The working sequence.
9 sectionsFollow the document’s sequence, or jump to the step you need.
- 01Part 1: Test Readiness Assessment
- 02Part 2: Choose Your Test Type
- 03Part 3: Contamination Prevention
- 04Part 4: Results Calculation
- 05Part 5: Decision Rules (Fill In Before Testing)
View 2 more sections
Jump to a section
Put this guide to work
Choose a test, define comparison/lag/contamination and preregister decisions before interpreting results.
Your task: Draft the test decision rule. Record the source, date range and reviewer beside the result. Leave anything you have not checked marked Not checked.
How to interpret it: A before/after comparison alone cannot isolate incrementality. Define the holdout, sample/power, conversion lag, contamination risks and uncertainty before starting. Any calibration factors below are planning assumptions, not measured causal ROAS.
Related work: Measurement Maturity Assessment · Choose an attribution view for the decision you need to make.
Part 1: Test Readiness Assessment
Before you design a test, make sure you have the prerequisites.
Minimum Requirements
- Document test cost and decision value; a$10K spend filter is a planning heuristic, not universal viability
- Run a power/sample calculation using baseline variance and the smallest decision-relevant effect; 100 weekly conversions is not a universal sufficiency threshold
- Clean conversion tracking verified in the last 30 days (broken tracking invalidates everything)
- No major campaign restructuring planned during the test window
- Stakeholder agreement on what you'll do with the results (pre-commit to decision rules)
Data You Need Before Starting
- 90 days of performance data by geographic region (DMA or state)
- Conversion lag distribution: what percentage of conversions arrive at 1, 3, 7, 14, and 30 days
- Current brand search spend as a percentage of total spend
- Organic brand traffic baseline (from Google Search Console)
- Your breakeven ROAS threshold
Part 2: Choose Your Test Type
Brand Search Before/After Diagnostic (Exploratory)
Use this to observe substitution hypotheses. A pause compared with preceding days cannot isolate incrementality or answer the causal question in a guaranteed 7–10 days.
- Verify organic rank is #1 for your top 10 brand terms
- Screenshot current Auction Insights for brand campaigns
- Record 30-day baseline: brand search clicks, organic brand traffic, total site traffic, total revenue
- Set up daily monitoring: organic brand clicks (Search Console), total site sessions, total revenue
- Pause all brand search campaigns
- Monitor for 7 days during a "normal" week (no sales, promotions, or major email campaigns)
- Calculate traffic delta only with a known nonzero baseline: (Pre-test traffic - Test period traffic) / Pre-test traffic; otherwise mark undefined
- Calculate organic recovery only with a known nonzero decrease in paid brand clicks; otherwise mark undefined
- Record paid/organic changes separately. Do not convert an organic-click recovery ratio into incremental conversions or CPA; clicks and conversions have different denominators.
Interpretation: Report this as exploratory before/after evidence. Use a suitable controlled design for a causal allocation decision.
Geo-Split Test (Requires Design and Power Review)
Consider when the proposed design has adequate power for the decision-relevant effect. A 250-conversion weekly count alone does not establish high confidence.
- Choose geographic units (DMAs recommended for most tests)
- Pull 90 days of performance data by geo unit
- Match treatment and control markets on:
- Historical conversion rate (similar rates over past 90 days)
- Revenue per visitor / AOV
- Population and traffic volume
- Demographics (age, income, urban/rural mix)
- Historical revenue trend direction
- Seasonal patterns
- Size the holdout using power, variance, cost and spillover analysis; 15–20% is an illustrative allocation
- Verify no geographically adjacent DMAs are split between groups
- Define which campaign/geographic controls the estimand requires; keep other treatments comparable and review cross-market exposure
- Run 1-2 week baseline with both groups under identical conditions
- Evaluate pre-specified baseline balance, trends and variance; a 5% difference alone does not validate exchangeability
- Execute campaign changes in holdout markets only
- Freeze all other campaign changes during test period
- Monitor daily: conversions, revenue, organic traffic by test geography
- Run for minimum duration based on conversion lag (typically 3-4 weeks)
- Add observation window after test ends (equal to your longest conversion lag)
Part 3: Contamination Prevention
Every item here can invalidate your test results.
- No adjacent DMAs split between treatment and control groups
- Required campaign/geographic controls and spillover checks follow the pre-specified estimand; no blanket all-campaign change is assumed
- No geo-targeted promotions running in only treatment or only control markets
- Document national promotions and assess differential geographic effects; nationwide availability does not establish equal impact
- No new campaigns launched during the test window
- No bid or budget changes during the test window
- No major creative refreshes during the test window
- Competitor activity monitored for unusual changes
Part 4: Results Calculation
Core Metrics
Incremental Revenue:
Treatment market revenue: $_______ Control market revenue (raw): $_______ Control market population share: % Expected control revenue at treatment scale: Control revenue x (Treatment share / Control share) = $ Incremental revenue: Treatment revenue - Expected control revenue = $_______
Incremental ROAS (iROAS):
iROAS = Estimated incremental revenue / Incremental treatment ad spend relative to control = _______x
The simple population scaling above assumes comparable per-capita baseline outcomes. Validate or replace it with a pre-specified baseline-adjusted estimator; report confidence intervals, balance and spillover checks. A zero/unknown spend difference makes the ratio undefined, not 0.
Platform ROAS (for comparison):
Platform ROAS = Platform-Reported Revenue / Ad Spend = _______x
Incrementality Factor:
Factor = iROAS / Platform ROAS = _______
Apply a calibration only within the tested channel, population, spend range and period, with uncertainty; it is not a permanent account-wide factor.
What Your Factor Means
Review the factor and its interval alongside contribution margin, experimental validity and the precommitted decision rule. No fixed factor band alone establishes profitable scaling or a need to pause.
Part 5: Decision Rules (Fill In Before Testing)
Pre-committing to decisions prevents reinterpretation of uncomfortable results.
If lift is below _____%, we will: ________________________________
If lift is between _____% and _____%, we will: ________________________________
If lift is above _____%, we will: ________________________________
If results are inconclusive, we will: ________________________________
Part 6: The 0.7x Quick Reference
The following multipliers are illustrative sensitivity scenarios. They are not derived from this account and cannot substitute for a controlled test or establish causal ROAS:
| Account Type | Multiplier | Example |
|---|---|---|
| Heavy brand search (>15% of spend) | 0.50x | 6.0x ROAS becomes 3.0x |
| Mixed campaigns (standard e-commerce) | 0.70x | 5.0x ROAS becomes 3.5x |
| Primarily non-brand Shopping | 0.85x | 4.0x ROAS becomes 3.4x |
| Non-brand Shopping only | 0.90x | 3.5x ROAS becomes 3.15x |
Replace these assumptions with appropriately designed measurements when available. Report uncertainty and the observation window; do not present a scenario multiplier as measured lift.
Part 7: Testing Calendar
| Quarter | Test | Why This Timing |
|---|---|---|
| Q1 | Brand search incrementality | Example slot; choose actual seasonality, cost and decision value. |
| Q2 | PMax incrementality (3-cell) | With brand / without brand / holdout. |
| Q3 | Channel comparison (Shopping vs Search) | Results inform Q4 budgets. |
| Q4 | Review peak-period risks | Test or monitor according to power and business risk; do not reuse factors outside their validated scope. |
Refresh triggers (run a new test when any of these occur):
- Campaign structure changed significantly
- New major competitor entered or exited
- 90+ days since last measurement
- Business model changed (new products, pricing, markets)
Next Step
For the complete team-operated workflow, review the Google Ads AI Agentic System. The product page lists the current deliverables, price, license terms and setup responsibilities.
- Complete geo-split test design with market matching methodology
- Brand search decision framework with published case study data
- PMax isolation protocols (3-cell methodology)
- YouTube/Demand Gen geo holdout protocol for measuring halo effects
- Statistical tools for reading results without a data science degree
- Quarterly testing calendar with refresh triggers
- Ready-to-use test design template, geo-split calculator, and results dashboard
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 →