Account Audits
Part of Search ad experiments
Recording an inconclusive search campaign test
Document an uncertain search experiment without declaring a false winner. Capture the result, limits, business decision and next action.
Record a search campaign test as inconclusive when the evidence cannot support a dependable choice between the original and the trial. Keep observed results separate from the decision they cannot justify. “No clear winner” does not mean the versions are equal, and a higher point estimate alone does not make the trial a winner.
Preserve the question and decision rule
Start the record with the hypothesis, intended change and outcome chosen before launch. State the smallest improvement the business would find useful and the downside it agreed to tolerate. Otherwise, a favourable secondary number can be mistaken for success after the results arrive.
Record the campaign and experiment IDs held by the account team, original and trial settings, traffic split, dates, conversion actions and business outcome source. Check and record the final serving state separately from the written conclusion.
Record what happened
Use “unknown” where evidence was not obtained:
| Field | Record |
|---|---|
| Delivery | Serving dates, traffic by arm and interruptions. |
| Primary platform result | Each arm’s result, estimated difference and interval if available. |
| Business result | Accepted enquiries, retained orders or sales, including pending cases. |
| Timing | Conversion and assessment delay; incomplete recent periods. |
| Deviations | Changes to ads, pages, bids, budget, goals or targeting. |
| Decision | Retain, extend within a limit or stop without a winner, with the reason. |
Record any available uncertainty measure for the preselected metric. A no-clear-result label alone does not identify which explanation applies.
Compare the interval for the preselected metric with the business’s decision threshold. An interval still compatible with both a worthwhile improvement and an unacceptable outcome leaves the decision open.
A narrow interval around a commercially trivial difference suggests a different next step from a wide interval spanning major benefit and harm.
If the business outcome is assessed outside Google Ads, the platform’s interval does not describe uncertainty in that separate outcome.
Key metrics to track in inconclusive tests
- Preselected metric interval
- Range showing uncertainty around estimated difference
- Decision threshold
- Minimum improvement the business finds useful
- Conversion delay
- Time between click and final business outcome (e.g. sale)
- Mature data check
- Confirmed when all conversions are assessed within window
- Uncertainty measure
- Standard error or confidence interval for key metric
Explain the limit and next action
Name the limitation as specifically as the evidence allows: too few mature outcomes, uneven or failed delivery, multiple changes, or insufficient business follow-up. More than one may apply. Use the account’s conversion delay and window to decide whether recent data is mature.
For example, a hypothetical record could say: “The trial estimate was higher, but the interval still includes results below our minimum useful improvement. Several recent enquiries remain unassessed. Keep the original offer while those records mature, then decide whether another test is justified.”
If delivery was sound but uncertainty remains large, a planned extension may be reasonable within the agreed spend limit. If the comparison was compromised, design a fresh test and retain the flawed result as history.
If the plausible upside is too small, keeping the original can be a sensible operating choice without claiming proof that it is superior. Add an owner, review date and the evidence that would change the decision; preserve later updates as separate dated entries.
