One-liner: Stop wasting ad dollars on products that never sell. Automatically invest more in winners, cut losers fast.
| Fact | Impact |
|---|---|
| 95,000 products in catalog | Impossible to manually manage |
| 99.3% have zero sales in 600 days | No data to make smart decisions |
| Current system uses margin only | High-margin products with no demand get top-tier treatment |
High Margin + Zero Demand = Wasted Ad Spend
Example: A $2,000 chandelier with 35% margin gets promoted heavily.
It gets 300 clicks. Zero purchases. $XXX wasted.
We're paying for clicks on products nobody wants to buy.
| Waste Category | Annual Impact |
|---|---|
| Products with 200+ clicks, 0 conversions | $XX,XXX |
| Products promoted while out of stock | $XX,XXX |
| Below-target ROAS on non-converters | $XX,XXX |
Exact figures to be calculated from 4-year Google Ads data analysis (Phase 1)
Instead of margin alone, use multiple signals to rank products:
| Tier | What It Means | Google Ads Treatment |
|---|---|---|
| HERO | Proven winners (sales + margin + reliable vendor) | Maximum investment |
| MOMENTUM | Growing performers | High investment |
| POTENTIAL | High margin, not yet proven | Test budget |
| DISCOVERY | Catalog presence only | Minimal spend |
| EXCLUDED | Out of stock, bad data, margin too low | Zero spend |
| Feature | What It Does |
|---|---|
| Smart Kill Switches | Auto-demote based on product type: 14 days for cheap items, 30 days for expensive chandeliers |
| Vendor Scoring | Products from bad vendors (high returns, slow shipping) get penalized |
| Daily Updates | Tiers recalculate every night based on latest data |
| Borrowed Confidence | New products inherit scores from similar products that sell |
| Budget Spillover | If HERO products saturate (>90% impression share), budget flows to MOMENTUM |
| Seasonal Resurrection | "Zombie" products get a second chance when their season arrives |
| Fail-Safe Feed Updates | If nightly job fails, yesterday's feed stays active (no corrupted data) |
| Tier | Campaign Type | Why |
|---|---|---|
| HERO | Performance Max | Let Google optimize across all channels for proven winners |
| MOMENTUM | Performance Max | Grow promising products with algorithmic help |
| POTENTIAL | Standard Shopping | Manual CPC to force exposure and collect conversion data first |
| DISCOVERY | Standard Shopping (low priority) | Minimal spend, data collection only |
| EXCLUDED | None | No ads served |
Key Insight: POTENTIAL products don't go straight to PMax. They earn their way in via Standard Shopping first. Once they get 2+ conversions, they graduate to MOMENTUM PMax.
HERO (40%) ──────────────────────────────█████████████████████
MOMENTUM (30%) ───────────────────────────███████████████
POTENTIAL (15%) ───────────────────────────████████
DISCOVERY (10%) ────────────────────────────█████
Catch-All (5%) ──────────────────────────────███
Philosophy: Invest heavily in what works. Test cheaply. Cut losses fast.
| Metric | Current (v3) | Target (v5) | Improvement |
|---|---|---|---|
| HERO tier ROAS | ~2.5x | 4.0x+ | +60% |
| Wasted spend on 0-conversion products | $XX,XXX/yr | 50% reduction | Significant savings |
| Time to identify a loser | Never (manual) | 14 days (automatic) | Faster reaction |
| New product discovery | No path | 30-60 day graduation | New revenue |
| Phase | Duration | Deliverable |
|---|---|---|
| Phase 0: Data Validation | 2-3 weeks | Analyze 4 years of Google Ads data. Validate assumptions. Set thresholds. |
| Phase 1: Foundation | 2 weeks | Vendor scoring, kill switches, margin calculations |
| Phase 2: Core Engine | 2-3 weeks | Tier calculation engine, daily refresh, Google Ads integration |
| Phase 3: Dashboard | 2 weeks | Management UI, manual overrides, reporting |
| Phase 4: Launch | 1-2 weeks | A/B test → full rollout |
Total: 10-12 weeks
| Risk | How We Handle It |
|---|---|
| v5 performs worse than v3 | A/B test in shadow mode before switching |
| Kill switch too aggressive | Price-based windows: 14 days for cheap items, 30 days for expensive |
| Budget saturates HERO tier | Auto-spillover: excess flows to MOMENTUM when impression share > 90% |
| Seasonal products buried | Resurrection protocol: auto-promote when season arrives |
| New vendors unfairly penalized | Dynamic weights: reallocate vendor signal to style/price when no vendor data |
| Nightly job corrupts feed | Fail-safe: abort export if job fails, yesterday's feed stays active |
| Can't rollback | Keep v3 intact, one config change to revert |
Before setting the kill switch threshold, we'll analyze:
"If we had killed products after X clicks, how much profit would we have lost from products that converted on click X+1?"
Rule: If lost profit exceeds 10% of savings, the threshold is too aggressive—raise it.
Seven structural improvements address cold-start problems and over-aggressive automation:
| # | Refinement | Problem Solved |
|---|---|---|
| 1 | Category-dependent kill switches | Chandeliers need 30 days, not 14—longer decision cycles |
| 2 | POTENTIAL → Standard Shopping first | PMax can't optimize without conversion data; earn it first |
| 3 | Budget spillover logic | If HERO saturates (>90% IS), overflow to MOMENTUM |
| 4 | Dynamic similarity weights | New vendors don't penalize products; weight shifts to style/price |
| 5 | Zombie resurrection protocol | Seasonal products auto-promote when their season arrives |
| 6 | Fail-safe feed updates | If nightly job fails, yesterday's feed stays (no corruption) |
| 7 | False negative validation | Sprint 0 must prove kill threshold doesn't kill profitable products |
Before building anything, analyze historical data to answer:
- What click threshold catches non-converters without false positives?
- Which categories have seasonal patterns?
- Do current tiers correlate with actual Google Ads performance?
- How much are we actually wasting?
This phase validates the entire strategy with real numbers.
| Resource | Effort |
|---|---|
| Engineering | 8-10 weeks |
| PPC/Marketing | Campaign setup + ongoing optimization |
| Analytics | Threshold tuning, reporting |
- Pull 4 years of Google Ads performance data
- Cross-reference with sales, inventory, vendor data
- Produce validation report with specific recommendations
- If data validates the approach → proceed with build
- If data shows different patterns → adjust strategy
| Question | Answer |
|---|---|
| What's broken? | We promote high-margin products that never sell |
| What's the fix? | Use multiple signals (sales, vendor, inventory) not just margin |
| How does Google Ads change? | Winners get PMax, losers get cut, unknowns get tested cheaply |
| What's the upside? | 50%+ reduction in wasted spend, higher ROAS on top products |
| What's the risk? | Low—A/B test before switching, easy rollback |
| What's the timeline? | 10-12 weeks total, Phase 0 validates first |
- Approve Phase 0 - Data analysis to validate assumptions and set thresholds
- Review findings - Go/no-go decision based on actual numbers
- Build in sprints - Incremental delivery with checkpoints
- A/B test - Prove v5 beats v3 before full switch
- Launch - Cutover with monitoring
For detailed technical implementation, see: ppc-tier-system-v5-strategy.md