Feed Integrity Blocks Performance Max Recovery at Northline Outdoor
Northline Outdoor sells patio furniture and grills: 4,200 active SKUs, average order value $340, and a $12,000/day Google Ads budget split between one Performance Max campaign and a legacy standard Shopping campaign. Six weeks after shifting 80% of spend to Performance Max, blended ROAS fell from 5.1 to 3.9 while conversion volume stayed flat. Merchant Center diagnostics show 1,730 products missing GTIN, 2,100 missing product type, 640 disapproved for images below 500x500, and 310 missing condition. Automatically generated headlines in the asset group include strings such as 'SKU-4482-B' and 'Save on 4482', pulled from truncated product titles. The team wants to raise budget 30% and add six more search themes to 'give the AI more room to learn'. Finance has frozen incremental spend until ROAS recovers, and the next quarterly review is in three weeks.
1. Rank the four feed problems by expected impact on Performance Max performance and justify the ranking, including what each problem actually breaks.
Read the worked answer
Ranking: GTIN, product type, disapprovals, condition.
- GTIN (1,730 SKUs, about 41% of the catalog). GTIN is the identity key that lets Google match an item to the same product across merchants and surfaces. Without it, matching falls back on title and brand - which is exactly why generated headlines read 'SKU-4482-B'. These products lose eligibility on product-listing surfaces and compete worse on broader queries, so the damage is systemic rather than per-item.
- Product type (2,100 SKUs). This attribute feeds category signals for asset groups and listing groups. Missing it blurs patio furniture against grills, so the system cannot learn category-level performance differences.
- Disapprovals (640). These remove roughly 15% of the catalog from serving. Painful, but the fix is mechanical and the blast radius is bounded to those SKUs.
- Condition (310). Lowest impact because most items are new and the default is usually correct.
Ordering by catalog share multiplied by severity puts GTIN first: it is the largest block and the one that degrades matching and generated copy simultaneously.
2. Finance asks whether a 30% budget increase plus six additional search themes will recover ROAS. What do you recommend, and how would you sequence the alternative?
Read the worked answer
Recommendation: do not increase budget now.
ROAS fell from 5.1 to 3.9 while conversions stayed flat. That pattern points to efficiency loss, not demand saturation. More budget on a feed where 41% of products lack GTIN and 15% are disapproved buys more impressions against weaker matching, so it scales the loss rather than recovering it.
Search themes are query hints, not a repair mechanism. Adding six broad themes to a campaign already struggling to match products pulls in more loosely relevant traffic - the opposite of what finance wants.
A defensible sequence:
- Fix GTIN and product type for the top 500 SKUs by revenue first; that cohort typically carries most of the revenue.
- Clear the 640 image disapprovals.
- Hold budget flat for two weeks and compare ROAS on the fixed cohort against unfixed SKUs as an internal reference group.
- Release the 30% increase only if the fixed cohort shows measurable lift, staged in 10% steps.
This gives finance an evidence gate instead of a bet, and it fits inside the three-week review window if the top-SKU work starts immediately.
3. Beyond the feed, what do you do about generated headlines that read 'SKU-4482-B', and why not simply pin better headlines?
Read the worked answer
The generated headlines are a symptom of the title field, not a separate creative problem. Generated assets draw from the feed and the landing page, so a title like 'SKU-4482-B Patio Set' gives the system nothing semantically useful to work with.
The fix is to rewrite titles into a consistent pattern: brand + product type + key attribute + size or color, for example 'Northline 4-Piece Teak Patio Set, Brown'. Do this for the revenue-weighted top of the catalog first, then batch the rest.
Pinning headlines is the wrong first move. Pinning forces a specific asset into every combination and reduces the number of combinations the system can test, which lowers the chance of finding a better performer. It also treats the output rather than the input. Pin only where brand or legal language is mandatory.
After titles are fixed, allow two to three weeks for asset regeneration and read the asset-level report before judging the creative. If certain headlines still surface incorrectly, use text exclusions rather than pins.