Structure exists to create different decisions
A Google Shopping campaign deserves to be separate only when you will control it differently. A different budget, efficiency target, country, campaign type, promotion window or operational owner is a reason. “We have many categories” is not, by itself.
Too little structure lets strong products absorb the evidence for weak ones. Too much structure leaves campaigns without enough conversion data, duplicates settings and makes budget movement arbitrary. The right structure is the smallest number of decision boundaries the business can genuinely operate.
Prepare the product map
Export active item IDs and add category, brand, price, margin band, stock cover, launch date and strategic status. Then join 60–90 days of product-level cost, conversions and conversion value. The current Google Ads Products report can provide a cross-campaign retail view.
Mark each product with the treatment it needs:
- standard target and shared budget;
- different efficiency target because economics differ;
- protected budget because it is strategically important;
- learning allocation for a new product;
- reduced exposure because stock or margin is constrained;
- excluded until an eligibility, offer or page problem is fixed.
If most products receive the first treatment, keep them together. Do not manufacture complexity to make the account look managed.
Choose Standard Shopping, Performance Max or both
| Need | Structure implication |
|---|---|
| Product-group bids and more constrained Shopping inventory | Standard Shopping may be useful |
| Cross-channel reach with product data and creative | Performance Max may be useful |
| Clean test of a distinct product cohort | Make inventory mutually exclusive |
| Different targets by margin or market | Separate only the affected cohort |
Do not describe a campaign-type comparison as an experiment when both campaigns can serve the same item IDs. Create mutually exclusive inventory partitions and hold geography, offer and conversion goal stable. Performance Max has additional channels, so compare total commercial outcome and use Google’s channel performance reporting to understand delivery; do not pretend the test isolates Shopping inventory alone.
Build product groups and listing groups
Standard Shopping groups inventory inside ad groups using attributes such as category, product type, brand, item ID, condition, channel and custom labels. Google’s product-group documentation is the implementation reference. Performance Max uses listing groups to determine which Merchant Center products are included.
Use stable feed fields. Product type can reflect the retailer’s own taxonomy; custom labels can hold commercial classifications. Keep one meaning per custom-label column and document the values. If margin is stored in custom_label_0, do not also put season values in that column.
Is campaign structure hiding product performance?
Upscale maps product economics to campaign controls and removes structure that has no decision behind it.
Book My Free Ad AuditWorked example: a 1,200-product catalogue
Consider a hypothetical homeware retailer with 1,200 active variants. Bedding has 55% gross margin and dependable stock. Furniture has 28% margin, high order value and expensive returns. A 40-SKU summer collection has only six weeks of useful demand. The remaining long tail has mixed data.
A defensible first structure might be:
| Campaign | Inventory | Reason for separation |
|---|---|---|
| Core bedding | High-stock bedding | Higher allowable acquisition cost and stable supply |
| Furniture | Furniture only | Different margin, return cost and target |
| Summer launch | 40 launch SKUs | Protected budget and fixed end date |
| Long tail | Remaining eligible products | Discovery within a capped budget |
That does not mean four campaigns are permanently optimal. After enough data, the long tail may be merged, split or excluded. The key is that every separation has a budget or target consequence.
When to split
Split when a cohort is large enough to evaluate and needs a materially different action. Useful evidence includes sustained budget competition, a margin difference that changes break-even return, a promotion with a hard deadline, stock constraints, or a category whose search and conversion behaviour is being hidden by a larger group.
Do not split on a single week of volatility. Set a minimum evidence threshold appropriate to the business, such as at least 30 conversions or spend equal to several target acquisition costs across the proposed cohort. The threshold is a decision aid, not a universal Google requirement.
When to merge
Merge structures that share the same target and budget logic, repeatedly underspend, or generate too little data for separate decisions. Before merging, preserve reporting through product type or custom labels. A simpler campaign does not require less granular analysis.
A warning sign is a weekly management meeting spent moving small budgets among ten near-identical campaigns. Another is campaigns named by category but governed by exactly the same target, creative and priority. In both cases, the structure is describing the catalogue rather than controlling it.
Migration without losing the baseline
- Export settings, product membership and 60–90 days of item performance.
- Fix the feed labels and confirm they populate all intended products.
- Build mutually exclusive groups and check “Everything else” exclusions.
- Set budgets from expected demand, not equal shares.
- Launch with a change log and avoid simultaneous title, price and landing-page changes.
- Review product coverage, spend distribution and conversion lag before judging outcome.
Measure the migration with eligible product count, products receiving impressions, spend concentration, conversion volume, revenue and contribution by the original cohorts. Campaign-level ROAS alone cannot tell you whether the new structure improved allocation.
Structure failure cases
Campaigns by category with identical treatment: five campaigns each use the same target, budget logic and market. The split describes the website navigation but creates no control. Preserve category reporting with product type and consider consolidation.
Margin segmentation with stale data: products enter a high-margin campaign based on an annual spreadsheet while discounting and fulfilment costs change. Refresh the label from a governed source or remove the claim that the campaign is profit-led.
New-product campaign without an exit rule: products remain in “New” for a year, fragmenting data. Define entry date, minimum learning allocation and a move-to-core or exclude decision.
Overlapping inventory presented as a test: Standard Shopping and Performance Max can serve the same IDs. Make product cohorts mutually exclusive and document the limitations caused by different channel inventory.
Monthly structure review
Review product membership, spend concentration, products excluded from every active group, cohorts constrained by budget and campaigns below the conversion volume needed for their separate decisions. Compare the current catalogue with the feed labels that drive inclusion.
Set a merge or split proposal only after estimating the result. If 80 products proposed for separation generated four conversions in 90 days and need the same target as the parent, a new campaign is unlikely to create useful control. If a margin cohort generated 300 conversions and its break-even ROAS differs by 150 percentage points, separation has a clearer commercial case.
Use the custom-label governance guide before basing structure on feed classifications.
Preserve an “unclassified” route for products with missing labels and monitor its size. Silently excluding unknown inventory can remove new launches; silently including it in a priority campaign can defeat the commercial boundary. Decide which behaviour is safer and alert when unknown coverage exceeds the agreed tolerance.
Annotate every inventory-rule change and compare intended product counts with processed listing-group counts before budgets are allowed to move.