Campaign diagnosis
Creative and destination troubleshooting matrix
Organization byline: Measurement Press editorial desk. This page does not list an individual author or reviewer. Read the editorial method and corrections policy.
About 11 min read
When a campaign response looks weak, uneven, or surprisingly strong, the first answer should not be that the media worked or failed. The result may be coming from context fit, creative promise, page friction, offer quality, routing coverage, or the comparison design.
Use this matrix after a campaign readout arrives and before a renewal, creative change, landing-page rebuild, budget shift, or lift-test plan is approved. The goal is to locate the most plausible weak link without turning a troubleshooting note into a causal claim.
Start with the result pattern
Campaign response is easier to diagnose when the readout names the exact pattern. A single total can hide whether the issue begins before the click, after the click, after the form, or inside the comparison.
| Observed pattern | First place to inspect | What to check | Next action | Claim boundary |
|---|---|---|---|---|
| High impressions and viewability, low clicks. | Context and creative promise. | Eligible page group, placement, creative headline, offer clarity, frequency, and whether the call to action matches the reader's job. | Narrow the context or rewrite the message before judging the destination. | Can describe low observed click response; cannot prove the audience had no interest. |
| Strong clicks, low engaged sessions. | Destination match and page performance. | Load time, mobile rendering, message match, above-the-fold proof, navigation clarity, and accidental-click indicators. | Fix speed and message continuity before changing the media package. | Can flag weak visit quality; cannot assign the cause without page and device evidence. |
| Engaged sessions, few form starts. | Offer and call to action. | Whether the offer is useful enough, the form is visible, the next step is concrete, and the page gives proof before asking for information. | Revise the offer path or test a lower-friction action. | Can say the page did not convert engaged visitors under the current offer. |
| High form starts, low completions. | Form friction. | Required fields, validation errors, mobile keyboard behavior, privacy text, autofill, error recovery, and abandonment points. | Repair the form before changing creative or inventory. | Can identify conversion friction; cannot call the source low quality. |
| High submitted leads, low qualification. | Qualification definition and form filter. | Duplicate rules, disqualification reasons, fit fields, incentive quality, geography, consent status, and contactability. | Tighten qualification fields or adjust the offer promise. | Can say submitted volume outpaced qualified demand under the current filter. |
| High qualified leads, low sales acceptance. | Routing and follow-up. | Owner assignment, response time, lead notes, rejection reasons, stage definitions, and sales coverage by source. | Fix routing and disposition coverage before judging lead source quality. | Can limit the readout because acceptance coverage is incomplete or inconsistent. |
| Desktop response strong, mobile response weak. | Mobile destination and device mix. | Mobile load time, layout, form field order, click-to-call or calendar behavior, creative size, and placement position. | Keep desktop learning separate and run mobile destination QA. | Can recommend device-specific fixes; cannot pool desktop and mobile into one creative verdict. |
| Native card response differs from display response. | Format and reader action. | Native headline, sponsor label, thumbnail, destination promise, scroll depth, and whether the card asks for a different action than display. | Read native and display as separate format lanes. | Can compare format response descriptively; cannot rank inventory without holding copy and action constant. |
| Attributed conversions look strong without a holdout. | Comparison design. | Prior intent, retargeting exposure, seasonality, source overlap, attribution window, matchback rate, and whether a counterfactual exists. | Use directional renewal language or plan a designed comparison. | Can report attributed outcomes; cannot call them incremental lift. |
Minimum evidence packet
The matrix only works if the report keeps enough fields alive. If these fields are missing, the next action is usually to repair reporting before debating creative quality.
Media context
Package ID, placement ID, page group, device class, geography, date range, viewability, frequency, and buying path.
Creative record
Creative ID, format, size, headline or message theme, offer, call to action, destination URL, and approval state.
Destination record
Page version, load issue, message match, form starts, form completes, abandonment point, and mobile behavior.
Outcome record
Qualified visit, lead status, duplicate flag, disqualification reason, routed owner, first response time, and accepted status.
Comparison record
Prior flight, matched context, reserved holdout, market split, audience split, or explicit no-comparison note.
Decision record
Renew, narrow, revise creative, fix destination, repair routing, or design a stronger test before scale.
When the packet needs a field-level join specification, use the campaign data-layer spec. For a row-level private-marketplace example, compare the private marketplace campaign walkthrough before generalizing the diagnosis.
Fix, narrow, hold, or retest
The best troubleshooting output is a bounded next action. A weak result does not always mean stop, and a strong result does not always mean scale.
| Decision | Use when | Required wording | Do not say |
|---|---|---|---|
| Fix destination | Engaged traffic reaches the page but mobile speed, message continuity, form behavior, or routing is weak. | Observed response is limited by destination readiness or follow-up coverage. | The media source produced poor-quality people. |
| Revise creative | Delivery is in scope but click or visit quality is weak across otherwise healthy pages. | The current message or offer did not earn enough useful response in this context. | The audience is not interested in the category. |
| Narrow package | Response concentrates in specific contexts, placements, devices, or formats. | The narrower lane produced the clearest observed response and should be preserved separately. | The full package proved lift. |
| Hold launch or renewal | Fields are missing, creative and destination records conflict, or lead status is too incomplete to interpret. | The report is not decision-grade until the source trail is repaired. | The campaign failed. |
| Retest with a comparison | The next decision requires causal language, budget scale, or a stronger claim than the current readout supports. | The current result is descriptive or directional; the next flight should reserve a counterfactual. | The campaign drove incremental demand. |
Worked diagnosis note
The example below shows how to keep a diagnosis useful without overstating it. The values are structure examples, not benchmarks.
| Finding | Evidence | Diagnosis | Next action |
|---|---|---|---|
| Upper-body desktop placements produced the strongest qualified visits. | Higher engaged sessions, lower bounce, and more complete lead status in the approved measurement guide context. | Context and placement fit look useful. | Renew a narrower placement set and keep row-level fields. |
| Mobile clicks did not become form starts. | Mobile visit quality was acceptable, but form starts dropped sharply after page load. | Destination or offer friction is more plausible than broad media failure. | Run mobile destination QA and test a shorter step next action. |
| Native card response was strong but different. | Native copy used a different promise and a softer action than display. | Format and message changed together. | Report native separately and do not use it to rank display placements. |
| Attributed conversions rose versus the prior flight. | No protected holdout; prior-flight comparison may include timing and audience differences. | Useful directional signal, not lift evidence. | Plan a holdout or matched-market test if the next budget decision needs causality. |
Supportable close
Recommended wordingThe campaign produced its clearest observed qualified response in upper-body desktop guide placements. Mobile traffic and native-card response should be interpreted separately because device experience, message, and action differed. The next flight should renew the narrower context, repair mobile destination friction, and reserve a comparison cell if the buyer needs incremental lift evidence.
Meeting script
- Which result pattern are we diagnosing: clicks, engaged visits, form starts, completions, qualified leads, sales acceptance, or attributed outcomes?
- Do we have placement, creative, destination, device, and outcome fields for the same reporting rows?
- Did the creative promise match the page context and the destination headline?
- Where did mobile, native, or lead-routing behavior diverge from the main readout?
- Which next action is most defensible: fix the destination, revise creative, narrow the package, hold the decision, or retest with a comparison?
- Does the final sentence use descriptive, directional, or causal language with the right confidence level?
Evidence trail and a simulated triage packet
This matrix separates delivery, destination readiness, lead quality, and comparison strength because each record answers a different question. Google's destination-not-working policy says an ad destination should function on common browsers and devices and return usable responses to Google Ads crawlers. Its destination-experience guidance also flags destinations that are unnecessarily difficult or frustrating to navigate, including pages that load slowly or interfere with the requested content. For the creative record, Google's responsive display specifications define required text and image asset fields; those fields make version and format comparisons auditable, but they do not predict performance. For downstream lead quality, Google's enhanced-conversions-for-leads guidance describes retaining identifiers such as GCLID and user-provided data so later qualified outcomes can be connected to the originating interaction. These sources support a troubleshooting packet; they do not establish that any particular creative, placement, or campaign caused an outcome.
| Source boundary | What it supports in this matrix | What remains unproven |
|---|---|---|
| Destination not working | Check common-browser/device access, HTTP responses, redirects, and the actual final or expanded URL before blaming media. | A passing destination does not prove message fit, conversion quality, or campaign impact. |
| Destination experience | Inspect speed, navigation, pop-ups, interstitials, unexpected click areas, and other friction that can interrupt the requested content. | A usable page does not show that a creative caused a qualified action. |
| Responsive display specifications | Retain the headline, description, image, business-name, and call-to-action assets needed to compare versions and formats. | Required or valid assets do not predict response or identify a winning creative. |
| Enhanced conversions for leads | Preserve joinable identifiers, conversion name/time, qualification state, and privacy-safe first-party matching details when connecting later outcomes. | A matchable lead record does not establish incrementality, complete CRM coverage, or causal lift. |
Reading rule: Use the source-backed checks to decide which record needs repair and which claim must be narrowed. Treat the simulated counts below as a reproducible routing demonstration, not as evidence that any production destination, creative, or lead stream passed these checks.
Simulated worked example -- not a real campaign, benchmark, forecast, or lift result: Assume 12 diagnostic records are reviewed: 4 have a destination-readiness failure, 3 have a creative/context mismatch, 2 have incomplete lead-routing status, 2 have a missing comparison, and 1 passes the initial triage. The mutually exclusive arithmetic is 4 + 3 + 2 + 2 + 1 = 12; the shares are 33.3%, 25.0%, 16.7%, 16.7%, and 8.3%. The first action is to repair the destination failures, then separate creative/context review from lead-routing review, while describing the two no-comparison records as directional only. These invented counts demonstrate the routing method and should not be read as Measurement Press inventory or buyer performance.
| Simulated bucket | Records | Share | Next action | Claim boundary |
|---|---|---|---|---|
| Destination readiness failure | 4 of 12 | 33.3% | Check HTTP response, mobile route, message match, load behavior, and form path before changing media. | Shows a repair queue; does not prove destination caused every weak result. |
| Creative/context mismatch | 3 of 12 | 25.0% | Split format, placement, message, and reader task; revise or narrow the affected lane. | Describes an observed mismatch; does not prove audience interest or disinterest. |
| Lead-routing status incomplete | 2 of 12 | 16.7% | Preserve source and lead identifiers, owner, status, and disposition reason before judging lead quality. | Missing downstream status limits interpretation; it is not a low-quality-lead finding. |
| Comparison missing | 2 of 12 | 16.7% | Use directional wording or reserve a holdout/matched comparison for the next decision. | Attributed or before-after movement is not incremental lift without a defensible counterfactual. |
| Initial triage pass | 1 of 12 | 8.3% | Continue to read the record by device, format, destination, and outcome quality. | A pass means no first-pass blocker was found; it is not proof of performance. |
Assumptions and limits: Each simulated record belongs to one bucket, so the shares sum to 100%. A real packet must define the record grain, time window, device and format slices, destination version, qualification state, and comparison rule before calculating shares. The matrix cannot infer causality from a troubleshooting count, and the Google policy/specification sources do not validate a campaign's business outcome.
Reproducible decision packet
The bucket counts above describe where a record needs attention. A decision still needs a second pass that tests whether the records can be joined and interpreted together. The following packet is a simulated worked example, not a real campaign, benchmark, forecast, or lift result. Use 80% as the operating pass threshold for this illustration: passes = records meeting the gate / 12. A comparison gate below the threshold holds causal or scale language even when delivery looks healthy.
| Diagnostic gate | Records passing | Calculation | Packet decision | Why it matters |
|---|---|---|---|---|
| Result pattern assigned to a defined record grain | 12 of 12 | 12 / 12 = 100.0% | Pass | A pattern cannot be compared until its event, time window, and unit are named. |
| Creative, placement, destination, and version joinable | 9 of 12 | 9 / 12 = 75.0% | Revise | Three unjoinable rows can make a creative or placement explanation look stronger than the source trail supports. |
| Destination check covers common device routes | 8 of 12 | 8 / 12 = 66.7% | Revise | Repair browser, mobile, message-match, and form-path checks before attributing weak response to media. |
| Lead status, disposition, and owner are complete | 7 of 12 | 7 / 12 = 58.3% | Revise | Incomplete downstream status limits a lead-quality conclusion; it is not evidence of low-quality traffic. |
| Format and device slices retained | 10 of 12 | 10 / 12 = 83.3% | Pass | Most rows can be read without pooling materially different experiences. |
| Counterfactual or explicit no-comparison note recorded | 4 of 12 | 4 / 12 = 33.3% | Hold | Only four rows can support a comparison discussion; the packet cannot support incremental-lift language. |
Simulated packet result: 2 of 6 gates pass (2 / 6 = 33.3%), 3 require revision, and 1 is a hard hold. The next action is to repair the join, destination, and lead-status fields, then preserve a counterfactual for any decision that needs causal language. As a sensitivity check, completing two of the three missing lead-status rows would produce 9 / 12 = 75.0%, which still fails the illustrative 80% threshold; all three would be needed for 10 / 12 = 83.3%. This is a method for exposing decision fragility, not a score for a real campaign.