Marketing performance investigation

This playbook defines how an agent investigates an anomalous divergence in marketing metrics, cross-references performance data across delivery platforms, obtains owner context, agrees remediation with authorized implementers, and verifies execution before monitoring recovery. One run evaluates a single bounded performance anomaly identified across a designated channel, campaign, or funnel stage. The agent operates in read-only analysis and drafting mode across configured reporting tools, advertising exports, analytics platforms, and customer databases; direct API connectors are candidate integrations that fall back to manual export files if unconfigured. Budget reallocations, bid changes, creative pausing, and tracking code modifications remain under the exclusive authority of human owners, and approval alone does not constitute proof of execution.

• Inspect the initial anomaly alert or manual reporting export to define the investigation boundary, noting the affected channel, campaign ID, evaluation timeframe, and primary divergent metrics such as cost per acquisition or lead conversion rate against company-defined illustrative tolerance baselines.
• Retrieve campaign delivery records from candidate advertising connectors or manual CSV exports, recording gross spend, impressions, raw clicks, and delivery state changes across the anomaly window and the preceding baseline period.
• Compare platform-reported clicks with destination site analytics sessions and UTM attribution logs; treat traffic divergences as working hypotheses rather than conclusive proof of tracking failure, auditing site release logs and tag manager version histories to determine whether discrepancy correlates with technical deployments or genuine audience drop-off.
• Audit downstream conversion event telemetry between site analytics events and CRM contact creation timestamps to test hypotheses regarding tracking script dropouts, form validation errors, or backend ingestion delays.
• Compile a factual discrepancy summary contrasting reported platform spend, landing page traffic, recorded conversion tags, and verified CRM pipeline records, explicitly identifying unverified assumptions and data gaps where reporting connectors or logs are unavailable.
• Information request. Present the discrepancy summary and working hypotheses to the responsible campaign manager, requesting operational context regarding recent creative refreshes, budget alterations, targeting adjustments, or concurrent landing page changes.
• Evaluate the channel manager's response against platform delivery logs and technical audit findings to establish whether the performance shift stems from confirmed tracking telemetry failure, intentional experimental configuration, audience saturation, or external platform volatility; if the manager provides requested adjustments, revise the diagnostic assessment accordingly.
• Approval. If corrective intervention is warranted, submit an explicit remediation work order to the marketing lead specifying the exact proposed changes, the designated authorized implementer (such as a named media buyer or web developer), and expected post-fix performance criteria; if the lead declines the proposal, document the decision and transition the case to declined or unresolved status without applying changes.
• Once the remediation proposal is approved, require the designated authorized implementer to submit an explicit execution receipt or system read-back (such as ad network change history logs or code deployment commit hashes) confirming the approved changes are live before initiating recovery monitoring; if receipt is unconfirmed or implementation stalls, hold the case awaiting verification rather than assuming completion.
• Following verified implementation read-back, monitor the target channel and backend data pipeline across an illustrative post-fix observation window to verify that conversion telemetry stabilizes, spend aligns with expected targets, and CRM lead ingestion resumes normal volume.
• Exception. If post-remediation monitoring reveals persistent variance exceeding defined tolerance thresholds, if candidate reporting tools remain unavailable, or if tracking remains broken after technical handoff, flag an escalation alert to the growth lead and technical operations owner with raw diagnostic logs, pausing automated evaluation until manual recovery is completed.
• In a fictional acquisition campaign, the agent detects an illustrative 48% increase in cost per qualified lead over a 72-hour window while reported ad clicks remain stable. Treating the click-to-session gap as an initial hypothesis, the agent checks Google Analytics 4 sessions against HubSpot form submissions and discovers that form submissions plummeted following a Friday website header release that inadvertently stripped the form tracking container. The agent submits an information request to the paid search lead, who confirms targeting was unchanged and identifies the responsible deployment ticket. The marketing lead approves an emergency tag restoration work order assigned to the web operations lead. The web developer reverts the script and submits the deployment read-back commit hash. The agent then monitors subsequent traffic for an illustrative 24-hour observation window, verifying that form submission tracking recovers to within illustrative baseline tolerances and matches CRM lead creation.
• Conclude the case when verified performance telemetry confirms agreed remediation has restored metrics within acceptable boundaries, or record the investigation as unresolved if underlying platform volatility, permanent market shifts, or human decision to decline intervention leaves the anomaly active, keeping campaign continuity under direct human governance.