Costs
Part of Search advertising: an intent and outcome framework
Search advertising improves when queries get inspected before bids do
How to improve search advertising: diagnose queries, promises, destinations, outcomes, values, economics, bidding, exclusions, capacity, and experiments in order.
What to take away
- Trace one query from search term to booked revenue before changing bids.
- Search-term reports are incomplete evidence, but they still support a useful review.
- Fix the earliest broken step, then test one change under a written rule.
How to improve search advertising begins with diagnosis, not a recommendation score. Choose a completed period and trace representative search terms through matching, ad copy, destination page, tracking event, qualification, sale, cancellation, refund, contribution, and customer experience. Mark unknowns. Do not fill them with guesses.
Inspect actual search evidence
Google states that its search terms report shows searches with significant activity, and it notes privacy-related omissions and differences from grouped insights. Use the report to refine campaigns. Label it as incomplete evidence.
Classify visible search terms
- Customer situation
- Eligibility
- Product
- Location
- Stage
- Risk
- Record whybelongs, not, review
Classify visible terms by customer situation, eligibility, product, location, stage, and risk. Record why a term belongs, does not belong, or needs review. A term with few clicks can still matter if it signals an unsafe claim or a systematic mismatch.
Worked example. A plumbing account bids on the broad match term "plumber near me" at a $60 target CPA. The search terms report shows "plumber courses" and "plumber apprenticeship". Both arrived through broad match.
The ad promises 24/7 emergency service, so a course seeker will not convert. Add "course" and "apprenticeship" as negative keywords in a shared list.
Move "emergency plumber" plus the city into an exact match ad group. Point that group at a page listing the callout fee and the response window.
Count a phone call of 60 seconds or more as the conversion. As an illustration, 40 calls at $45 each, with a 30 percent close rate on $180 jobs, would pay back. Raise the group budget and keep the negatives.
Check forecasts against your own data
Keyword Planner forecasts come from your settings and historical information. They are not guarantees of future traffic, prices, or outcomes. Save the date, geography, language, network, and assumptions. Compare the forecast with your own cost per conversion before you move a bid.
Diagnostic layers and first repairs
Diagnostic layer
- Eligibility
- Can serve now?
- Matching
- Why did query enter?
- Promise
- Ad remains accurate?
- Destination
- Page fulfills request?
- Outcome
- Event unique and valuable?
- Economics
- Next dollar creates contribution?
Question
- Eligibility
- Update geography, schedule, stock
- Matching
- Adjust structure, match, negatives
- Promise
- Rewrite, pin, restrict, remove
- Destination
- Send to specific accessible page
- Outcome
- Repair tracking and downstream status
- Economics
- Change limits, offer, operations, stop
First repair
- Eligibility
- Matching
- Promise
- Destination
- Outcome
- Economics
| Diagnostic layer | Question | First repair |
|---|---|---|
| Eligibility | Can the business serve this request now? | Update geography, schedule, stock, or exclusions |
| Matching | Why did this query enter? | Switch to exact or phrase match, split ad groups, add negative keywords |
| Promise | Does the ad match the request? | Rewrite, pin, restrict, or remove unsupported assets |
| Destination | Does the page fulfill the request? | Send to a specific accessible page |
| Outcome | Is the event unique and valuable? | Repair tracking and downstream status |
| Economics | Does the next dollar create contribution? | Lower tCPA or tROAS targets, shift budget to campaigns above target, or pause |
- Freeze a baseline before major edits.
- Fix claims, destinations, tracking, and capacity before bids.
- Write the hypothesis, outcome, guardrails, split, and duration.
- Review the result after the outcome window closes.
As a typical rule, wait for 30 to 50 conversions per variant before you judge it. Below that, treat the gap as directional. One search term needs roughly 100 clicks before its cost per conversion is stable.
Change one bid strategy at a time. A typical move is Maximize clicks to Maximize conversions with a tCPA cap. Keep the old strategy in the control group.
Protect interpretation during the test. Avoid changing offers, pages, location rules, conversion values, or sales response unless the design accounts for them. Record incidents and exposure. Decide under the prewritten rule.
After the decision, verify what operators actually applied. A partly implemented winner can look weaker later, and a losing treatment kept in some markets contaminates the comparison. Store the rollout date, scope, exceptions, owner, and follow-up result.
Assign ownership and verify rollout
Two public-sector references sit outside search advertising: the GOV.UK technology selection guidance, which covers technology choices, data control, and security review, and the CISA software acquisition fact sheet, which covers supply-chain exposure. Treat them as procurement context only.
Assign each step to a named role and require an observable finish condition. The analyst must be able to reproduce the result. The decision owner can explain the action and the stop rule.
Include one broken-data case and one revoked-access case. Record the repair, time required, and vendor help, so the written process reflects normal operation rather than a prepared demonstration. A written process still needs a media buying decision brief before any budget moves.
For how to improve search advertising, keep the evidence record beside the decision so a reviewer can reproduce the reasoning without relying on memory.
Common questions
What should be improved first?
Fix material claim, eligibility, destination, tracking, fulfillment, access, or financial errors before tuning delivery.
How many changes belong in one test?
Use one interpretable contrast unless a factorial design and enough evidence support more.
When is improvement proven?
When a credible comparison shows a practically useful outcome under acceptable cost, safeguards, execution, and uncertainty.
