The working sequence.
14 sectionsFollow the document’s sequence, or jump to the step you need.
- 01Phase 1: PROFILE - Extract Product-Level Data
- 02Phase 2: AGGREGATE - Combine for the True Picture
- 03Phase 3: CLASSIFY - Custom Label Segmentation
- 04Phase 4: EXECUTE - Per-Tier Optimization Tactics
Keep product evidence together before choosing an action.
P.A.C.E. separates extraction, aggregation, classification and the next account decision. Insufficient evidence is not a losing product.
- Profile
Retain product identity, campaign coverage and the date window.
- Aggregate
Combine matching product rows without double counting.
- Classify
Compare the covered observations with business economics and uncertainty.
- Execute
Review the proposed tier action, then read back the account state.
Read the full source: Phase 1: PROFILE - Extract Product-Level Data · Phase 2: AGGREGATE - Combine for the True Picture · Phase 3: CLASSIFY - Custom Label Segmentation · Phase 4: EXECUTE - Per-Tier Optimization Tactics
Jump to a section
Put this guide to work
Extract and aggregate product observations, classify for investigation, and plan a controlled per-tier correction.
Your task: Review a product cohort. Record the source, date range and reviewer beside the result. Leave anything you have not checked marked Not checked.
Related work: Find the next fix in a sample Shopping feed. · Plan feed and campaign segments with clear roles.
Review Product Evidence Across Campaigns
Campaign averages can hide variation between products and traffic sources. Export available product-level reports and reconcile compatible dates/metrics before classifying. Google provides product and asset-group reporting; it does not hide all product data, and asset groups do not have independent bid strategies. External cases without their original evidence are not used here to claim that consolidation harms performance.
The P.A.C.E. Framework
We developed this framework after auditing 20+ brands. The sequence matters - each phase builds on the previous one.
P = PROFILE → Extract per-product data across ALL campaigns
A = AGGREGATE → Combine data to see true performance
C = CLASSIFY → Label products by performance tier
E = EXECUTE → Apply optimization tactics per tier
Why this sequence? Gather compatible source rows before aggregation and classification. Retain missing-data and low-evidence states before briefing any change.
Skip a step and you're back to guessing. (We've watched people try. It doesn't work.)
Each phase has specific actions. Initial setup: 2-4 hours. Weekly maintenance: 30 minutes.
Phase 1: PROFILE - Extract Product-Level Data
Goal: Get performance data for every product, from every campaign.
What You Need to Extract
For each product (by item_id or SKU):
- Impressions
- Clicks
- Cost
- Conversions
- Revenue
- ROAS
- Campaign/Asset Group source
How to Extract
Option A: Google Ads Interface
- Go to Products & Listings → Products
- Segment by Campaign (important!)
- Choose dates suited to volume, seasonality and conversion lag; 30 or 60-90 days are planning examples
- Export to Google Sheets or CSV
Option B: Google Ads Scripts Use a script to automatically pull product performance across campaigns. Mike Rhodes' PMax scripts are a good starting point.
Option C: Third-Party Tools
- Producthero
- BlueWinston Labelizer
- DataFeedWatch
- Custom Looker Studio dashboards
The Critical Step Most Miss
Retain the campaign dimension in the export or query. A compatible segmented export can contain multiple campaigns; separate files are not inherently required.
Inspect product-level spend and attributed value by campaign as well as in aggregate. A difference between campaign ratios is a review question; do not infer a transfer opportunity without comparing traffic selection, lag and marginal economics.
Data Hygiene Checklist
Before moving to Phase 2:
- Observation window documented against volume, seasonality and conversion lag
- Review sample size, baseline variance and conversion lag; 100 clicks is not statistical significance
- All PMax and Shopping campaigns included
- Data exported by campaign source, not just aggregate
- Product IDs consistent across exports
Phase 2: AGGREGATE - Combine for the True Picture
Goal: See how each product actually performs when you combine all campaign data.
The Aggregation Process
Step 1: Create a Master Sheet
Columns needed:
| Column | Description |
|---|---|
| product_id | Unique SKU/item_id |
| product_title | Name for reference |
| total_impressions | Sum across all campaigns |
| total_clicks | Sum across all campaigns |
| total_cost | Sum across all campaigns |
| total_conversions | Sum across all campaigns |
| total_revenue | Sum across all campaigns |
| aggregate_roas | attributed conversion value / total_cost (undefined if cost is 0 or unknown) |
| campaign_count | How many campaigns product appears in |
| best_campaign_roas | Highest single-campaign ROAS |
| worst_campaign_roas | Lowest single-campaign ROAS |
Step 2: Calculate Aggregate Metrics
For each product:
Aggregate attributed ROAS = compatible attributed conversion value ÷ total cost across deduplicated report rows; it is not net profit. Keep zero/unknown denominators undefined.
CTR = Total Clicks ÷ Total Impressions (undefined if impressions are 0 or unknown)
CPA = Total Cost ÷ Total Conversions (undefined if conversions are 0 or unknown)
Step 3: Identify Fragmentation
Flag products where:
- campaign_count > 1 (shared product coverage; investigate before calling it fragmentation)
- High spend in low-ROAS campaigns
Review these alongside margin, lag and traffic mix; they are not ranked opportunities without that evidence.
Inspect the Observed Distribution
Calculate actual product shares by tier for this catalog and period. Low volume is insufficient evidence, not proof that a product cannot convert. A product appearing in multiple campaigns is a review flag, not established waste.
Aggregation Quality Check
Before moving to Phase 3:
- All products have compatible aggregates or an explicit unknown/undefined state
- Fragmentation flagged (campaign_count, ROAS variance)
- Duplicate report rows removed while legitimate same-item rows across campaigns/markets remain distinguishable
- Revenue and cost data validated against Google Ads totals
- Ready to classify by tier
Phase 3: CLASSIFY - Custom Label Segmentation
Goal: Label every product by performance tier using custom_labels in your feed.
The Four-Tier Classification System
Apply data sufficiency first: missing values, inadequate lag/sample or unmatched tier ranges stay Unclassified, not 0 or automatically Villain. Tier names and catalog percentages below are illustrative preferences; measure your own distribution and margins.
HEROES (Review for Scaling)
- Definition: ROAS more than 25% above your target (e.g., >5x if target is 4x)
- Typical %: 5-15% of catalog (your money printers)
- Action: Review marginal contribution and capacity before proposing a bounded budget or target change
- Goal: Assess additional economically useful demand; past ROAS does not guarantee it
SIDEKICKS (Optimize Carefully)
- Definition: ROAS 80%–125% of target in this example (3.2x–5x for a4x target), after sufficient observed data
- Typical %: 15-25% of catalog
- Action: Moderate budget, standard tROAS
- Goal: Improve efficiency, watch for promotion to Hero
VILLAINS (Reduce Spend)
- Definition: ROAS more than 20% below target (e.g., <3.2x if target is 4x)
- Typical %: 25-35% of catalog (yes, a third of your products might be villains)
- Action: Reduced budget, higher tROAS floor
- Goal: Break even or cut entirely
ZOMBIES (Test or Remove)
- Definition: <100 impressions OR <10 clicks in 60 days
- Typical %: 25-40% of catalog
- Action: Dedicated test campaign OR exclusion
- Goal: Establish eligibility, demand and data sufficiency before a keep/test/remove decision
Setting Your Thresholds
Step 1: Define Your Target ROAS
What ROAS do you need to be profitable after:
- COGS
- Shipping
- Returns
- Operating costs
Example: If you need 3.5x ROAS to break even, your target might be 4x.
Step 2: Set Tier Boundaries
| Tier | ROAS Range (4x target example) | Logic |
|---|---|---|
| Hero | >5x | More than 25% above target |
| Sidekick | 3.2x - 5x | 80%–125% of target in this illustrative scheme |
| Villain | <3.2x | More than 20% below target |
| Zombie | Illustratively <100 impressions, applied before ROAS bands | Low-exposure review cohort, not evidence of failure |
Adjust based on your margins and targets.
Implementing Custom Labels
In Google Merchant Center:
-
Create a supplemental feed with:
- id (match the primary Merchant Center product ID)
- custom_label_0: performance_tier (Hero/Sidekick/Villain/Zombie)
- custom_label_1: ROAS range (optional, for granular segmentation)
- custom_label_2: margin tier (optional, for profitability layer)
-
Upload as supplemental feed
-
Schedule refresh (daily or weekly)
Feed Structure Example:
id,custom_label_0,custom_label_1
SKU001,hero,roas_above_5x
SKU002,sidekick,roas_3_2x_to_5x
SKU003,villain,roas_below_3_2x
SKU004,zombie,low_impressions
Automation Options
Manual: Update spreadsheet weekly, upload to GMC Semi-Automated: Google Sheets + scheduled supplemental feed upload Fully Automated: Tools like Producthero, BlueWinston, or custom scripts
For accounts with 1000+ SKUs, automation is essential. Manual labeling doesn't scale.
Classification Quality Check
Before moving to Phase 4:
- Every product has an evidenced tier or explicit Unclassified/unknown state
- Thresholds based on YOUR profit requirements
- Custom labels implemented in GMC
- Supplemental feed connected and refreshing
- Labels visible in Google Ads campaign targeting
Phase 4: EXECUTE - Per-Tier Optimization Tactics
Goal: Apply different strategies to each product tier.
Campaign Structure by Tier
Illustrative Setup (use only where distinct roles, funding and evidence justify separate campaigns):
Campaign 1: Heroes (Scale)
├── Asset Group: Hero Products
├── Budget: 60% of total PMax budget
├── tROAS: Derived from contribution and test-risk limits; no automatic discount
└── Goal: Investigate additional economically useful demand
Campaign 2: Sidekicks (Optimize)
├── Asset Group: Sidekick Products
├── Budget: 30% of total PMax budget
├── tROAS: Documented economically justified campaign target
└── Goal: Review efficiency and sufficiently evidenced tier changes
Campaign 3: Villains + Zombies (Contain/Test)
├── Asset Group A: Villain Products
├── Asset Group B: Zombie Products (optional separate)
├── Budget: 10% of total PMax budget (combined)
├── tROAS: One campaign-level target; both asset groups share it
└── Goal: Break even or identify candidates for removal/graduation
Budget Allocation Strategy
The Illustrative 60/30/10 Allocation:
- 60% to Heroes - These products have proven conversion data. More budget can change marginal efficiency; no linear volume or ROAS result is guaranteed. Review marginal demand rather than treating past conversions as prepaid future sales.
- 30% to Sidekicks - They're close to target. With optimization headroom, some will graduate to Hero status.
- 10% to Villains/Zombies - An example test allocation, not a minimum required to evaluate the cohort; fund only a justified, bounded test.
Why these ratios? Based on the 80/20 pattern we see in audits - your top products drive disproportionate revenue.
Campaign controls can express commercial priorities that need explicit input. Smart Bidding also allocates bids; segmentation does not establish an improvement over its current allocation.
Adjust based on your catalog. If you have few Heroes, start with 50/35/15 until more products prove themselves.
Bidding Strategies by Tier
Heroes: Lower tROAS than target
- A 4x-to-3x target change is only a test hypothesis if the margin model and agreed loss limit support it
- You want VOLUME from winners
- A lower target can change bidding and volume; it does not guarantee conversions or profit
Sidekicks: Target tROAS exactly
- Derive the target from the documented margin/cost basis; revenue ROAS alone is not a profit requirement
- Standard optimization
Villains: Higher tROAS than target
- A 4x-to-5x/6x target change is illustrative and may suppress traffic; review demand and economics first
- A higher target can constrain delivery; it does not guarantee less wasted spend
- You're limiting exposure while still collecting data
Zombies: A supported PMax conversion/value-based strategy with explicit test budget; Maximize Clicks is not supported for PMax
- Goal is DATA, not efficiency
- Investigate eligibility and demand; impressions alone do not establish economic performance
Weekly Optimization Checklist
Every week, review and adjust:
Heroes:
- Getting sufficient budget (not limited)?
- Impression share acceptable?
- Any Heroes dropping to Sidekick?
- Would any proposed target change remain within the contribution and risk limits?
Sidekicks:
- Any ready for Hero promotion?
- Any dropping to Villain?
- Test different asset combinations?
Villains:
- Any improving? Re-evaluate for Sidekick?
- Any should be excluded entirely?
- Wasted spend decreasing?
Zombies:
- Any generating data now?
- Reclassify based on new data?
- Remove persistent zombies?
Monthly Relabeling Cycle
Once per month:
- Re-run aggregation (Phase 2)
- Recalculate tiers based on fresh data
- Update custom_labels
- Adjust campaign budgets if tier composition shifted
- Document changes for trend analysis
Products move between tiers. A Hero can become a Villain (seasonality, competition). A Zombie can become a Hero (trends, reviews).
Monthly relabeling keeps your structure aligned with reality.
Quick Reference Checklist
Initial Setup (2-4 hours)
- Export product data from all PMax/Shopping campaigns
- Aggregate by product ID in master sheet
- Calculate aggregate ROAS, identify fragmentation
- Define tier thresholds based on profit requirements
- Classify sufficiently evidenced products; retain Unclassified/unknown states
- Create supplemental feed with custom_labels
- Build campaign structure by tier
- Set tier-specific tROAS targets
- Document funded tier/test allocations summing to 100%; 60/30/10 is illustrative
Weekly Maintenance (30 min)
- Check tier performance in Google Ads
- Identify products near tier boundaries
- Adjust bids if needed
- Note products for reclassification
Monthly Review (1-2 hours)
- Re-run full aggregation
- Relabel products based on 30-day performance
- Update supplemental feed
- Adjust budget allocation if tier mix changed
- Document patterns and learnings
30-Day Action Plan
Week 1: Profile & Aggregate
Day 1-2: Extract product data from all campaigns Day 3-4: Build master sheet, aggregate by product Day 5: Identify fragmentation, calculate variance Day 6-7: Review findings, define tier thresholds
Week 2: Classify & Build
Day 8-9: Classify all products into 4 tiers Day 10-11: Create supplemental feed Day 12-13: Upload to GMC, verify labels showing Day 14: Build new campaign structure
Week 3: Execute & Monitor
Day 15: Launch tier-based campaigns Day 16-21: Monitor implementation, economics and risk; review timing depends on lag, with urgent defects handled immediately Day 21: First performance check
Week 4: Optimize & Refine
Day 22-28: Weekly optimization based on data Day 28-30: First monthly relabeling cycle Day 30: Document results, set ongoing schedule
What to Measure
After the first review, record product coverage, unknown/low-data items, actual tier spend and contribution margin. Recheck after a window suited to volume and lag. Classification does not establish profit, automatically reduce waste or guarantee margin improvement in 30/90 days. Numeric tier/volume/time/budget bands are planning heuristics; labels require sufficient evidence and an explicit unknown/unclassified state.
What We Got Wrong
The P.A.C.E. Method exists because we made expensive mistakes first. (The recurring theme of this business.)
Mistake: Over-segmenting too early. We used to create 8+ tier campaigns immediately. Bad idea.
Excess segmentation made evidence sparse in those accounts. The three-campaign example is one option; neither three campaigns nor 50 monthly conversions is a universal requirement. Retain or split groups from actual volume, distinct roles, lag and maintenance cost.
Mistake: Relabeling too frequently. Weekly relabeling felt proactive. Turns out it caused chaos - products bounced between tiers before the algorithm could optimize.
Monthly is a planning cadence, not a universal optimum. Choose a schedule from stock, volatility, lag and sufficient evidence; fix incorrect labels promptly.
Mistake: Ignoring margin data. Early versions only classified by ROAS. At the same revenue, a 5x ROAS product with a 10% pre-ad contribution margin leaves −10% of revenue after ads, while 3x ROAS at 40% leaves about 6.7%, before other omitted costs. Document that margin basis; ROAS alone cannot rank profitability.
Now we layer margin data into classification. Some "Villains" by ROAS are actually profit Heroes.
These failures shaped the method. Follow P.A.C.E. and you skip the learning tax we paid.
Next Steps
If You Want to Do This Yourself
- Block 4 hours this week for initial setup
- Follow the P.A.C.E. Framework step by step
- Commit to weekly 30-min maintenance
- Document your results
If You Want Expert Help
We help e-commerce brands implement per-product PMax optimization. Our typical engagement:
- Complete product-level audit
- Custom_label segmentation setup
- Campaign restructure
- Ongoing optimization support
Want us to audit your account? Reach out at tegra.co
Summary
Product reports can be combined across campaigns for a clearer descriptive view; Google provides reporting, and aggregate attribution is not causal truth.
The fix:
- Profile: Extract per-product data from all campaigns
- Aggregate: Combine to see true performance
- Classify: Label by tier using custom_labels
- Execute: Optimize each tier differently
Stop optimizing campaigns. Start optimizing products.
Want the Product Classification Template?
We built a Google Sheet that does the aggregation and classification automatically. Drop in your exports, it spits out your Heroes, Sidekicks, Villains, and Zombies.
Browse the resource library for the next relevant checklist or template.
Choose who handles the next step.
Use the reference yourself, book a focused paid working session, or have Tegra handle the ongoing channel work.
Search, Shopping, Performance Max and Demand Gen campaigns, plus the research, product feeds, static and video creative, and landing, offer and pre-sale pages behind them.
Explore managed Google Ads → · See paid working sessions →