MOps Pipeline Intelligence
Last analyzed: —
Demo project · Apr 2026

AI-assisted
pipeline analysis.

Synthetic MOps data flows through a scheduled pipeline into BigQuery. Click Run Analysis — Claude reads current state and returns a health score, findings, anomalies, and ranked recommendations as structured JSON.

Architecture
Google Sheets Source
Fivetran Pipeline
BigQuery Warehouse
Claude API Analysis

Free tier throughout. Fivetran syncs every 15 minutes. Claude Haiku 4.5 handles structured-output analysis. Monthly cost at light traffic: under $2.

01 — Pipeline health
—
Not analyzed

Click Run Analysis to generate insights.

02 — Funnel

Funnel

Total leads
—
last 6 months
MQL rate
—
of all leads
MQL → SQL
—
of MQLs
Close rate
—
of opportunities
03 — Velocity

Velocity

Average days in each stage, by source
Pipeline count by stage — last 8 weeks
04 — Anomalies

Anomalies

No analysis yet. Click Run Analysis to surface anomalies.

05 — Findings

Findings

No analysis yet.

06 — Sources

Sources

Source Leads MQL rate SQL rate Close rate AI rank
07 — Campaigns

Campaigns

Cost per lead vs. cost per MQL — top campaigns
08 — Recommendations

Recommendations

No analysis yet.

09 — About this tool

About this tool

What this is, how it works, and what it would look like at scale.

This is a working demonstration of AI-assisted marketing operations analysis. A synthetic dataset flows through a scheduled, managed pipeline into Google BigQuery, where tables for leads, campaigns, touchpoints, and weekly pipeline snapshots are organized as queryable views.

When you click Run Analysis, the app queries BigQuery for current pipeline metrics and passes them to Claude with a structured prompt. Claude returns a JSON object with a pipeline health score, prioritized findings, anomalies, source rankings, and ranked recommendations. The frontend renders that structure.

What's here

  • — Segmentation: source-level and campaign-level slicing of the funnel
  • — Orchestration: scheduled sync, cached analysis, rate-limited endpoints
  • — Insights generation: Claude reads current state and surfaces what matters
  • — Predictive framing: pipeline health score summarizes forward-looking risk

Why Claude

Claude is a strong fit for structured-output tasks with tight JSON schemas and domain-specific reasoning. For classification-heavy work or high-volume lightweight tasks, a different model choice would make sense. A production version would route tasks by shape.

What this would look like at scale

The source becomes Salesforce, HubSpot, or Marketo. The schedule runs unattended against fresh data. Claude outputs write back to Slack, email, or a read-only status page. Non-technical MOps leaders check the page instead of opening BigQuery. Governance lives in the schema: canonical definitions, owned fields, versioned changes.

Stack

  • — Source: Google Sheets
  • — Pipeline: managed connector on a 15-minute sync
  • — Warehouse: Google BigQuery
  • — Analysis: Claude Haiku 4.5 via Anthropic API
  • — Frontend: HTML + Chart.js
  • — Hosting: Railway

Total build cost: $0 (all free tiers). Monthly operating cost at light traffic: under $2 in Claude API usage.

Built by Kyle Baudour. kylebaudour.com