Google Ads experiments and offline conversions: testing on booked jobs, not clicks

How a custom experiment splits traffic

Google's custom experiment setup page describes these mechanics:

• the experiment shares the original campaign's traffic and budget, and you choose the share given to the experiment
• for Search campaigns you can split by cookie or by search; Display always splits by cookie so each person sees one version
• you can schedule a future start, which the page suggests if new ads need time for review
• with auto-apply, you choose whether a winner needs statistical significance or only a directional result

Eligibility rules matter for small HVAC accounts. Campaigns must be active, use Smart Bidding, and include exact or phrase match keywords. Shared budgets are not allowed, only one experiment runs at a time, and edits during a run make results harder to interpret.

Cookie split or search split?

A cookie split keeps each person in one arm, which is cleaner when people search several times before calling, as many homeowners do before a replacement. A search split assigns each search independently, which builds sample faster but lets one person see both versions. For replacement and installation campaigns, we generally prefer cookie splits. For emergency repair, where most people search once and call, a search split can be acceptable.

Why calls are the wrong finish line

If an experiment optimizes for calls, a variant that attracts more short, low-value calls can "win". Google's call conversion help lets you set a minimum call length, which filters the worst of it, but it cannot tell a booked furnace replacement from a price shopper.

Illustrative case: arm A produces 60 qualifying calls and 15 booked jobs; arm B produces 75 calls and 14 booked jobs. On calls, B wins by 25%. On jobs, A wins. Without offline data, the account would roll out the wrong version.

Offline conversion imports in brief

Google's offline conversion imports page explains the core loop:

1. Google assigns a click ID (GCLID) to each ad click that reaches your website.
2. You store that ID with the lead in your CRM or booking system.
3. When the lead becomes a sale, you send the ID back with the conversion type and date.
4. Google Ads records the conversion against the original click.

The same page recommends enhanced conversions for leads over standard imports. It adds hashed first-party data, such as an email address, to improve matching. Imported phone-call conversions are supported too, so call-led businesses can report which keywords led to sales calls.

2026 changes to note

Google's page lists two dated changes. Starting April 2026, enhanced conversions for web and for leads are combined into a single setting. Starting June 15, 2026, offline conversion imports and enhanced conversions for leads uploads moved to the Data Manager API and are blocked in the Google Ads API. If an older integration stopped uploading this summer, check that change first.

A practical HVAC import design

• Conversion 1, primary for bidding: booked job, imported from your CRM.
• Conversion 2, secondary: qualified call, measured in Google Ads.
• Value: use a conservative value per job type, agreed with your bookkeeper, rather than quoted prices.
• Schedule: upload on a fixed day each week, so experiment arms receive data on the same rhythm.

Mark the booked-job conversion as the one the experiment is judged on, and read the result with an interval as in our results guide.

Privacy and consent

Enhanced conversions use hashed customer data, and Canadian privacy law applies to how you collect and share it. Make sure your privacy notice explains it, and get advice from a qualified privacy professional if you are unsure. We set up measurement; we do not give legal advice.

Common mistakes

• Optimizing an experiment for calls when booked jobs are available.
• Uploading conversions irregularly, so one arm looks better for a week.
• Editing the base campaign mid-test after a cold-weather spike.
• Forgetting the conversion delay: replacement jobs can book weeks after the click, so leave time after the test ends before reading results.

For the full setup as a service, see measurement setup. For more method pieces, visit the newsroom, or call, text or email.