Why it matters
Google campaign changes are hard to read from before/after dashboards. Seasonality, auction dynamics, and budget pacing confound single-campaign trends. Experiments create a simultaneous control so you can compare target ROAS (tROAS), CPA, or conversion value under matched conditions.
That matters when activating user-level pLTV: you need to know whether new conversion values and bidding targets improved incremental ROAS or simply shifted mix. Experiments are one Google-native tool alongside conversion lift studies and custom holdout tests that withhold the pLTV signal on control campaigns.
Experiments have guardrails: traffic split limits, learning phase interactions, and metrics that reflect Google's attribution window, not full cohort LTV. Treat experiment readouts as directional unless paired with first-party measurement at cohort maturity.
Google Ads Experiments
Google Ads Experiments fit the test layer after signal design:
- Inputs: First-party data and revenue history in your data warehouse.
- Model and calibrate: User-level pLTV with calibration checks before live bidding.
- Activate: Churney sends values directly to Google Ads via Google Ads Conversion API (web) or GA4 Measurement Protocol on a Firebase-linked setup (app).
- Experiment: Create a Google Ads Experiment comparing BAU conversion values vs pLTV-enhanced value events and/or tROAS targets.
- Readout: Evaluate experiment metrics, then confirm with data warehouse incrementality or conversion lift study where finance needs causal proof.
The data warehouse is an input to modeling. Experiments measure how Google responds to the activated signal; they do not replace data warehouse-based LTV readout.
Category variants
| Model | How Google Ads Experiments show up |
|---|---|
| Ecommerce / DTC (web) | Split test tROAS or Maximize conversion value after enabling pLTV via Google Ads Conversion API; compare value per conversion and volume. |
| Subscription app | Experiment on Firebase-linked app campaigns after GA4 MP value events; supplement with data warehouse trial-to-paid readout. |
| SaaS / PLG (web) | Test value rules on lead or signup campaigns; align platform "conversion" with qualified pipeline definitions. |
Common mistakes
- Ending during learning phase. Both arms need stable signal volume before conclusions.
- Unequal value event setup. Control and experiment must differ only on the intended lever (value signal, bid strategy, or target).
- Ignoring conversion action mismatch. Experiment compares Google-reported metrics; if pLTV values only hit one arm, verify Google Ads Conversion API routing.
- Using experiments as sole incrementality proof. Split tests show relative performance; finance may still require conversion lift study or holdout test for causal incrementality.
- App path confusion. Web experiments rely on Conversion API; app value tests need Firebase-linked GA4 MP, not MMP import alone for tROAS.
- Changing too many variables. Simultaneous creative, audience, and value changes make readouts uninterpretable.
Advertiser lens
| Role | What they ask | What good looks like |
|---|---|---|
| Head of Performance / UA | Can we test pLTV without killing scale? | Documented traffic split, BAU arm, and minimum runtime through learning. |
| VP Growth / CMO | Did the experiment justify broader rollout? | Pre-registered success metrics plus data warehouse cohort check at maturity. |
| Marketing Analytics / Data Science | Is the experiment design valid? | One primary change, power/volume plan, and analysis calendar. |
| Data Engineering | Are value events scoped to the experiment arm? | Campaign mapping and monitoring for mis-routed pLTV deliveries. |
| Finance / Procurement | What triggers scale or renewal? | Agreement on incremental vs platform metrics before launch. |
FAQ
What are Google Ads Experiments?
An in-platform tool to split campaign traffic between a control and an experiment variant to compare performance on a defined change (bidding, budget, targeting, or creative).
How do Google Ads Experiments differ from conversion lift studies?
Experiments split traffic between campaign variants you configure. Conversion lift studies estimate incremental conversions from ad exposure using platform-managed control withhold designs. Use experiments for relative A/B style tests; use lift for incrementality questions when eligible.
Can Google Ads Experiments test pLTV value signals?
Yes, when control and experiment arms receive different value events or bidding strategies and values are delivered via Google Ads Conversion API (web) or the correct app path (GA4 Measurement Protocol / Firebase-linked setup).
What split should we use?
Balance signal volume and statistical power. Heavier experiment splits speed learning but reduce scale on the control; follow Google guidance and pre-agree minimum runtime.
How long should an experiment run?
Through learning phase and enough conversions for stable comparison, then align with cohort maturity for data warehouse LTV validation on delayed-value businesses.
Do experiments replace holdout tests?
Not always. Experiments compare campaign variants; custom holdout tests may better isolate withholding pLTV from all overlapping traffic. Many programs use both.
What metrics should we track?
Conversion volume, conversion value, CPA, ROAS or tROAS performance, and impression share where relevant. Supplement with incremental ROAS and cohort metrics from your data warehouse.
Not the same as
| Term | Difference |
|---|---|
| Conversion lift study | Platform incrementality product with randomized ad withholding; different question and setup. |
| Holdout test (custom) | You control signal withholding across campaigns; not limited to Google experiment UI. |
| Campaign draft (no experiment) | Staged changes without simultaneous control traffic split. |
| Geo experiment | Geography-based test design; may run outside Google Ads Experiments tooling. |
| A/B test (creative only) | Asset-focused test; experiments can include bidding and value changes too. |
