The pro fault tree: when the numbers drop
- You decompose every drop into three measurable parts before hunting causes.
- You follow a fixed tree per branch instead of a hunch.
- You recognise when a drop is not a drop — the noise threshold.
- You check causes by frequency, not by how interesting they are.
This is the closing lesson of the bonus track, and it binds everything before it into one tool. The trigger is always the same: a number has dropped, and the question is why. Most sellers answer it with a hunch and then work on that hunch — sometimes for weeks, sometimes in the wrong place. Professionals work a tree instead. It takes twenty minutes and nearly always ends with a cause rather than a suspicion.
1Step 0: is it a drop at all?
Before you search for anything, check whether there is anything to search for. From Growth L11 you know the noise threshold: at an average of 19 sales a day the normal band runs roughly from 10 to 28 — a day at 14 is not news.
| Observation | Assessment |
|---|---|
| One day outside the band | Notable, but not a case on its own — look again tomorrow |
| Three days in a row in the same direction | A pattern with a cause — start the tree |
| One day far outside (halved or zero) | Start the tree immediately — that is not chance |
| A weekly average below last week's band | Start the tree, even if no single day stood out |
Reacting to a single weak day: cutting the price, raising bids, swapping the image. The next day the numbers rise — because noise returns to the mean — and the action is credited as the saviour. That is how accounts accumulate years of measures that never worked but permanently cost margin.
2Step 1: the decomposition
Every revenue figure breaks into three factors. Before hunting causes, establish WHICH one fell:
Revenue = sessions × conversion rate × average order value
| What fell | What it means | Where to look next |
|---|---|---|
| Sessions | Fewer visitors arrive | Branch A: visibility |
| Conversion rate | The same number arrive but buy less often | Branch B: persuasion |
| Order value | Same buyers, less money per purchase | Branch C: mix and price |
| None of them clearly | The drop is spread out | Branch D: external causes |
This single question saves half the work. It answers whether you have a visibility or a persuasion problem — and those two share not one cause.
A shop takes less money. Three possibilities: fewer people come in, fewer of those who come in buy, or they buy less each. Before you rebuild the window display, stand at the door and count. Almost any shopkeeper would do that — and almost every seller skips it.
3Branch A: sessions fell — visibility
Check in order of frequency, not in order of interest:
- Is the offer buyable at all? Featured Offer lost, listing suppressed, stock at zero, offer inactive. The most common case of all, checked in thirty seconds (Listing L14).
- Was the advertising running? Budget exhausted, campaign paused, payment method declined. The second most common — and the most embarrassing when found late.
- Did anything change in the catalog? Title, category, images, attributes — by you or by someone else (lesson 1).
- Is indexing still there? The ASIN test with exact phrases (Listing L12).
- What does the search query data say? Which term lost impression share — and was it share or market volume (lesson 2)?
- Season and market. Is search demand falling overall? Then your share may be stable and you are hunting a fault that does not exist.
4Branch B: conversion fell — persuasion
- Price in context. Not your price but your distance to the competition — even when you changed nothing.
- Star rating and review count. A drop in rating, a jump in count (a variation split, lesson 4) or a new, highly visible criticism.
- Delivery promise. Longer delivery time, Prime badge gone, dispatch from another country.
- Page content. Image swapped, A+ content in review, bullets shortened, text stripped of formatting.
- Competitive field. A new seller with a better offer on the same terms — visible in the item comparison report.
- Traffic quality. New campaigns bring broader queries; more visitors with less intent lower conversion without anything being broken.
The AURELO set's revenue falls 34 % in a week. The decomposition shows: sessions minus 4 % (within noise), conversion from 14 to 9 % — branch B. Point 1: price unchanged, competition too. Point 2: review count down from 410 to 373. The cause is found in four minutes: the variation review split. Without the decomposition the search would have started at ranking and advertising — and found nothing there, because there was nothing there.
5Branches C and D: order value and external causes
- Branch C — order value: product mix shifted (a cheaper variant is pulling, lesson 4), a quantity discount is biting, a bundle expired, a promotional price is still active.
- Branch D — external causes: holidays, weather, a news event, an Amazon event day your category is not part of. These causes are real, but they are the LAST branch — not the first, because they feel so comfortable.
6The twenty-minute run
- Minute 0–2: check the noise threshold. A three-day pattern or an outlier far outside?
- Minute 2–5: decompose. Sessions, conversion and order value of the last seven days against the seven before.
- Minute 5–8: check buyability — Featured Offer, stock, listing status, ad budget. Four glances that between them settle the majority of cases.
- Minute 8–15: work the relevant branch, in the stated order.
- Minute 15–20: note the finding: date, affected metric, cause found, action, expected effect and a check date. Without those five lines you will repeat the whole search next time.
- Noise threshold checked before any searching started.
- Decomposed into sessions, conversion and order value — before hunting causes.
- Buyability checked first: Featured Offer, stock, status, ad budget.
- Branch worked in frequency order, not by interest.
- Your own changes of the last 14 days cross-checked (lesson 11).
- Finding recorded with a check date.
- Only ONE action taken — otherwise the next diagnosis is worthless.
7Expert insight: frequency beats interestingness
The mistake that makes diagnosis genuinely expensive is not a knowledge error but an ordering error: people check the causes they find most exciting first — ranking algorithms, sabotage, changes at Amazon — and the ones that most often apply last. The right order follows from frequency, not from curiosity.
As a rough ranking, from how often cases occur in seller accounts:
| Rank | Cause | Time to check |
|---|---|---|
| 1 | Something is not buyable: stock, Featured Offer, listing status | 30 seconds |
| 2 | One of your own changes in the last 14 days | 1 minute with a register |
| 3 | Advertising: budget, pause, payment method | 1 minute |
| 4 | Price or competitive field shifted | 3 minutes |
| 5 | Reviews: rating or count jumped | 1 minute |
| 6 | Season or market volume | 5 minutes |
| 7 | Catalog or indexing problem | 10 minutes |
| 8 | Outside interference or sabotage | last, with a timeline |
The remarkable thing is the ratio: the four most common causes are checked in under six minutes between them — and cover the large majority of cases. The most interesting cause sits at the bottom and takes longest. Invert the order and you spend hours on ranking theory while an offer has been unbuyable since Tuesday.
Three rules turn the tree into a reliable tool:
- Search first, act second. Taking countermeasures during the diagnosis changes the system you are measuring. Walk the tree to the end, then act.
- One action, one cycle. Change three things after a diagnosis and the next drop leaves you knowing nothing again — the next diagnosis starts from zero (lesson 1).
- “No cause found” is also a result. It usually means it was noise, or the market moved. Neither justifies an action. Doing nothing is a decision, and in those cases the right one.
And the thought that holds the whole bonus track together: every lesson in this track is a branch of this tree. Catalog mechanics explains ranks 2 and 7, search query data makes branch A measurable, the advertising lesson covers rank 3, variations and price mechanics ranks 4 and 5, fees and reimbursements explain why stable revenue can still yield less money, account health and brand protection cover the rare but serious cases, and the season lesson tells you when a drop is not a drop. Master the tree and you do not need to memorise the lessons — you know which one to look in.
A second case on the AURELO set: revenue down 60 % on a Monday. Rank 1 in thirty seconds: stock at zero. Cause found, diagnosis over. The actual work starts afterwards and is a different question — why did the reorder not trigger in time (Growth L7)? That is exactly the value of the tree: it separates the symptom question from the cause question and closes the first in minutes, leaving time for the second.
The human appetite for a good story is the strongest opponent of clean diagnosis. “Amazon changed the algorithm” explains everything and cannot be refuted — which is why it is so popular. An explanation that could explain any drop explains none. If a cause requires you to be unable to check it, it is not yet a cause but a consolation.
The pro track assumes the other three and repeats nothing from them. It opens once you have completed all three in full — every lesson quiz at 80 % or better, and every final quiz passed.
Not logged in? Your progress only counts with a free account.
Check yourself
6 quick questions — one at a time, instant feedback. With a free account your progress is saved.
Frequently asked
How often do I run the fault tree?
Only on a trigger — it is a diagnostic tool, not a routine. The trigger is in step 0: three days in the same direction, an outlier far outside the band, or a weekly average below the previous week's. Run it without a trigger and you will find causes for fluctuations that have none.
What if two causes apply at once?
That is more common than it sounds — a price attack and an expired promotion in the same week, say. Then: fix the larger one first, let it take effect, and address the second afterwards. Tackling both at once is understandable and makes measuring the effect impossible.
When may I assume an attack?
Only when the tree yields no cause AND several anomalies fall into a narrow window that does not match your own changes. That order matters: the sabotage explanation is comfortable because it contains no fault of your own — and that is exactly why it sits at the bottom of the tree.
The free Listing Check scores any ASIN from 0 to 100 in a minute — the fastest way to see whether the mechanics from this lesson actually hold on your own listing.
Go deeper: the complete guide to optimizing Amazon listings →
Everything in this academy comes from day-to-day selling practice — the same playbook behind Listimo, the tool that turns product photos into complete Amazon listings.