UPLIFY19 checks

Demonstration report

A PMax audit that tests where average ROAS came from

A strong average does not prove growth. We separate brand, remarketing, products, feed quality and new customers.

Evidence status: Demonstration report
Source: Synthetic, internally consistent data
Period: Synthetic 30 day window

Names and metrics are synthetic. The arithmetic demonstrates the format, not a client result.

Control snapshot

240,000 spend synthetic monetary units
816,000 Ads revenue synthetic model
3.40 average ROAS 816,000 / 240,000
480 orders customer status partly unknown

The 90 second view

Protect Keep the product group with confirmed orders and bidding stable during measurement review.
Separate Isolate brand and verify new customer share.
Test Use a capped high margin inventory test with a stop rule and rollback.

What was checked and where confidence ends

Sources

Google Ads, PMax and change history
Merchant feed and product diagnostics
GA4 and synthetic backend
Search categories, assets and landings

Coverage

All campaigns, asset groups, product groups and conversion actions in the synthetic account.

Limitations

Long term margin and true new customer status usually require backend or CRM data.

Measurement confidence

Arithmetic verified, business truth partial

Spend of 240,000 and attributed revenue of 816,000 produce ROAS 816,000 / 240,000 = 3.40, reconciled to 480 synthetic orders.

19 checks

Full audit map

The three decisions above give an executive summary. The complete review register below shows the checks behind those decisions.

This page shows 19 of 19 checks in the demonstration register. A client report also links each check to its source, accountable role, and working artifact.

Check	Priority	Decision	What we established
01 Measurement and ROAS limitsAverage ROAS mixes demand sources and is not an instruction to scale.
PM-01	P0	Protect	Keep the working purchase conversion.	Before bid changes, one purchase must pass with ID, value, and currency without duplication.
PM-02	P1	Separate	Classify new, returning, and unknown.	Without customer status, average ROAS does not show the cost of new demand.
PM-03	P1	Verify	Reconcile Ads revenue with the internal ledger.	Platform reporting can differ by window, returns, and attribution.
02 Brand, remarketing, and incrementalityBrand and repeat demand account for 40% of synthetic revenue, which requires a controlled separation test.
PM-04	P1	Experiment	Test brand exclusion.	326,400 / 816,000 = 40%; part of this revenue may occur without PMax.
PM-05	P1	Protect	Maintain brand coverage during the test.	Exclusion without a separate control risks losing inexpensive demand.
PM-06	P2	Measure	Compare incremental revenue with redistribution.	Moving conversions between campaigns does not equal business growth.
03 Feed, products, and SKU economicsThe feed supplies products but does not fully support decisions by margin and stock.
PM-07	P1	Fix	Add reliable margin and stock labels to every SKU.	300 of 1,200 products, or 25%, lack reliable segmentation.
PM-08	P1	Reconcile	Check 20 random SKUs against the source system.	Field coverage does not prove field accuracy.
PM-09	P2	Isolate	Place products with spend and no orders on a watchlist.	The decision follows an adequate window and stock verification.
04 Asset groups, signals, and adsCreative completeness is assessed with group purpose rather than one interface score.
PM-10	P1	Complete	Fill or close incomplete asset groups.	A group without enough copy and media cannot represent product intent.
PM-11	P2	Add	Introduce category search themes and audience signals.	Signals should fit the product group rather than repeat broad terms.
PM-12	P2	Extend	Add verified callout, price, and promotion assets.	Every promise must be visible and true on the landing page.
05 Campaign structure and budgetThe high-margin group is tested separately without rebuilding the account.
PM-13	P1	Experiment	Separate 180 high-margin products.	The group has orders and stock but is mixed with the rest of the inventory.
PM-14	P1	Prevent overlap	Use mutually exclusive product IDs.	Overlap obscures the result and creates internal competition.
PM-15	P2	Hold budget	Do not change budget with structure.	A single variable is required for a causal conclusion.
06 Change control and protected winnersThe report defines changes, protected elements, and rollback conditions.
PM-16	P0	Protect	Do not disable a campaign with confirmed orders.	The issue is average interpretation, not the absence of results.
PM-17	P1	Document	Record baseline and date for every change.	Without a baseline, the team cannot separate impact from seasonality.
PM-18	P2	Limit	Set an experiment budget cap.	The test needs a controlled maximum risk.
PM-19	P2	Roll back	Return inventory when a guardrail is breached.	Rollback is defined before launch, not after an unwanted result.

Priority decisions with evidence

Verify

Average ROAS mixes brand and acquisition

Source: Ads, search categories, backend model. Period: 30 days. Status: synthetic. Confidence: medium. Coverage: 240,000 spend, 816,000 revenue, 480 orders.

Observation: 326,400 of 816,000 revenue comes from brand and returning demand, a 40% share. Some customer status is unknown.
Decision logic: ROAS 3.40 cannot be scaled mechanically, but the campaign still produces orders and should not be stopped wholesale.
Action and verification: Analyst adds a backend flag and brand exclusion test. Acceptance: new, returning and unknown revenue, control, cap and margin rollback.

Fix

The feed hides product economics

Source: Merchant feed and product report. Period: 30 days. Status: synthetic. Confidence: high. Coverage: 1,200 products.

Observation: 300 of 1,200 products, or 25%, lack reliable margin and stock labels.
Decision logic: Delivery works, but the manager cannot see contribution margin.
Action and verification: The feed specialist adds labels with sample QA. Acceptance: 1,200 / 1,200 active products labeled and 20 random SKUs match backend.

Experiment

The high margin group needs a test, not an account migration

Source: product performance and margin model. Period: 30 days. Status: synthetic. Confidence: medium. Coverage: 180 products.

Observation: The group has orders and stock but is mixed with the remaining inventory.
Decision logic: Standard Shopping may add control and cannibalization risk, so one variable changes.
Action and verification: Google lead creates an inventory split. Acceptance: no ID overlap, incremental revenue, margin stop rule and rollback.

What works and should not change

Keep bidding stable during data labeling

Order volume is sufficient; interpretation, not delivery, is the main risk.

Close new customer and margin gaps before a separate bidding test.

Target state and change rules

Target state

ROAS is split by demand type and SKU economics; strong campaigns are protected and experiments are isolated.

How we test

Brand exclusion and the high-margin group launch separately with stable budget and non-overlapping products.

When we roll back

The test stops if incremental revenue is not confirmed or guardrail CPA / ROAS breaches the agreed threshold.

The 7, 30 and 90 day plan

7 days

Freeze baseline and map data gaps.
Owner
Analyst and feed specialist

Acceptance criteria
The report separates brand and non-brand demand, customer status and margin, with unknown values explicitly labeled.
30 days

Run brand exclusion and inventory split tests.
Owner
Google Ads lead

Acceptance criteria
Test and control inventory do not overlap, spend remains within the cap, and breaching the numeric stop threshold triggers rollback.
90 days

Scale verified groups only.
Owner
Growth lead

Acceptance criteria
Contribution margin and new customers verified.

What the client receives

ROAS decomposition
PMax, asset and URL expansion review
Feed economics register
Experiments with rollback
New customer measurement tasks

Method and appendices

ROAS split by new / returning / unknown
Before → after campaign map
Supplemental feed and custom-label rules
Budget, metrics, window, and rollback for each test

What is PMax actually scaling?

We separate the platform average from brand, product economics and verified new customers.

Discuss an audit for your business