Practical methods, not a client performance claim. Any worked example is fictional. Actions require the relevant evidence, authority and spending boundary.
Start with the decision that is blocked
A useful campaign review ends with a reasoned action, an owner and a spending boundary. This guide groups recurring problems by the evidence I would ask for first. It is a map to the methods in this portfolio, not a list of universal thresholds or a record of client outcomes.
Use the same sequence: define the event and population, verify the measurement, check maturity and economics, then choose the smallest defensible change. If a necessary input is missing, “not ready to decide” is a valid conclusion.
The source reports clicks. Where are the visitors?
First check: distinguish billed clicks, redirect requests, successful page responses and measured sessions. Align the time window and trace a permitted sample through the actual landing path.
Decision: investigate the first demonstrated failure before judging source quality. A gap or short visit alone does not prove fraud. Stop exposure to a broken route when required by the agreed safeguard.
Follow the click-to-visit diagnosis and fictional worked example.
The test has spent money and has no conversions
First check: is the intended event being recorded, and is the acquisition cohort old enough for that event and its approval process? Check cost freshness and the remaining test allocation.
Decision: do not label a placement permanently unviable from a small, immature sample. At the cap, stop additional exposure even if the performance question remains unanswered. Holding a decision is not unlimited buying.
RedTrack has a conversion, but the source does not
First check: identify which postback leg failed. Match one event to its captured source identifier, event mapping, outbound request and destination report. An HTTP 200 is a technical response, not proof of commercial approval.
Decision: repair the demonstrated path and verify with supported diagnostics or an isolated destination. Do not send fake FTDs into live optimization to make a test look successful.
Several postbacks arrive. How many customers converted?
First check: distinguish delivery attempts, event IDs, event types and unique customers. A retry can repeat an event; one customer can also generate several legitimate events. Closely timed requests from one IP are not a complete explanation.
Decision: reconcile the event and deduplication contract before counting outcomes. Keep deposit events, unique depositors, FTDs and approved FTDs separate.
Read the four-postback example: retries, event updates and separate conversions.
Registrations are cheaper, but valuable outcomes are not improving
First check: compare comparable acquisition cohorts at the same age and carry pending approvals separately. Ask whether the downstream event is a paid order, validated lead or approved FTD, and whether its value is actually known.
Decision: do not move budget solely toward the cheapest upstream action. If downstream data is unavailable, describe the quality question as unresolved rather than claiming profit.
CTR improved, but acquisition did not
First check: the creative's promise, landing-page message and downstream outcome. A more clickable claim may attract people who are less suited to the offer. Traffic mix and delivery allocation may also have changed.
Decision: test one creative or one landing-page element while holding the other stable. Define the downstream decision metric and risk cap before the test; more clicks alone do not select the winner.
Is this creative fatigue or a change in the traffic?
First check: compare the same placement, GEO, device, offer and cohort age. Inspect exposure or frequency where the source actually provides it, along with price and delivery changes. Frequency unavailable means unknown, not zero.
Decision: frame fatigue as a hypothesis. If the tracking path and offer still work, test one replacement creative against the control under comparable delivery. Changing targeting and the lander at the same time would muddy the answer.
CPA rose after a budget increase
First check: whether extra spend bought the same mix of placements, devices and hours. Separate the original and additional acquisition cohorts, allow outcomes to mature, and check simultaneous changes.
Decision: evaluate the marginal traffic inside a pre-agreed scale test and rollback boundary. A blended before-and-after average cannot isolate why performance moved or establish incrementality.
The campaign is not spending its budget
First check: campaign and creative eligibility, schedule and timezone, account balance, targeting restrictions, source availability, bid strategy and actual budget limits. Check the destination too; a rejected or unavailable page may be part of the problem.
Decision: increasing budget is not a diagnosis. If bidding is the supported constraint, define a bounded bid experiment and hold other controls steady. Different sources expose different controls; use their current documentation.
Distinguish a bid experiment from budget scaling.
Detailed method: Campaign not spending? Find the constraint →
Retail ROAS looks good. Is the campaign contributing anything?
First check: the revenue basis, order status, discounts, refunds, product costs, fees and fulfilment. Check stock and the promotional price on the advertised variant. Do not sum overlapping attributed revenue from several platforms.
Decision: reconcile contribution and operational constraints before expanding. If costs or returns are unresolved, the profitability claim remains provisional even when the ad dashboard looks strong.
The automation says it changed a setting. Did it?
First check: the destination state after the write, actual subsequent delivery and whether another writer could have overwritten it. Keep supported live actions separate from shadow proposals and scheduled work that has not yet run.
Decision: record a mismatch as unresolved. A successful request is not sufficient proof. Keep spending expansion behind the required approval, and define recovery scope explicitly.
See the Worker-first architecture and ownership boundaries.
Detailed method: Did the campaign change actually take effect? →
What should a buyer hand over at the end of a review?
First check: whether someone else can reproduce the comparison. Record the source and campaign slice, acquisition window, timezone, event definition, billed-cost status, mature and pending outcomes, and the leading alternative explanation.
Decision: name one next action, its owner, approved cap, next review time and reversal condition. State “no change” when evidence does not justify one. A screenshot without those fields leaves the next buyer guessing.