Run a Paid Search Test Across Branches Without Confusing Seasonality With Lift

Why a simple before-and-after test misleads

A chain increases local search budgets in October and sees more enquiries. That does not isolate the effect of the spend: festival demand, store hours, inventory, competitors and branch staffing may also have changed. A branch-based test can be useful, but only if the comparison is designed before results are visible.

Choose a test question narrow enough to answer

Examples: does adding a service-specific campaign improve qualified bookings at eligible branches? Does a revised local landing page increase completed appointments from paid search? Does a different budget allocation improve contribution margin? Do not change budget, landing page, offer and routing at the same time if you need to know which change mattered.

Match branches before assignment

Build pairs or groups using pre-test characteristics: service mix, historical lead volume, catchment overlap, operating hours, conversion rate, capacity and seasonality. Exclude branches with planned renovations, opening events or major stock constraints unless those conditions are part of the question. Assign the test and comparison conditions before launch and document why each branch was included.

Define the scorecard in advance

Pick one primary outcome—such as qualified leads per eligible branch-week or completed bookings per spend—and a short list of guardrails: cancellation rate, lead quality, customer complaints, cost and capacity. Use a fixed measurement window that includes the typical delay between click and outcome. If conversions mature slowly, do not judge the test from the first few days.

Protect the comparison

A nearby branch may receive spillover from the test branch's campaign or from shared brand search. If catchments overlap, record it and decide whether to exclude the pair, analyse the pair together or accept contamination as a limitation. Keep offers and operating conditions comparable unless they are the tested variable.

Analyse honestly

Compare change in the test group with change in the comparison group, not just their final totals. Inspect branch-level variation: an average lift driven by one high-volume location may not generalise. Report uncertainty and sample size limitations. If there are too few branches for a robust inference, call it a pilot and use the result to improve the next test rather than claiming proof.

Example plan

Question: does a service-specific search campaign increase qualified repair bookings? | Units: 12 branches | Matching: prior 8-week qualified-booking rate, service capacity and region | Primary outcome: completed qualified bookings per branch-week | Guardrails: cost per booking, cancellations, capacity | Duration: set before launch | Exclusions: branches with planned closures.

Stop and rollback rules

Pause if a campaign sends leads to branches unable to fulfil the service, if tracking fails, or if the test condition leaks into the comparison group. Record all deviations. Do not retroactively redefine success because results are disappointing.

Sources and limitations

Platform experiment features vary by campaign type. This is a business-level test design; verify the current Google Ads implementation and use a statistician for high-stakes budget decisions.