Reading Amazon in three dimensions.
The free notes let you see the shape. This course gives you the method to find it on your own account: the decomposition, the queries, the controls and the checklists, unit by unit.
Operators with data
Sellers with real volume, in-house ad leads and agency operators who already know the console and want to know why the numbers move.
A method, not a verdict
Ways to decompose a metric, test whether ads are incremental and catch a reading trap before it costs you a decision.
Setup and promises
No account setup walkthroughs, no compliance advice, no tool recommendations and no promised results.
Six modules. One skeleton in every unit.
Each unit opens with a popular claim or something a seller actually said, then walks the same eight steps: how the crowd reads it, the premises, how we take it apart, the evidence, what would prove it wrong, how you run it yourself, and where it stands in the ledger.
The Shadow Problem
- Every report is a shadowACOS, AMC and SQP each see one face of the account.
- Dashboard vs. mapLagging indicators versus structure; what matters is whether the account grows.
- Why one operator can do it nowUsing AI to organize and join AMC with the other reports for an investigation.
- How to read this courseExperience vs. hypothesis, the four verdicts, and how to use the public ledger.
- Time is the third axisNumbers alone are flat. Where shoppers are in their journey changes every number, which is why customer mix comes first.
The Mountain: Ticket, Auction, Reaction
- The mountainShoppers are already on the mountain; your bid only decides how hard you reach for them. Why ads cost so much, especially at the start.
- Not a second-price auctionThe bid is close to the price you actually pay.
- One seller, one entryWhen several of your campaigns chase the same term, only one enters the auction.
- When bids stop bindingIf actual CPC sits well below the bid, raising the bid does nothing.
- Premium is a bidA rest-of-search placement premium acts exactly like a higher bid.
- Reaction comes lastConversion rate can't be read until the ticket is solved; customer mix and placement move it as a whole.
- Open question: more ticketsHow more shoppers become eligible to see you. Unsolved, and tracked in the open.
Five Reports, Five Calibers
- What each report countsBusiness Reports, SQP, ad reports, AMC and finance; three layers of visibility.
- Time trapsAd report days follow US Pacific time, and the latest week's attribution isn't finished.
- Reports that keep changingBusiness Report figures still revise downward six weeks later.
- Window mismatchesWeekly against daily data; duplicate rows across a variation family.
- Attribution breaksA change in attribution rules that looks like a collapse.
- Profit that adds upThe finance caliber, and why the deposit is a fraction of what you sold.
- Watch the intent group, not the wordRead SQP by groups of searches with the same buying motive, and compare your add-to-cart rate with the market's.
Decompose, Then Prove
- Why ACOS risesSplit it into CPC, conversion rate and order value; split orders into impressions × CTR × conversion rate.
- Three reasons exposure fallsYou shrank, the market shrank, or eligibility withered. The remedies are opposite.
- Is it incremental?Four tests for whether ads are eating organic orders.
- Controls that tell the truthHoldout groups, placebo groups and same-product controls.
- Free experimentsZero-spend windows and the true-zero gate.
- When did it really break?Lower the time grain to find the break; run placebo tests in time.
- The residual is a ceilingWhat remains after the measured parts is an upper bound, not an answer.
The Customer Cycle
- Who counts as newProduct-level vs. brand-level new customers; measure the lookback window instead of assuming it.
- The fake 100% newAn illusion created by AMC privacy thresholds.
- Repurchase curvesEach product has its own repurchase cut-off.
- Entry vs. terminal productsCross-purchase: which products bring people in and which are the last stop.
- Acquisition vs. harvest flowsWhere acquisition and harvesting spend actually flow.
- Did we reach new people?Compare warm-to-cold ratios, not the share of returning customers.
- New and returning, by placement and positionEach placement, and each position within it, faces a different mix of people.
Designing What You Watch
- Direction firstDecide growing, flat or declining before looking for heroes or culprits.
- Drill to the keywordStopping at the channel level is not an answer.
- Freshness tiersDon't drill into incomplete weeks; that is how fake collapses are born.
- Right pair, right roleAcquisition reads CPM × CTR; harvesting reads conversion rate × ACOS.
- Three-layer cadenceCompare to yourself, to the product family, then the monthly review.
- Good news needs action tooA positive signal calls for a move as well.
- Top-10 concentrationHow much of a keyword group the top ten products hold, and which rivals are buying exposure, holding a stable base, or winning clicks.
- Write the prediction firstBefore you change anything, write down what should happen, when you'll check, and what would make you stop.
Free lets you see it. The course lets you do it.
The free notes are meant to be true enough to trust on their own. That is where the trust is earned.
| Part of each unit | Free note | Course |
|---|---|---|
| ①–③ The claim, the crowd's reading, the premises | Full | Full |
| ④ How we take it apart | The conclusion and one key screen | The complete decomposition, including how AI joins the reports |
| ⑤ Evidence | One example | All the evidence, several cases, the failures too |
| ⑥ What would prove this wrong | One sentence | The full set of disproof conditions |
| ⑦ How you run it | — | Step-by-step operations, query templates, checklists, the research-method handbook |
| ⑧ Ledger status | Public | Public, plus updates when a verdict changes |
Trust built from premises, not from screenshots of revenue.
I can't show you who I am, which accounts these are or what they sell. So the course is built to be checked instead of believed.
Every claim carries its own disproof condition
Each unit states the result that would overturn it. If you see that result on your account, the claim is wrong for you, and I want to hear about it.
Premises first
Before any conclusion, the premises are laid out and marked ✅ ⚠️ or 🔶. You can reject a premise and see exactly which conclusions fall with it.
Reproducible on your account
Part ⑦ gives you the steps, queries and checklists to run the same test on your own data.
Reversals are published
When a verdict flips, the reversal stays in the public ledger with the premise that broke.
Experience and hypothesis, labelled
You always know whether a claim was measured on my own accounts or is a framework still under test.
No conflicts of interest
No tools for sale, no affiliate links, no testimonials. Scale is stated once, in ranges, and never shown off.
Bringing AI into your analysis
AI assists; the judgment stays yours. Each lesson shows how to bring AI into your own operating analysis: which reports to hand it, how to line up their definitions and time windows, how to cross-check a claim from several angles, and how to catch it when it gets something wrong. The questions, the verdicts and the disproof conditions remain human work.
Screens shown in the course use de-identified or demonstration data. No client, product or absolute revenue figures appear anywhere.
The full series
Price to be announced
- All 6 modules, 41 units, video and text
- Part ⑦ in every unit: steps, queries, checklists
- Downloadable attachments and the research-method handbook
- Ledger updates when a verdict changes
Personal use only. See terms.