Statistical Waste Detection for Google Ads

01The Problem

Most PPC optimization tools compare your metrics to industry benchmarks and flag anything below average as "waste." This approach is fundamentally broken for one reason: a campaign that launched 3 days ago should have a terrible ROAS — it's still in Google's learning phase.

The result is a false-positive epidemic. Advertisers either panic-pause campaigns that were still gathering signal, or they ignore all alerts because the tools have cried wolf too often. Both behaviors cost money.

Google's own optimization recommendations have a well-documented conflict of interest — many of them (raise budget, enable broad match, expand audiences) directly increase your spend on Google's platform. Third-party tools that rely on benchmarks aren't much better.

ClickCatalyst takes a different approach: before classifying any waste, the system first asks whether the campaign has enough data to be judged at all.

02Minimum Viable Spend (MVS)

The core primitive is a per-campaign statistical threshold called Minimum Viable Spend. It's derived from the standard binomial sample-size formula:

n = z² × p × (1 - p) / e²

where:
  n = minimum clicks needed
  z = z-score for desired confidence (1.645 for 90%)
  p = campaign's observed conversion rate
  e = acceptable margin of error (absolute)
      we use relative error: e = 0.30 × p (±30% of true CVR)

Minimum conversions: C_min = n × p
MVS (dollars):       varies by bidding strategy ↓

The critical insight: MVS adapts to the bidding strategy. A Target CPA campaign and a Target ROAS campaign need completely different spend thresholds to reach statistical significance:

Target CPA:   MVS = C_min × target_CPA
Target ROAS:  MVS = C_min × (avg_revenue_per_conversion / target_ROAS)
CPC-based:    MVS = n × avg_CPC   (fallback for manual/maximize-clicks)

Any campaign below its MVS gets protected — the 6 standard waste categories are suppressed. We call this The Shield. No false alarms on campaigns still building signal.

03Worked Example

Bidding:    Target ROAS (target = 4.0x)
CVR:        3.2%
Avg CPC:    $4.50
Avg Rev/Conv: $85

Step 1 — Minimum clicks
  e = 0.30 × 0.032 = 0.0096
  n = (1.645² × 0.032 × 0.968) / 0.0096²
    = (2.706 × 0.03098) / 0.0000922
    ≈ 909 clicks
  C_min = 909 × 0.032 ≈ 30 conversions

Step 2 — MVS (dollars)
  Strategy = Target ROAS
  MVS = C_min × (avg_rev / target_ROAS)
      = 30 × ($85 / 4.0)
      = 30 × $21.25
      = $637.50

Step 3 — Classification
  Spent $400    → EXPLORATION (below MVS, protected by Shield)
  Spent $1,275  → 2× MVS, 0 conversions → BLACK HOLE ☠️
  Spent $1,913  → 3× MVS, volatile → CALIBRATION TRAP ⚠️

04Interactive Calculator

Plug in your own numbers. The calculator implements the same formula the production system runs in BigQuery daily.

05The 7-Category Waste Taxonomy

Once a campaign clears the MVS bar (or blows past it), waste gets classified into 7 categories. Categories 1–6 are standard waste types. Category 7 (Exploration Waste) is the MVS-derived type that most tools miss entirely.

06What This Finds in Real Accounts

The typical Google Ads account leaks 15–25% of total spend across these 7 categories. Here's what we found in one mid-size e-commerce account:

The 5 Shopping campaigns had been in Google's "learning" phase for months with nobody noticing — because standard dashboards show them as "active." The MVS model classified them as Black Holes (spend >2× MVS, zero conversions). Every dollar spent daily was waste.

07Tech Stack

Frontend:    Next.js (App Router)
Backend:     Python + BigQuery
Data:        Google Ads API + GA4 via OAuth (read-only)
Pipeline:    Daily scheduled BigQuery queries
             → 90-day rolling window, keyword-level
             → maturity classification → waste detection
Auth:        Google OAuth 2.0 (read-only scopes)
ML:          None. Statistical thresholds only.
             No black-box models — every threshold
             is derived from sample-size statistics.

The entire pipeline runs once daily as scheduled BigQuery queries — no per-request compute costs on the frontend. The UI serves pre-computed results from the daily pipeline. No ML models, no neural networks — just careful statistical thresholds tuned to how Google's auction system actually behaves.

08Try It

The free waste audit connects via OAuth (read-only — we never modify your campaigns), runs the full 7-category analysis on your actual Google Ads data, and delivers results in about 60 seconds.