Overview

Props Media is a New York-based advertising agency specializing in influencer marketing. Creators publish content through their platforms, and Props runs paid media campaigns through creator profiles on behalf of major national brands. Their business depends on granular, accurate performance data across every ad platform, creator, episode, and conversion event.

When Props came to MetricMaven, they were in a difficult position. The SuperMetrics consulting program had ended in mid-2025, and the agency was referred to another firm to take over their data infrastructure. Months into that engagement, the migration had stalled at roughly 30% completion. The new data models didn’t work, the consultancy had effectively walked away, and Props was left with a half-finished system and no path forward.

The original Supermetrics pipelines were still running in the background, but they lacked the functionality Props needed for newer ad platforms and custom attribution windows. The team couldn’t move forward with the broken new system, and they couldn’t scale with the outdated legacy one.

MetricMaven was brought in to audit the failed implementation. What we found was a data architecture crisis that required a complete rebuild—not another patch job. What began as an audit evolved into a long-term strategic partnership to transform Props’ entire analytics infrastructure.

The Challenge

The scope of the problems went far beyond the incomplete migration. Years of accumulated technical debt had created an infrastructure that was actively holding the business back.

Fragmented, Undocumented Data Models

The existing dbt project contained over 100 scattered models with duplicated logic and conflicting transformations. Critical business logic—creator attribution, episode tracking, historical reclassification, margin calculations, and blacklist management—was buried across undocumented models with no clear lineage. Prior contractors had encountered this complexity, attempted rewrites, and abandoned the work.

Manual Reporting Bottleneck

Processing delays of two to three days meant the media team was always working with stale data. Over 40 hours per week were consumed by manual data pulls, reconciliation, and report assembly. The team responsible for optimizing client campaigns was instead spending their time explaining data discrepancies.

Attribution Limitations

Props’ media team required 1-day view-first attribution to accurately measure campaign performance—the standard in their industry. No off-the-shelf ETL platform (Supermetrics, Airbyte, or Fivetran) could deliver this natively. Custom conversion events required manual platform pulls, and platform-specific limitations made the problem worse: Google’s API only reports certain conversion metrics at the campaign level, and Facebook’s dynamic ad structure created additional complexity for standard integrations.

Limited Platform Coverage

The existing infrastructure only supported Facebook and Google. There was no integration for TikTok, Snapchat, Pinterest, LinkedIn, or Roku—despite growing client demand across all of these channels. Every new platform request risked destabilizing the already fragile system.

Eroding Client Trust

Data accuracy hovered around 85%, and the reporting fed into Props’ Bubble-based client application, which meant inaccurate data wasn’t just an internal problem—it was visible to their clients. Props needed a complete architectural rebuild. Another band-aid fix wouldn’t cut it.

The Solution

We executed a full transformation across three workstreams, maintaining continuous reporting with zero operational downtime throughout the engagement.

Data Architecture Rebuild

We rebuilt the entire data infrastructure in BigQuery using a medallion architecture with four distinct layers:

• Bronze: Raw data lands here directly from Supermetrics and our direct API integrations. We keep data as-is from each source but join client tables together and standardize column naming.
• Silver: Data is separated by subject matter with standardized field names across channels. Facebook’s adset_id becomes ad_group_id to match TikTok and Google’s naming conventions. Performance data is aggregated across platforms at this level.
• Gold: Finalized fact and dimension tables ready for reporting. Fact tables include base ad-level data plus breakdowns for age/gender, region, DMA, platform, and device. Dimension tables cover episodes, creators, and blacklists.
• Presentation: The final reporting tables that power Props’ Bubble application. Custom conversions across all platforms are pulled from direct integrations at this layer to cover all attribution windows and event types.

This approach consolidated the scattered models into a streamlined, well-documented architecture. All business logic—creator attribution, episode tracking, client-specific transformations, margin calculations—moved into reusable dbt macros with a single source of truth. We implemented comprehensive documentation for data lineage, automated testing to catch discrepancies before production, and a multi-tenant configuration system that standardized client onboarding.

Metadata from Props’ Bubble application—including episodes, creators, ad group targeting, ad corrections, image recognition data, and margins—flows into the bronze layer through a direct BigQuery integration managed by the Props team. Tupling between their operational platform and the warehouse ensures dimensional data stays current without manual intervention.

Direct API Integrations

Where standard ETL tools fell short, we built custom integrations deployed on Google Cloud. The architecture uses a deliberate hybrid approach: Supermetrics manages standard metrics (impressions, spend, clicks, video plays) for efficiency, while direct API connections handle the advanced requirements that no off-the-shelf tool could deliver.

• Facebook Insights API: 1-day view-first attribution with chunked historical pulls across 37 months of data. Each client gets dedicated tables for actions, platform breakdowns, DMA, region, age/gender, and device—all at the ad ID level by date.
• TikTok Events API: Custom event and conversion data pulled directly, including form submissions and landing page views that Supermetrics couldn’t provide accurately. Ad group targeting data pulled separately for audience analysis.
• Google Ads GAQL: Granular conversion action data at the ad ID and date level, addressing Google’s limitation of reporting certain metrics only at the campaign level.
• Pinterest & Snapchat APIs: Ad group-level targeting data pulled via Google Cloud Functions for accurate event and audience reporting.
• Roku Direct: Automated daily ingestion of performance data, enabling connected TV reporting alongside digital platforms.

All custom pipelines run daily starting at midnight EST, with each function fully documented in-code. The infrastructure uses a combination of Google Cloud Run jobs (for newer pipelines) and Cloud Functions (for earlier integrations), with credentials centrally managed and rotation procedures documented.

Platform Expansion & Reporting

We expanded platform capabilities from two advertising platforms to seven—Facebook, Google, TikTok, Pinterest, LinkedIn, Snapchat, and Roku.

The presentation layer feeds directly into Props’ Bubble-based client reporting application, where brands and creators can view campaign performance in real time.

To protect the system long-term, we implemented continuous monitoring with automated Slack alerts. The alerting system compares live database values against validation benchmarks, notifying both the MetricMaven and Props teams the moment a discrepancy appears—before it ever reaches stakeholders.

The Outcome

The transformation delivered measurable impact across every dimension of Props’ data operations:

• Processing time dropped by 60%: Workflows that took hours now complete in minutes
• Data accuracy improved from ~85% to 99.9%: Restored executive confidence in every report
• Data freshness issues eliminated: Replaced 2–3 day delays with reliable daily updates
• Platform coverage expanded from 2 to 7: Full omnichannel attribution unlocked
• Creative-level reporting tracks 50,000+ unique assets: Granular performance insight
• 40+ weekly hours freed: Manual reporting replaced with full automation
• True 1-day view-first attribution achieved: Full reconciliation across all platforms

The team that once spent over 40 hours weekly on manual reporting now focuses entirely on campaign optimization and client strategy. New clients are onboarded into the data infrastructure in under an hour, and when bugs surface, they’re resolved within a single business day. The system scales automatically as Props adds new ad accounts—no manual pipeline reconfiguration required.