Every additional target splits the same traffic into thinner slices, and thin data can't be read. You end up with three hundred keywords that each have four clicks and no keyword you can make a decision about.
Fewer targets, each with enough volume to produce a verdict. Coverage comes from discovery campaigns; control comes from a short list of proven terms.
What a high-ACoS campaign can look like when it's working
When I'll tell you not to run ads
Ads amplify whatever the listing already does. Amplifying a broken offer makes the problem bigger and more expensive, so there are conditions under which the right advice is to spend nothing until they're fixed.
Stock is about to run out
Advertising into a stockout burns spend and produces a rank drop you then pay again to recover. Days out of stock is a metric I check before budget.
Featured Offer share is unstable
You are buying clicks to a page where the buy button may belong to someone else.
The listing isn't retail-ready
Images, reviews, A+ Content and delivery promise decide the conversion rate. Ads amplify a listing; they cannot rescue one.
Price is materially off the category
No bid strategy survives being the expensive option on an identical product. That's a pricing decision, not an ads decision.
The query doesn't match the product
If the search asks for a feature the product lacks, every click is pre-failed. Block it — don't discount it.
What Amazon's AI layer actually changed
Amazon retired the Rufus brand in May 2026 and folded it into Alexa for Shopping, which now sits in the search bar itself rather than off to one side. Underneath it is COSMO — a semantic layer Amazon published research on in 2024, deployed across eighteen categories, that maps products to human context: what something is for, who it suits, when it gets used.
The practical consequence is narrow and specific. A shopper can now ask for “shoes for flat feet” and get sent products whose listings never contain that phrase, because the connection was learned from behaviour rather than matched from text.
scroll to see the full diagram →
- The product catalogue and your detail pages
- Customer reviews
- Community Q&A on the listing
- Information from across the web
Every one of those is a surface you already control. None of them is new.
“I don't think brands need to do anything different… what's most important is ensuring you've got the most up-to-date, factual information about your products available in the store.”
Amazon VP of Conversational Shopping, December 2025
Worth holding onto when someone quotes you for an AI optimisation package. The work is real, but it is listing work with a new reason behind it — not a new discipline.
Same four surfaces, different job
Attributes
Filled in once at listing creation, then never opened again.
Amazon states that complete, accurate attributes help the assistant recommend your products. Item type, intended use, audience, material, compatibility, dimensions — these are read.
Bullets
Keyword stacks. 'Durable premium BPA-free lightweight portable.'
A model composing an answer can quote a sentence. It cannot quote a keyword stack. Bullets that answer a specific buyer question are usable; adjective lists are not.
A+ Content
A design surface. Images with the copy baked into them.
Text is read. Copy locked inside a JPEG is invisible to anything reading the page, which now includes the shopper's assistant.
Q&A
Left to whatever customers happened to ask.
Amazon confirms community Q&A is one of the sources. An empty Q&A means the answer gets assembled from your reviews instead — which may not be the version you'd choose.
Sponsored Prompts — and you are probably already running them
Amazon generates suggested questions about your products, marks them Sponsored, and charges for the clicks out of your existing campaign budgets. They went generally available in the US in March 2026, and enrolment is automatic — so for most accounts this is already spending money.
Unlike the rest of the AI layer, it is measurable. There is a report: Ads Console → Reports → Sponsored Products → Prompts
Two reasons to pull it. It shows spend you may not know you have — and the prompt text tells you, in Amazon's own words, how its model currently describes your product. That is free feedback on whether your listing says what you think it says.
The long tail stopped being an edge case
People ask an assistant for things they would never type into a search box. Queries get longer and more specific, and those phrases arrive in the search term report looking like low-volume noise — one or two conversions each, easy to scroll past.
They are usually the opposite of noise. A shopper who specified the use case, the size and who it's for has pre-qualified themselves, and nobody is bidding hard against them.
Pull 90 days of search terms. Filter to queries of six words or more with at least one order. Anything on that list without its own exact-match target gets one, at a bid derived from what the click is actually worth.
Campaign architecture doesn't change for this. Where the budget sits does.
What nobody can tell you — including me
- There is no AI-attributed traffic in any seller report. Discovery through the assistant lands in the organic line, indistinguishable from a keyword search.
- Amazon has never published which listing fields the semantic layer weights, or by how much.
- Anyone showing you an 'AI visibility score' for your catalogue has built it themselves. It is an inference, not a reading.
- 'A10' is not an Amazon algorithm. A9 is the only version Amazon has ever named.
The first month on someone else's account
The most common way a new manager destroys value is restructuring in week one. Rebuilding campaigns resets the performance history they've accumulated, so a tidy-up you didn't need can cost weeks of learning.
Read before you write
Baseline the account and reconstruct what has actually been done to it. An inherited account's history is the single most useful thing nobody hands over — and without it, every number you see later is uninterpretable.
Stop the obvious bleed
Negating proven waste and pausing dead targets is reversible, cheap, and doesn't disturb anything that's working. It is the only thing I'll do fast.
Leave the structure alone at first
Restructuring resets accumulated performance history. On an account that's merely untidy rather than broken, a rebuild costs weeks of learning to fix something cosmetic. Structure gets changed where structure is provably the constraint — not because it isn't how I'd have built it.
Then rebuild, in sequence
Retail readiness, then proven intent, then discovery, then scale. Each stage gated on the one before it.
A business I took over, in its own numbers
The owner was running his own Sponsored Products campaigns and losing money on every click. Eighty clicks had produced a single order — so this wasn't an account short of spend, it was one short of relevance. The diagnosis said structure, not bids, and the rebuild came before any bid moved.
$74.17 spend → $59.99 sales.
80 clicks · 0.50% CTR · 1 order · $0.93 CPC
$35.01 spend → $149.97 sales.
17,843 impressions · rebuilt on auto close-match
Cutting ACoS by throttling an account is trivial and worth nothing — spend less, sell less, report a better ratio. That is not what happened here. Sales went up 2.5× while spend roughly halved. The ratio improved because the structure did.
Small budget: this one is about the structure, not the scale. Both console screenshots are on the results page so you can read the numbers yourself rather than take mine for them.
What I won't do to your account
- deny
account_wide_changescause and effect stop being separable - deny
scale_from_thin_samplea good ratio on nine clicks is not a result - deny
discovery_before_retail_readyyou pay to generate unreliable data - deny
blended_brand_reportingdefence quietly subsidises acquisition - deny
spend_to_hit_the_budgetthe ceiling is not a target - deny
acos_cuts_by_throttlingsmaller is not the same as better
Terms I'll use in your audit
Grouped by where each one gets used rather than alphabetically, because the grouping is itself part of the method. Defined once here so the report reads like a report instead of a decoder ring.
scroll to see the full diagram →
Reading a week
The comparison vocabulary. Most bad decisions start with an unexamined delta.- TW / LW / W-2This week / last week / two weeks ago
- Mon–Sun periods, read as three parallel columns rather than a chart. One delta gives direction; three columns give trend, and trend is what stops you overreacting to a single bad week.
- WoWWeek over week
- The percentage change between periods. Useful for spotting movement, useless for explaining it — and worthless entirely if either window contains an event.
- LY comp weekLast year, matched by ISO week
- Matched on week number, not calendar date, so the days of the week and the seasonal position line up. Month-over-month year comparisons quietly compare a five-weekend month to a four-weekend one.
- Change logThe account's edit history
- Every change, dated, with what was touched and why. The working standard: if it isn't in the log, it didn't happen. It's also the first thing I read when a number moves — half of all 'sudden' changes have a deliberate edit behind them.
The catalogue
Where a parent-level symptom turns out to have a child-level cause.- PASINParent ASIN
- The umbrella listing holding the variations. Reviews and ranking largely live here, which makes it the right level to start a review at — and the wrong level to act from.
- CASINChild ASIN
- The individual buyable variation. Stock, price and Featured Offer are decided here, so a parent-level problem nearly always resolves to one child.
- Featured Offer %Formerly Buy Box %
- The share of page views where your offer owned the buy button — a continuous percentage per child, not a yes/no per product. Losing share suppresses conversion and ad eligibility at once.
- Days out of stockCounted days at zero
- Distinct from current stock level. Stockout days suppress organic rank, so the count explains why a recovered product still underperforms. Read alongside inbound units to estimate when recovery actually starts.
Business Reports metrics
From Seller Central. These describe the listing, and include organic traffic.- SessionsUnique visits to the listing
- All traffic, not just paid. Comparing sessions to ad clicks is how you see what share of your visits you are renting.
- CVRUnits ÷ sessions
- The conversion rate that matters, because it is measured against every visit. Paired with sessions it becomes a fork: which of the two moved tells you whether to look outward at traffic or inward at the page.
- OPS / ASPOrdered product sales / average selling price
- OPS is revenue; ASP is OPS ÷ units. A flat OPS with a falling ASP means you sold more units for the same money — a mix shift, not growth.
Advertising Console metrics
From Campaign Manager. Paid traffic only, attributed on a 7-day window.- ACoS / TACoSAd cost of sales / total ACoS
- ACoS is ad spend over ad sales. TACoS measures it against total sales, which is the one that shows whether paid is building something. Falling TACoS with growing sales means organic is compounding underneath the ads.
- CPC / RPCCost per click / revenue per click
- What a click costs and what a click is worth. Bidding is the business of keeping the first below the second — which is why RPC, not ACoS, is the number a bid actually gets set from.
- Top-of-search ISImpression share at top of search
- How often you appeared in the top placement out of the times you were eligible. High conversion with low share is a budget problem; low share with a competitive CPC is a bid problem. Different fixes.
- 7-day attributionThe default reporting window
- An order counts against a click for seven days after it. It's why yesterday's numbers are never final and why reading a change too early misleads you.
- Blended ROASAll sales ÷ all spend
- The number that hides the most. Brand and non-brand averaged together lets defence subsidise acquisition, and the total looks fine while new-customer traffic loses money.
Targeting
The distinction most of the work lives inside.- Target vs search termWhat you bid on vs what they typed
- You bid on a target; the shopper types a search term; one target matches many terms. Nearly all real optimisation happens in the gap between the two — and conflating them is the clearest sign someone hasn't opened a search term report.
- Match typesExact · Phrase · Broad · Auto
- Exact and Phrase carry intent that has already proven itself. Broad and Auto are discovery — they find new terms and they generate noise when the product isn't ready to convert.
- Auto sub-typesClose · loose · complements · substitutes
- An Auto campaign is four different campaigns wearing one budget. Close match usually converts and is where harvesting comes from; loose match is where most Auto waste lives; substitutes puts you on competitor pages; complements puts you beside things people buy alongside yours. Bidding them as one number is leaving the obvious efficiency on the table.
- BleedSpend with nothing to show for it
- Spend against queries that have never produced an order over a meaningful window. In accounts that haven't been negated properly it's routinely a fifth to two fifths of the budget — which makes it usually the fastest money available in an audit.
- Lost impression shareLost to budget vs lost to rank
- The impressions you were eligible for and didn't get. Losing them to budget means the campaign ran out of money; losing them to rank means the bid didn't win. Same symptom, opposite fixes — and raising the bid on a budget-capped campaign just makes it run out sooner.
- NTBNew-to-brand
- Orders from shoppers who haven't bought from the brand in twelve months. Reported on Sponsored Brands and Display, not Sponsored Products — which is why SB gets undervalued when it's judged against SP on ACoS alone. Acquiring a customer and re-selling an existing one are not the same purchase.
- Harvest vs negateTwo different jobs
- Harvesting promotes a converting search term into its own controlled Exact target. Negating blocks a non-converting one from spending again. Different criteria, different effect — using the words interchangeably usually means doing neither properly.
- Negative levelsAd group · campaign · account
- Ad group stops your own campaigns cannibalising each other. Campaign blocks a term irrelevant to that product. Account blocks one that never converts anywhere. Choosing the wrong level either leaks spend or suppresses a term that works elsewhere.
- SP / SB / SDSponsored Products, Brands, Display
- Direct conversion, brand defence and upper funnel, remarketing and conquest. Different bid mechanics and different jobs — treating them as one 'campaigns' bucket is how budget ends up in the wrong funnel stage.
The reports I actually open
Which report answers which question, and in what order.- Search term reportSTR
- Every query that triggered an ad, with what it cost and what it returned. This is where harvesting and negating both come from, and it's the first report I open on any account. Paid traffic only — a term with no ad orders may still be selling organically.
- Search query performanceSQP · Brand Analytics
- The same queries, but total market — organic and paid together, with your share at impression, click, cart-add and purchase. It shows where you lose shoppers relative to competitors, which the STR structurally cannot. Needs Brand Registry. Complementary to the STR, not a replacement.
- Placement reportBy ToS / rest of search / product pages
- Where the money actually went. Top of search usually costs the most and converts the best — but not always. On consideration-stage queries, product-page placement often out-earns it, and you only find that out by looking.
- Budget reportCap-out timing
- Which campaigns hit their ceiling and when. A good campaign capping at midday is lost revenue; a wasteful one capping at midday is a structural problem. Same symptom, and the fix is opposite.
- Bulk sheetsBulk operations
- Campaign management as a spreadsheet — download, edit in bulk, upload. Past about fifty keywords, making changes one at a time in the UI stops being viable, and consistency across a systematic bid change is the whole point.
What the free audit actually gives you
It's steps one to four, run on your account, delivered as a ~20 minute video. Not a form, and not a PDF of screenshots you could have taken yourself.
- The root cause, traced through the sales chain rather than guessed
- The three highest-leverage fixes, in the order they should happen
- What to pause, what to protect, and what to leave alone for now
- What to test next, and the sample size it needs before it counts
- The checkpoint — what number, reviewed when, and the rule for continue, pause or reverse
If the audit finds the problem isn't advertising, it says so. That happens more often than the industry likes to admit.
Questions I get asked
Do you start by lowering bids?
No. Lowering bids is the most common first move and it's usually wrong, because it treats every problem as a price problem. A high ACoS has at least six root causes and only one of them is over-bidding — the rest are conversion, spend mix, missing negatives, budget distribution or an expensive auction. Bids come down when the evidence points there, not before.
Why look at the listing during a PPC audit?
Because ads only buy the visit. The listing and the offer are what turn that visit into an order. Sending more traffic to a page that can't convert produces more spend, not more profit — and it's the fastest way to make a small problem expensive.
What's a good ACoS?
There isn't a useful universal answer, and anyone who gives you one hasn't asked about your margins. A good ACoS sits inside the product's economics and the stage it's at — a launch buying rank and a mature product extracting profit should not be run to the same number. I start from break-even and work to a target the business can actually fund.
When do you use Broad, Auto or Sponsored Display?
After the listing, offer and product matching are sorted. Broad and Auto are how you find new search terms, but run them over a product that can't convert and you're paying to generate unreliable data. Wider display audiences come later still — a strong branded conversion rate says people already looking for you buy; it says nothing about whether a cold audience will.
How fast do you change bids and budgets?
In measured steps with a hold between them. Large swings either remove traffic faster than you can read the effect, or bury the cause under three simultaneous changes. I move enough to matter, then wait for enough data to decide whether to continue, pause or reverse.
What happens in the first month on an account you've inherited?
Mostly reading. I baseline the account and reconstruct what's been done to it, then stop obvious waste — that part is reversible and safe to do quickly. What I don't do is restructure immediately. Rebuilding campaigns resets the performance history they've accumulated, so on an account that's untidy rather than broken, a rebuild costs weeks to fix something cosmetic. Structure changes where structure is provably the constraint.
Would you ever tell me not to run ads?
Yes, and it comes up more than you'd expect. If stock is about to run out, the Featured Offer is unstable, the listing isn't retail-ready or the price is materially off the category, more traffic makes the problem more expensive rather than smaller. Fixing the constraint first is cheaper than advertising around it.
What if the data says the problem isn't PPC at all?
Then the audit says that. A fair amount of what looks like an advertising problem turns out to be stock, pricing, a lost Featured Offer or a listing that stopped matching what people search for. Selling you ad management for a conversion problem would waste your money and my time.
Want the root cause, not another bid change?
The free audit runs the first four steps on your account and tells you what's actually wrong.
Independent specialist. Not affiliated with or endorsed by Amazon.com, Inc.