How *Conversion Lift* studies earn trust

What Conversion Lift actually measures#

Conversion Lift answers one question: how many extra conversions did the ads cause beyond what would have happened anyway. Test cells receive the campaign, control cells are held back, and the gap between them is incrementality with statistical confidence attached.

This differs from A/B testing, which compares two strategies, and from attribution, which assigns credit. Lift asks whether the spend should exist at all. The hub frames the family, and our incrementality breakdown places lift inside the wider method.

Run lift where the budget question is existential: does prospecting add demand, does retargeting merely harvest, does the brand layer pay. Small tactical questions deserve cheaper reads.

Designing test and control cells#

Cells split the eligible audience randomly at user level, with the control held unexposed to the tested campaign while seeing everything else normally. Randomisation quality decides everything; sloppy splits corrupt conclusions before spend starts. Keep structures stable through the study: significant edits mid-test invalidate the read. Our A/B rules set the stability discipline.

Size cells for the lift you need to detect: small expected lifts need large audiences and patient budgets. Pre-commit the hypothesis, the primary metric and the read date before launch.

Isolate the tested variable ruthlessly: one campaign, one question, one read. Multi-variable studies answer nothing with confidence. Meta own guidance on stable experiment structure reinforces why untouched learning matters.

Budgets, duration and polling#

Budget for reach inside both cells, not efficiency: lift studies buy statistical power, and starved cells return the most expensive answer of all, which is inconclusive. Duration must cover at least one full buying cycle plus learning; short studies flatter novelty and punish familiarity.

Polling reaches control users to capture cross-device outcomes the pixel never sees, which matters where journeys span phones, desktops and stores. Design polling into the study from day one rather than bolting it on after.

Hold creative steady through the flight: rotating concepts mid-study confounds the very effect being measured. Freeze, run, read, then iterate. Learning windows set the minimum patience.

Reading results without self-deception#

Read the pre-committed primary metric at the pre-committed date, with confidence intervals attached. Peeking weekly and stopping at the first green number manufactures false wins; discipline here is the entire game. An honest inconclusive beats a confident fiction that misdirects next quarter.

Segment lift by funnel stage and audience where power allows: prospecting usually lifts harder than retargeting, and new customers differently from existing ones. Prospecting splits frame the segments.

Translate lift into calibration multipliers for everyday reporting so the study keeps paying after the read date. Calibration method shows the maths, and directional examples live among Meta own business success stories.

Mistakes that kill studies#

The classic kills: editing mid-flight, starving cells, peeking early, testing trivia. Each turns real money into decorative dashboards. A second tier corrupts quietly: overlapping concurrent tests contaminating cells, seasonal events landing mid-flight, sales teams unaware the control exists.

Coordinate the calendar: no site migrations, repricing or competing experiments inside the study window. Announce holds to stakeholders so nobody panics at control-group silence.

Keep a study ledger with hypothesis, design, outcome and the budget decision taken. Ledgers compound; memories evaporate. Scaling decisions should cite the ledger line.

Your first study this quarter#

Pick the biggest budget question, usually prospecting incrementality. Freeze one campaign, split cells, fund for power, pre-commit the read date, read once, calibrate, decide. That is the whole programme at minimum viable size.

Graduate from lift studies to geo designs as spend scales, and to quarterly cadence when the ledger proves its worth. The incrementality breakdown maps the path upward.

Ask our team for a lift-study design review before you commit budget. One hour of design scrutiny saves quarters of misread spend.

Frequently asked questions#