VERIXA / Evidence
What a VERIXA finding looks like.
VERIXA is new. There are no client results to publish here and we will not borrow anyone else's. What we can show you is the exact form the work takes, run end to end on questions we constructed ourselves.
Status is stated on everything / Nothing on this page is a client result
How to read this page
Three categories, and only one of them is currently populated.
Most firms show you their best case study. We would rather show you the shape of the work and be straight about which parts of the shelf are still empty.
01Real case studies
Published only when a genuine client agrees in writing to have their work described. None exist today, so this category is empty rather than filled with something that resembles one.
02Sample intelligence
Demonstrations built by VERIXA on constructed questions, using realistic mechanics. Every figure is illustrative. Each one is labelled SAMPLE where you read it, not in a footnote.
03Anonymised analysis
Real work with identifying detail removed, published only with the client's permission. None exist today, and an anonymised piece will be labelled as such when one does.
Sample intelligence
Five investigations, each run through all five stages.
These are demonstrations. The mechanics are real and the reasoning is the reasoning we would use. The numbers are invented to make the shape legible, and they are not results.
“We are wasting money on ads but nobody can tell me where.”
An operator reports advertising spend up year on year while net profit is flat. The internal view is that costs per click have risen.
Pull the search term report across a 60 day window, campaign settings for every active campaign, and placement performance. Record the state and the date before interpreting any of it.
41 of 118 active search terms carried spend and returned zero attributed orders. 34 of those sit inside one broad match campaign with no negative keyword list attached. Search term report and campaign settings screenshot, both dated and attached.
The cost per click rise is real but secondary. The larger effect is structural: one campaign has no negative coverage and is buying terms that have never converted.
Add the 34 terms as campaign negatives. One named owner, reversible within a day. Expected effect stated in advance: spend against those terms stops inside 48 hours, total campaign orders unchanged.
Checked at day 14 against the measure named before the action. Spend against the 34 terms reached zero. Total campaign orders unchanged, which is what was predicted, so the spend was not buying incremental sales.
“Sales fell on two products and the reports show nothing.”
Two parent listings show a sales drop with no corresponding traffic drop. Advertising continued running at full budget throughout.
Pull offer history for both listings across the month, advertising spend by ASIN for the same window, and the price change log.
Both listings held the featured offer continuously until a price change on the 14th, after which neither has held it. Advertising against both continued at full budget for the eleven days that followed. Offer history export and spend by ASIN, both dated and attached.
The featured offer was lost, which stops Sponsored campaigns from serving on that listing. The budget continued to be committed while the offer could not convert.
Restore the offer condition that preceded the change, then re-check eligibility before resuming spend. Named owner, expected effect stated in advance.
Re-checked after the change. Whether the featured offer returns is reported either way, including the case where it does not and the cause turns out to be a competing offer rather than the price change.
“One variation stopped converting and we cannot work out why.”
A single child variation shows a conversion drop that has not recovered. The other children in the same parent are unaffected.
Pull detail page sales and traffic by child across 90 days, the full listing change history for the same window, review history, and stock events.
No listing edit, price change, review change, image change or stock event was found in the same window as the conversion drop. Traffic by child and listing change history, both dated and attached.
Unexplained. The available evidence does not identify a cause, and we are not going to invent one to make the report feel complete.
No action on this finding. Acting now would change several variables at once and destroy the ability to identify the cause later. Recorded as open with the reason.
Kept under observation with a scheduled re-check, and a named list of what additional evidence would settle it if the client can reach it.
“Is this category worth entering?”
An operator is considering committing inventory to a category on the basis that the top listings have modest review counts.
Record the offer composition on the leading listings, price and offer movement across a 90 day window, review velocity rather than review totals, and keyword coverage.
Review totals are modest, but review velocity on the top three listings is rising, and their price has moved in a coordinated way three times in the window. Dated observation set attached, with the date of every competitor claim recorded.
The low review count reads as an opening but the velocity and pricing behaviour indicate actively managed listings. Stated with medium confidence, because seller intent is not directly observable.
Advise against a full inventory commitment on this evidence alone. Name the smaller test that would produce the missing information at lower risk.
Re-observe at a set interval, because a category read goes stale. The assessment carries its own expiry date rather than being treated as a standing fact.
“Revenue is up and profit is down.”
Top line growth over two quarters with net profit moving the other way. No single cost line looks obviously wrong.
Reconstruct unit economics per ASIN from the operator's own landed cost, then compare against fee records, returns, and advertising attribution for the same units.
Growth is concentrated in three ASINs whose true landed cost had not been updated after a supplier change. Two of the three are below contribution once fees, returns and ad spend are attributed. Fee records, returns data and the operator's own cost sheet, all dated.
The account is scaling its least profitable units. Nothing in the standard reporting shows this, because the reporting uses a cost figure that is out of date.
Update landed cost, then re-run bid decisions against the corrected figure. Named owner, expected effect stated before the change.
Re-check contribution per unit after the correction, and report whether the corrected figures changed the ranking of which products deserve budget.
One sentence, stated plainly, with no hedging and no supporting detail. It is written last, after the evidence is settled, and it is the only part of the report that is allowed to sound certain.
Why it is separated: if the finding cannot be said in one sentence, the analysis is not finished. Compressing it is a test we apply to ourselves before the report leaves.
The raw observation and where it came from, with a date. Nothing here is interpreted. You can open the attached export and count the rows yourself.
Why it is separated: this is the part that makes the rest checkable. A report where evidence and interpretation are written as one paragraph cannot be audited, because there is no way to agree with the observation while disputing the conclusion.
The pattern inside the evidence. Still factual, still checkable, but it is the observation organised rather than the observation raw.
Why it is separated: the difference between a scattered problem and a concentrated one changes the recommendation completely. Forty one loose terms is a hygiene task. Thirty four in one campaign is a structural fault with a single fix.
What we believe the evidence means, stated as a belief and carrying its own confidence level. This is the first part of the report that is an opinion, and it is labelled as one.
Why it is separated: so you can accept the evidence and reject our reading of it. Where confidence is medium or low we say so, and where we cannot explain something we write unexplained rather than reaching for a plausible story.
The argument against our own recommendation, written by us. Every action has a way of going wrong and this is where it is named before you decide.
Why it is separated: a recommendation with no stated downside is a sales pitch. If we cannot describe how our own advice could hurt you, we have not thought about it hard enough to give it.
The action, with one owner, one expected effect stated before the work happens, and a way back. The expected effect is written in advance precisely so it can be checked afterwards.
Why it is separated: naming the expected result before acting is what stops the measure being quietly changed later to make the outcome look better.
The return visit. It compares what happened against what was predicted, using the measure named in the recommendation and no other.
Why it is separated: this is the part most reports do not have. When the expected effect does not occur, the verification note says so in the same plain language, and the finding goes back to interpretation rather than being quietly dropped.
Real case studies
This section is intentionally empty.
No client case studies yet
When a client agrees in writing to have their work published, it will appear here with their name, their numbers, and the same five stage structure as the demonstrations above. Until then this space stays empty, because a case study is the easiest thing on a website to fabricate and the hardest for a buyer to check.