Test broader search matching with a defined learning budget

Name the demand the test should discover

A useful test starts with a hypothesis, such as finding customers who describe a service using related wording the account currently misses. 'Get more traffic' is too vague to judge. Write the services, area, and customer situations you can fulfill, then list the mismatches you must avoid. Broader matching can reach related searches, so the business needs a clear boundary and a success measure tied to suitable work.

Source checks: About keyword matching options · About the Experiments page

Check bidding signals before widening eligibility

Confirm which conversion actions the campaign uses, and whether they represent meaningful inquiries rather than scrolling or repeated button taps. Review primary/secondary settings and any custom goals. Google's broad-match guidance emphasizes Smart Bidding, but automation still depends on what you ask it to value. If the current measurement is unreliable, fix it first or make the test's purpose explicitly diagnostic rather than presenting it as a qualified-lead growth plan.

Choose a test structure you can interpret

Where supported, use a campaign experiment to separate base and trial traffic while changing one material variable. Check which experiment type and settings your account currently offers. Otherwise document the limitations of a sequential test. Keep the service promise, destination, qualification rule, and reporting period stable where practical. Changing targeting, bids, page copy, and staff handling together prevents a clear interpretation of which change mattered.

Write a spend boundary before launch

Decide how much uncertainty the business can afford and who can stop or revise the test. This is a management guardrail, not a universal Google minimum or a guarantee that the system will learn within a particular amount. Include the normal sales-cycle delay in the review plan. Specify immediate stop conditions for broken tracking, wrong-service promises, or unhandled inquiries, while distinguishing those failures from an ordinary short-run absence of conversions.

Review the demand found, not just the click total

Classify visible new query themes and reconcile spend that cannot be classified. Review qualified inquiries, requested services, booked jobs, and contribution when available. A trial that buys inexpensive clicks but produces unsuitable requests has not met a service-quality goal. A trial with a small number of promising inquiries may still be inconclusive. State that uncertainty rather than calling every favorable percentage a statistically established winner.

Make the adoption decision explicit

Choose whether to continue collecting evidence, apply the change, revise its boundaries, or stop. Record the result using the original hypothesis and guardrails. Explain any staffing change, seasonal event, or reporting defect that affected the comparison. Keep the base configuration so the business can review what changed. The useful outcome is a controlled decision about broader demand, with known limitations, rather than a permanent expansion triggered by an attractive recommendation card.

Worked example: illustrative figures

For example, the owner permits a $200 targeting test. The base sample spends $180 and produces 6 qualified inquiries, or $30 each. The trial spends $200 and produces 5 qualified inquiries, or $40 each, although its click count is higher. That does not meet the stated qualified-inquiry efficiency goal. The owner also sees one valuable new service theme and may investigate it separately. Eleven total qualified inquiries are a small operational sample; the example does not establish statistical significance or a universal test duration.

Questions about this task