Why this matters
The advanced e-commerce level structured a catalogue by margin and role. That is a design, and a running programme drifts from it every week: products sell out, someone patches the feed with a rule, a freelancer excludes a range without telling the buyer, and the campaign's own report looks healthy throughout. The expert job is to judge the programme as it runs, fix it in the right order and move no budget until the evidence can carry the decision.
Two pages do that work here. The product feed and Performance Max audit scores the programme from the product feedA structured file of a store's products, prices and availability used to build shopping and catalogue ads. outwards, treats feed diagnostics and order values as gates, bands the result and ranks the fixes. The Performance Max uplift test decides whether Performance MaxA Google Ads campaign type that runs one goal across all Google inventory from shared assets. grows the account or only reshuffles credit, by comparing regions on total account value rather than on the campaign's own column.
The aquarium store and its catalogue campaigns
Thunorn Aquatics is an invented online retailer of aquariums, filters, fish food and live plants. Its live plants come from a grower whose availability list changes every week, which is where most of its feed trouble starts.
Thunorn Aquatics: the team, the catalogue and the audit evidence
Team: the owner, who holds the budget; a buyer who sets ranges, margins and stock; an ecommerce assistant who loads products into the store platform and uploads a stock file to Merchant Center every Friday; a freelance paid media manager; a part-time designer who shoots product photos; a bookkeeper.
| Range | Products | Contribution margin |
|---|---|---|
| Tanks and cabinets | 180 | 22% |
| Filters, heaters and lighting | 520 | 35% |
| Fish food and water treatments | 600 | 45% |
| Live plants, availability updated each Tuesday | 300 | 55% |
Blended contribution margin about 38%. Evidence window: the last six weeks, no sale inside it.
| Campaign | Spend | Reported conversion value | Set-up |
|---|---|---|---|
| Performance Max | £9,000 | £64,800 | One campaign, one asset group with the homepage as final URL, every product in one listing group, target ROAS 500% chosen from revenue, interest-based audience signals only, no brand exclusion, URL expansion at its default |
| Standard Shopping | £2,100 | £8,400 | Running since before Performance Max, one product group, no negative keywords |
| Search, brand | £1,200 | £15,600 | |
| Search, generic | £2,700 | £10,800 |
What the exports show:
- Merchant Center: 290 items disapproved or limited. Top reason, availability mismatch on live plants: between Friday uploads a feed rule sets every plant to "in stock". Second, missing identifiers on 34 filters, including the two best-selling canister filters. Nobody reviews diagnostics; the freelancer looks when spend drops.
- Titles are supplier codes, rewritten for the top 50 sellers only. Each plant species is one item, with the potted plant and the tissue-culture cup sharing it.
- No custom labels. 40 discontinued tanks are still advertised and out-of-stock plants still serve.
- Conversions: one purchase action is primary, but it fires again when the confirmation page is reloaded.
- Search term insights: brand queries carry 31% of Performance Max conversion value.
- Nobody compares platforms with orders. The order system recorded £74,500 of online orders in the window; the four campaigns reported £99,600 between them.
- Last month the freelancer excluded live plants from Performance Max because they returned less than 500%. The buyer found out from a customer.
- The owner's plan: double Performance Max from £1,500 to £3,000 a week, because its report shows a return of 7.2.
- Delivery: dry goods ship to all 16 of Thunorn's delivery regions; live plants and the large tanks only to 12. A spring aquascaping sale is planned for later in the season.
Score first, then test
The owner's plan skips two steps. A campaign built on a broken feed or miscounted values cannot be judged on its results: with best-sellers disapproved, every structure row describes what was left, and with a double-counting purchase action, bidding has chased the wrong figure. The uplift test freezes the feed for its window, because a feed change moves every shopping campaign at once. Test on today's feed and the result describes a programme about to change; fix the feed mid-test and it describes nothing.
So the order is fixed: score, fix the gates, rescore, test with the feed frozen, then decide the budget. The Shopping and Performance Max playbook supplies the weekly rhythm the fixes must survive and the split of product decisions that stops the drift returning.
Task
- Score every audit row as missing, partial or solid, with the evidence for each, gates first, and state the band and what it holds back.
- Write the findings as one fix list in the audit's priority order, with an owner, the guide behind each fix and a rescore date.
- Settle who decides about a product: name the merchandiser, feed owner, campaign manager, creative lead and analyst, and rule on the live plant exclusion, the discontinued tanks and the plant availability rule.
- Design the regional test that decides whether Performance Max earns more budget: arms, regions, primary metric, guardrails, window, what is frozen and the decision rule, written before launch.
- Say what happens to the owner's plan to double Performance Max under each result.
Worked answer
The scorecard and the band
Gates first. Feed diagnostics score missing: nobody owns them, best-selling filters are disapproved and the plant mismatch is patched downstream instead of fixed at the source. Order values score partial: only purchases are primary, but reloads add duplicates, so the target and structure rows are provisional. One missing gate makes the band red, which answers the owner's plan before any test does: the audit holds budget increases.
Thunorn Aquatics: the scorecard
| Row | Score | Evidence |
|---|---|---|
| Feed diagnostics routine (gate) | Missing | No owner; two best-selling filters disapproved; plant mismatches patched by a rule |
| Titles and attributes | Partial | Supplier codes; top 50 sellers rewritten |
| Price and availability agree | Missing | Plant availability mismatches flagged and left |
| Variants submitted correctly | Missing | One item stands for both the potted plant and the tissue-culture cup |
| Lifecycle and margin labels | Missing | No custom labels |
| New, seasonal and clearance rules | Missing | 40 discontinued tanks and out-of-stock plants advertised |
| Campaign type choice | Missing | Standard Shopping left running by default since before Performance Max |
| Shopping priorities and negatives | Missing | One product group, no negatives |
| Performance Max campaigns and listing groups | Missing | One listing group across margins from 22% to 55% |
| Asset group themes | Missing | One group, homepage as final URL |
| Signals and search themes | Partial | Generic interests only |
| Brand and URL controls | Missing | Defaults nobody chose; brand carries 31% of reported value |
| Order values on primary actions (gate) | Partial | Purchase only, duplicated on reload |
| Margin-aware targets | Missing | One 500% target from revenue |
| Orders reconciliation | Missing | Platforms report £99,600 against £74,500 of orders: a ratio of 1.34 nobody tracks |
| What Performance Max adds | Missing | Its own column only |
Band: red. Budget increases and new campaigns are held.
The fix list
The audit's order: gates, then the data that steers bidding, then structure and controls, then the evidence needed before any budget increase, with title polish last; within each group, the fix touching most revenue goes first.
Thunorn Aquatics: the fix list
| Order | Finding | Fix | Owner | Guide |
|---|---|---|---|---|
| 1 | Diagnostics unowned; plants mismatched; filters disapproved | Ecommerce assistant named feed owner; weekly diagnostics; plant stock synced from the store platform when it changes, and the "in stock" rule deleted; identifiers added at the source | Feed owner | Product feeds |
| 2 | Purchase duplicated on reload | Deduplicate the purchase action; reset the target from the cleaned figures | Freelancer, with the ecommerce assistant | Conversion action clean-up |
| 3 | No labels; discontinued and out-of-stock products advertised | A lifecycle label and a margin band label, populated from data and refreshed weekly | Buyer decides; feed owner populates | Product lifecycle |
| 4 | One target from revenue | Listing groups split where targets differ, each target from its band's break-even | Freelancer, with the buyer's margins | Performance Max structure |
| 5 | Defaults on brand and URL controls; one asset group | Brand exclusion and URL expansion decided and recorded; asset groups for tanks, filtration and plants with matching pages | Freelancer and designer | Building an asset group |
| 6 | Standard Shopping left running by default | Keep or retire it on purpose, and record why | Freelancer | Performance Max vs Standard Shopping |
| 7 | No reconciliation | Monthly reconciliation and a tracked ratio | Bookkeeper | Reconciling with sales records |
| 8 | No evidence of what Performance Max adds | The regional uplift test below | Bookkeeper proposes; owner approves; freelancer builds | The uplift test below |
| 9 | Supplier-code titles; pot and cup as one item | Rewrite titles and split variants, after the test window | Feed owner | Product variants |
Rescore: once six weeks have run on the corrected gates, so the rescore uses a window of the same length that sits wholly after the fixes, and at least a full month for bidding to adjust.
The last fix waits on purpose: retitling is exactly the feed change a test must not straddle.
Who decides about a product
The playbook's split, applied: the buyer is the merchandiser, the ecommerce assistant the feed owner, the freelancer the campaign manager, the designer the creative lead and the bookkeeper the analyst.
- The live plant exclusion. A campaign manager may exclude a product for poor returns against its margin, but plants were judged against one revenue target and against data spoilt by availability disapprovals. They return to the campaign under their own margin band, and any later exclusion is reported to the buyer, who decides whether a commercial reason keeps them.
- The discontinued tanks. The buyer marks them clearance or ended; the feed owner writes the stage into a custom labelFree-text product feed attributes a store fills in to group its products for campaign structure, bidding and reporting, such as margin tier or season., as the product lifecycle guide labels new, seasonal and clearance lines; the freelancer gives clearance its own group and excludes what is gone.
- The plant availability rule. Titles, attributes and rules belong to the feed owner, fixed at the source. From now on, the feed owner tells the freelancer before changing any label or attribute, because the listing groups are built on them.
The uplift test
Thunorn already runs Performance Max everywhere, so the first question is not whether doubling pays but whether the campaign adds anything. The test keeps it running beside Search and Shopping in treated regions and switches it off in control regions, the rest of the account unchanged. Doubling is a later question for the budget increase incrementality test.
Thunorn Aquatics: the test plan, written before launch
| Element | Choice |
|---|---|
| Regions | The 12 regions that receive the full range, matched into 6 pairs on 26 weeks of total online order value from the order system; one of each pair treated at random. The other 4 regions keep the current set-up and stay out of the read |
| Arms | Treated: Search, Shopping and Performance Max, with the brand exclusion Thunorn intends to keep. Control: Search and Shopping alone |
| Primary metric | Each region's total account conversion value, taken from the order system and set against the control prediction |
| Incremental return | Incremental value ÷ Performance Max spend in treated regions, against break-even 1 ÷ 0.38 ≈ 2.6 |
| Guardrails | Brand conversions claimed by Performance Max; new-customer share by region from order records; blended cost per conversion |
| Window | Matched pre-period, then 8 weeks of stable delivery after Performance Max is switched off in control regions, ending before the spring sale |
| Frozen | Titles, labels, rules, prices and promotions; the automatic stock sync runs in both arms alike; other campaigns' budgets and targets unchanged |
| Result | Decision |
|---|---|
| Incremental return at or above 2.6 | Keep Performance Max; test the doubling with the budget increase design |
| Value rises, return below 2.6 | Cut Performance Max back, or move its spend to Shopping and Search |
| Value flat while the campaign reports many conversions | Its reported conversions were credit taken from other campaigns; move the budget back to them |
| Value rises mostly from brand or returning customers | Tighten brand exclusion and audience signals, then retest |
The bookkeeper uses the incrementality testing guide to check whether the pairs can detect a lift that clears break-even; if not, the test grows or the decision is admitted to rest on judgement. Restricting the read to regions that receive the full range keeps delivery areas from biasing it. The decision rule is written into the plan before launch, and the read window is fixed in advance.
The owner's plan under each result
Doubling waits for gates solid at the rescore, a result clearing break-even and then a budget increase test. The owner sees the incremental returnRevenue caused by advertising, measured against a holdout, divided by the ad spend., never the campaign's own. If value is flat, the budget returns to the campaigns Performance Max overlapped and the report behind the plan is read for what it was: credit, not growth. If value is carried by brand or returning customers, the controls are tightened and the test rerun before anything is scaled.
Common mistakes
- Testing before scoring. A test run on a feed with disapproved best-sellers measures a programme that is about to change.
- Patching availability with a rule. A rule that marks every plant in stock creates the mismatch it hides.
- Letting the freelancer merchandise by default. An exclusion for poor returns is reported to the buyer and judged against the product's own margin.
- Testing the doubling first. Whether the campaign adds anything comes before how much more it should get.
- Changing the feed in the test window. Every shopping campaign moves at once, and unevenly where delivery areas differ.
- Reading the campaign's own return. Only total account value against control regions shows growth.
The shopping programme checkpoint that follows tests the same judgements on a different store.
