Placement Quality Audit — Performance Marketing Agentic Workflow

schema_version: 1
version: "1.0.0"
id: placement_audit_bot
title: "Placement Quality Audit"
description: "Analyze placement report for MFA risk, viewability issues, and quality scoring. Outputs exclusion list, allowlist recommendations, and quality scores."
overview: |
  A placement-quality analysis chain that ingests a display or programmatic placement report and scores each domain or app on made-for-advertising risk, viewability, invalid traffic indicators, and brand-safety signals. Problematic placements are grouped into an exclusion list; high-performing ones become an allowlist candidate set.

  Run it monthly or after any significant spend increase in open-exchange display. Output is intended to feed back into the DSP or network as an exclusion or allowlist update, and to drive budget re-allocation away from low-quality inventory.

category:
  level: task
  domain: marketing
  reusable: true
params:
  placement_data: string
  viewability_target: string
  brand_safety_level: string
steps:
  - type: llm.generate
    prompt: |
      You are a programmatic quality analyst auditing placement data.

      **Placement data:** {{placement_data}}
      **Viewability target:** {{viewability_target}}
      **Brand safety level:** {{brand_safety_level}}

      Analyze each placement against MFA signals and quality criteria.

      **MFA Detection (5-signal checklist per placement):**
      1. Ad density signals (>30% mobile / >50% desktop = fail)
      2. Ad refresh behavior anomalies
      3. Traffic source composition (>50% paid = arbitrage flag)
      4. Content quality (AI-generated, templated, non-unique)
      5. Site design/UX (cookie-cutter, aggressive popups)

      **Quality Scoring (0-100 per placement):**
      - Viewability (30% weight): >70% = 100, 50-70% = 70, 30-50% = 40, <30% = 0
      - MFA risk (25%): 0 signals = 100, 1-2 = 50, 3+ = 0
      - IVT rate (25%): <2% = 100, 2-5% = 70, 5-10% = 30, >10% = 0
      - Performance (20%): Above avg CPA = 100, avg = 70, below avg = 30, no conv = 0

      **Output format:**

      ## Placement Quality Summary
      - Total placements analyzed: [count]
      - Average quality score: [0-100]
      - MFA risk placements: [count and % of spend]

      ## Exclusion List (Quality Score <40)
      | Placement | Quality Score | Primary Issue | Spend Wasted |

      ## Watch List (Quality Score 40-60)
      | Placement | Quality Score | Risk Factor | Recommendation |

      ## Allowlist Candidates (Quality Score >80)
      | Placement | Quality Score | Viewability | Performance |

      ## Supply Chain Issues
      [Any ads.txt/sellers.json/SupplyChain validation failures]

      ## Estimated Savings
      [Total spend on excluded placements = recoverable budget]
      [Projected CPA improvement from exclusions]
    as: audit_result

  - type: variable.set
    name: result
    value: "{{audit_result}}"