Overview
Reporting inside a digital agency group was anything but uniform. Every platform had its own format. Broadcast media went further — column structures and granularity differed station by station. Planned CPM and actuals lived in separate files. Some metrics weren’t even available through APIs, forcing manual reconciliation. The correct way to assemble a report existed only inside each operator’s head.
The surface ask was “unify the platforms.” The probe revealed something deeper. What they actually wanted was reproducibility — the same quality of report regardless of who produced it. Eliminating silos, not aggregating dashboards, was the real core.
This isn’t a visualization tool. It’s a transplant of operational judgment into AI, so that every operator arrives at the same conclusion.
Approach
-
Automatic format detection. Reports dropped into the UI are classified by Gemini — platform, station, and period identified automatically. Current accuracy: 98%, reproducing the operator’s judgment.
-
APIs and manual intake, side by side. Where official APIs allow it, collection is automated. Where they don’t, the intake UI absorbs the gap — and, as a way to push closer to full automation, Gmail forwarding addresses capture inbound reports for automatic structuring.
-
Plan vs. actual auto-reconciliation. Planned CPMs and actuals that used to live in incompatible formats are mapped to a shared KPI schema. Regions that previously required manual reconciliation are now structured at intake.
-
UI that resists silos. React + TypeScript + Vite on the frontend, GCP Cloud Run as a lightweight backend. The operator reaches a conclusion in a few clicks. No tribal knowledge required.
Outcome
- Cross-platform ad-ops dashboard live and operational
- 98% AI classification accuracy breaks down reporting silos and delivers uniform quality regardless of operator
- Station-by-station format differences absorbed on the UI side
- Next phases: Gmail-based auto-ingestion pipeline, Windows support, and standalone productization for partners