Paid Media Data Quality Monitor

A lightweight Python project for validating daily paid media campaign files before they feed executive dashboards, pacing reports, or downstream attribution models.

This repo is framed for a Marketing Data Operations Analyst role at a streaming company: the sample data models the kind of cross-platform media delivery exports used to monitor subscriber acquisition, launch campaigns, and always-on performance marketing.

What It Checks

The monitor validates CSV rows at the campaign-date-platform grain and flags:

• Missing or invalid dates
• Duplicate campaign/date/platform rows
• Negative spend
• Spend recorded with zero impressions
• Clicks greater than impressions
• Conversions greater than clicks
• Stale data, based on the latest available reporting date
• Spend spikes versus recent campaign/platform history

The goal is not to replace a warehouse test framework. It is a practical first line of defense for recurring file drops from platforms, agencies, or internal media teams.

Project Structure

paid-media-data-quality-monitor/
  data/
    sample_paid_media.csv
  paid_media_quality/
    __init__.py
    checks.py
    cli.py
  tests/
    test_checks.py
  pyproject.toml
  README.md

CSV Schema

The validator expects a header row with these columns:

Column	Type	Notes
date	YYYY-MM-DD	Daily reporting date
campaign_id	string	Stable campaign identifier
campaign_name	string	Human-readable campaign name
platform	string	Example: Meta, TikTok, YouTube, Google Ads
spend	number	Media spend for the date
impressions	integer	Delivered impressions
clicks	integer	Delivered clicks
conversions	integer	Conversion count, such as sign-ups or trials

The included sample file is synthetic and modeled after common public paid media export schemas. It intentionally contains quality issues so the CLI has meaningful output.

For a real public dataset path, this repo includes data/public_facebook_ads_sample.csv, a validator-ready 150-row sample from the Facebook ads conversion dataset mirror hosted on Hugging Face, originally associated with Kaggle's Sales Conversion Optimization data. To regenerate it:

python scripts/fetch_public_facebook_ads_sample.py
python -m paid_media_quality.cli data/public_facebook_ads_sample.csv --fail-on error

The included file maps Impressions, Clicks, Spent, and Total_Conversion into the validator schema. The synthetic files remain in the repo because they are better for demonstrating clean and intentionally broken checks.

Quick Start

Requires Python 3.10+.

cd paid-media-data-quality-monitor
python -m paid_media_quality.cli data/sample_paid_media.csv

Expected behavior: the command prints a severity summary, issue details, and exits with status code 1 when issues are found. This makes it suitable for scheduled jobs or CI checks.

To allow warnings while still failing on errors only:

python -m paid_media_quality.cli data/sample_paid_media.csv --fail-on error

To tune freshness and spike rules:

python -m paid_media_quality.cli data/sample_paid_media.csv --max-age-days 2 --spike-multiplier 2.5 --spike-lookback-days 7

Example Output

Paid Media Data Quality Report
Rows scanned: 16
Issues found: 7

Severity summary:
  error: 5
  warning: 2

[error] duplicate_grain row 3: Duplicate campaign/date/platform row for STRM-001 on Meta at 2026-04-25.
[error] negative_spend row 8: Spend cannot be negative for STRM-002 on TikTok.

Running Tests

cd paid-media-data-quality-monitor
python -m unittest discover -s tests

Why This Matters

Marketing teams at streaming companies often make daily decisions from paid media performance data: where to shift budget, which title launch needs support, whether acquisition costs are drifting, and whether platform delivery is healthy. A few bad rows can distort pacing, ROAS, or subscriber forecasts.

This project shows how I would operationalize data trust:

• Codify checks around business-critical grain and metric rules
• Keep validation explainable for non-engineering partners
• Make the CLI automation-friendly for scheduled pipelines
• Use small, focused tests to protect expected behavior
• Document assumptions so Marketing Ops, Data Engineering, and Analytics can align quickly

Next Improvements

• Export results as JSON or Slack-ready Markdown
• Add per-platform schema adapters
• Store historical baselines for stronger anomaly detection
• Add CI workflow examples for automated validation on new CSV drops