A paid-search workflow that checks its work.
We built a system around the daily work of account management: interpreting the evidence, preparing changes and checking that approved updates landed correctly.
The numbers need context.
Paid search runs on a lot of small decisions. Which searches fit the service? Which ad deserves another test? Is a conversion a new lead, or another step in the same journey?
We built an internal workflow in Codex that combines account data with business context, prepares changes and checks the account after they’re applied.
One bad finding changed how we read the account.
An early review flagged repeated keywords as duplicates. A closer check showed that some were exclusions used to send searches to the right ad group. The structure was deliberate.
We withdrew the finding and added a check for keyword type, match type and live status before recommending a cleanup.
From account evidence to a checked change.
The account
Search terms, ads, settings and conversion definitions.
The business
Services, locations, priorities and customer intent.
- 01
Interpret the evidence
Check what each account setting actually does.
- 02
Prepare the change
The exact update, its reason and what to watch.
- 03
Review & approve
Validate the proposal and approve the specific update.
- 04
Apply the update
Use the reviewed change; create replacement ads paused.
- 05
Verify in the account
Compare the actual copy, URLs and settings with the plan.
Approved corrections inform the next account review.
The business sets the rules.
The workflow reads each client’s services, locations and priorities alongside the account. A search describing a household problem can be relevant even when it doesn’t name the service. A high click-through rate tells us something different from a booked job.
Recommendations name the evidence, the exact change and what we’d watch to judge it. Unclear intent or incomplete conversion tracking stays an open question.
Take the change all the way through.
We’ve used the workflow to add exclusions, adjust keyword routing and replace search ads. Each proposed batch is validated before approval, then checked against the current account before it’s applied.
For ad replacements, we create the new ads paused and check their copy and URLs. Activation gets a separate review. After each approved update, we read the account again and compare what changed with what was intended.
Keep the reasoning for the next review.
The record holds the change, its reason and the next check. Bryce’s corrections can become instructions for future reviews once he approves them.
After implementation, the next review looks at qualified outcomes and other changes that could affect the result. When there isn’t enough evidence, the test stays open.
Growth is being rewritten.
You’ve seen what other teams are doing with AI and wondered what more is possible for yours. We’ll help you find the opportunities worth pursuing and build a better way to work.