Google describes Conversion Lift as a user-based holdout experiment that estimates incremental conversions caused by ads. Access is limited, study requirements vary, and Demand Gen-only studies can include modeled delayed conversions. A Gmail-serving campaign can participate through Demand Gen, but the study must remain a campaign-level randomized experiment—not a placement split improvised from reporting columns.

VISUAL LESSON

What you will learn

  1. 01Define the decision and conversion contract for a lift study.
  2. 02Prepare an eligible user-based holdout design.
  3. 03Report delayed modeled and attributed metrics separately.
A Demand Gen audience divides into randomized exposed and holdout groups that feed an incremental conversion calculation
Conversion Lift depends on a protected comparison: the holdout, conversion definition, campaign grouping, and study window must stay intact.

ILLUSTRATIVE WORKED EXAMPLE

Read an illustrative holdout result cautiously

Eligible usersStudy universe
100k
HoldbackIllustrative only
10k
Observed differenceBefore modeling
+24
Power threshold metCheck live estimate
Unknown
Illustrative example—not a benchmark. Replace every sample value with your own campaign, market, and measurement data.

PRACTICAL INTERFACE MAP

Create the study without breaking the comparison

Contract01Choose the decision and outcomes

Record campaigns, conversion actions or value, attribution boundary, window, geography, audiences, other media, CRM imports, and minimum decision threshold.

Study02Use the current Conversion Lift flow

Confirm account access, select campaigns once, review overlap, set dates and holdback, inspect the power estimate, and obtain the required approvals.

Operate03Protect and report the experiment

Avoid unplanned campaign changes, log external events, monitor status, and separate attributed conversions, lift estimates, and delayed modeled results.

Conceptual walkthrough. Labels, controls, and availability can vary by account, region, plan, and interface version; verify the current screen before acting.

STEP-BY-STEP LESSON

Eligible population → randomized holdout → conversion evidence → incremental estimate

DefineMeaningful outcome
ProtectStable comparison
InterpretIncrementality, not attribution

THE LEAD ATLAS METHOD

Lead Atlas Data can research contacts for a campaign’s chosen market, categories, and locations; its outbound results should remain outside the randomized Demand Gen study unless the approved experiment design explicitly accounts for them.See how custom list research works ↗
01

Define the incrementality decision

State what decision the study will change, which Demand Gen campaigns belong together, the eligible population, geography, dates, conversion actions or value, conversion source, lookback, import schedule, privacy controls, other media, seasonality, and accountable owners.

Choose outcomes that represent real business value and are measured consistently for exposed and holdout users. A page view or shallow click can be technically convenient but may not answer whether advertising created an incremental qualified action.

02

Verify eligibility and power

Confirm that Conversion Lift is available in the account and review the current setup with the Google contact or interface guidance. Select the campaigns, conversion actions or value setting, study dates, holdback, budget, and audience, then inspect the study power estimate before approval.

Google notes that studies generally need more than 14 days even though a seven-day minimum may appear, and that access and requirements vary. Treat the live power estimate and study status as gates, not decorative forecasts.

03

Protect the randomized comparison

Group the intended campaigns once, document overlap with Search, YouTube, social, email, offline outreach, promotions, and other Demand Gen activity, and avoid moving campaigns in or out after the study begins. Freeze material targeting, budget, creative, conversion, and bidding changes unless the study owner approves them.

Do not attempt to recreate a holdout by comparing Gmail placement reports with another placement. The randomized experiment applies to eligible users and selected campaigns; platform placement totals answer a different descriptive question.

04

Monitor study and data health

Track study status, serving, budget, audience size, conversion volume, tag and import health, value and currency, CRM delays, duplicate actions, policy events, site outages, major promotions, and every approved campaign change. Keep dates and evidence for any external shock.

Demand Gen-only Conversion Lift studies can include delayed incremental conversions as modeled projections by default. Preserve Google’s exact metric definitions and separate observed counts, attributed conversions, estimated lift, and delayed modeled results.

05

Make a bounded decision

Report study design, population, holdback, campaigns, conversions, dates, status, power, lift estimate, uncertainty, delayed modeling, attributed results, placement delivery, limitations, and whether the prewritten decision rule was met. Do not translate a non-significant or underpowered result into ‘ads had no effect.’

Deliverable: incrementality brief, conversion contract, eligibility evidence, power estimate, campaign grouping, holdback record, overlap map, frozen-settings capture, change and incident log, metric dictionary, study report, decision memo, follow-up plan, and owner.

THE TAKEAWAY

Use Conversion Lift only with eligible campaigns, a stable conversion contract, adequate power, protected holdout, and precise reporting that separates incremental estimates from attributed conversions and Gmail placement delivery.

OFFICIAL REFERENCES

Check the platform’s current instructions.

Platform labels, eligibility, and workflows can change. These official help pages were used to validate this lesson.