Agent-driven anomaly detection for spend + pipeline

Problem

Issues surfaced late (tracking breaks, spend spikes, pipeline drop-offs) because monitoring was manual and reactive.

Context & constraints

• High daily spend across multiple channels
• Tracking dependencies across pixels, web events, and CRM
• Alerts needed clear ownership and escalation rules
• False positives required active management

Approach

• Define critical KPI invariants and acceptable ranges by channel and segment.
• Automate detection + triage summaries (what changed, suspected causes, next checks).
• Route alerts into the team’s workflow with clear ownership and escalation rules.

What Shipped

• KPI monitors with anomaly detection and severity tiers
• Agent summaries that surface likely causes and next actions
• Ticket/alert routing with ownership and escalation
• Postmortem templates and continuous improvements to reduce repeats

Operational outcomes

• Earlier detection of spend and attribution breakage
• Faster triage with likely-cause summaries and next checks
• Reduced downtime/spend waste through routing + runbooks
• Repeat incidents decreased via postmortems and ownership

Governance & Safety

• Thresholding + rate limits
• Human-in-the-loop escalation
• Audit logs for alerts/actions