Meta Ads Performance Analytics Pipeline

Case Study: Meta Ads Performance Analytics Pipeline

Executive Summary

This case study covers a marketing-analytics engineering engagement for an artisan D2C retail brand. The deliverable is a Node.js data pipeline that walks the Meta Graph API (v19.0) with cursor-based pagination, normalizes paid campaign insights, audience demographics, and placement breakdowns into flat CSV matrices, and merges Facebook Page and Instagram organic reach into a single daily JSON feed. Every number in this study is computed from the pipeline's own output artifacts.

The Business Challenge

Meta splits a brand's performance picture across products: Ads Manager reports paid spend and conversions, Page Insights reports organic Facebook reach, and Instagram insights report a third stream. Each surface paginates, rate-limits, and aggregates differently. For a small D2C team, the practical consequence was that paid and organic performance were never on the same axis — a spend spike on May 13 could not be compared against that day's organic reach without manual export gymnastics. Attribution questions ("did the campaign lift organic discovery?") were unanswerable.

System Architecture

The pipeline is a set of small, single-purpose Node.js collectors sharing one pagination engine:

1. Cursor-Paginated Fetcher: A generic fetchAllPages loop follows paging.next cursors until exhaustion, so every collector transparently handles Meta's page-size limits and returns complete result sets.
2. Ads Insights Collector: Queries the /insights edge with eight metric fields per request — spend, impressions, reach, clicks, CPC, CPM, actions, and inline link clicks — with a configurable time_range and time_increment for daily granularity.
3. Breakdown Collectors: Separate collectors pull age/gender demographic splits (391 rows) and per-platform placement performance (80 rows), normalized to flat CSV for spreadsheet-free analysis.
4. Organic Merger: Facebook Page page_impressions_unique and Instagram ig_reach daily metrics merge by date into one JSON series — 31 consecutive days in the audit window.

[Graph API v19.0] ──(cursor pagination)──> [Collectors: ads / demographics / placements]
        │                                          │
[Page + IG daily metrics] ──(merge by date)──> [Unified daily JSON feed] ──> [Dashboard prototype]

Key Engineering Decisions

• Cursor pagination as a shared primitive: Every Meta endpoint paginates; centralizing the paging.next walk in one function means no collector can silently truncate a result set.
• Eight fields, one query: Requesting spend, impressions, reach, clicks, CPC, CPM, actions, and inline link clicks in a single Insights call avoids N+1 API round-trips and keeps each daily row internally consistent — all fields describe the same attribution window.
• Flat CSV as the normalization target: Demographics and placement breakdowns land as flat matrices (471 rows total), so the brand can audit any aggregate back to its source rows without custom tooling.
• Merge-by-date, not merge-by-source: The unified feed is keyed on the calendar day, which is the axis the business thinks in. Paid spend, Page reach, and IG reach become columns of one row instead of three exports.

What the Audit Window Showed

The normalized series made the brand's real performance legible for the first time:

• ₹34,721 of spend across 18 ad rows reached 285,994 unique people and served 379,676 impressions — a spend-weighted CPM of ₹91.45.
• 8,203 link clicks put the blended link-click CTR at 2.16%, with 104 add-to-cart events attributed downstream.
• The organic series exposed reach spikes (217 IG accounts reached on May 13, 171 on May 17) that were invisible in weekly averages — both aligned with campaign flight dates.
• The 5,313-row product catalog export reconciles ad-driven sessions against the storefront's live SKU set.

Results & Retrospective

• Paid and organic Meta performance now share one daily axis; campaign reviews read from a single feed instead of three dashboards.
• Every aggregate is reproducible: re-running the collectors regenerates the same matrices from the API, and the CSV artifacts are the audit trail.
• The unified feed is the ingestion contract for the brand's dashboard prototype — the next phase visualizes the same JSON without touching collection code.