AI Operations & Reporting System

Stack: n8n (automation) · Supabase (storage) · Claude (AI insights) · Google Sheets / Airtable / People Ops API (data sources)

Overview

The AI Operations & Reporting System centralizes operational data from Sales, Project Delivery, and People Operations into a single automated reporting pipeline. Instead of manually pulling numbers from three separate systems, the workflow collects the data, validates and normalizes it, computes KPIs deterministically, generates AI-powered insights via Claude, and stores everything in Supabase for a unified dashboard view.

Architecture

• -Automation Workflow (n8n): pod1.app.n8n.cloud/workflow/qHMjuZq2muGULflF — orchestrates data collection, validation, KPI calculation, and AI insight generation.
• -Supabase Database: supabase.com/dashboard/project/sqchxhequznbkuqzqljr — the single source of truth for processed records, KPI snapshots, and AI-generated insights.
• -Dashboard: ai-operations-reporting-system.vercel.app — reads exclusively from Supabase, never hits the source systems directly, so what you see is always a consistent, stored view.

How It Works

1. -Trigger and configure. The workflow runs on three triggers: a Manual Trigger for ad hoc runs, a webhook that fires when someone selects a custom date range on the dashboard, and a schedule trigger that runs daily to keep the time-series data fresh for trend charts. Configuration then resolves the reference date and requested periods into last_30, last_90, and ytd period objects.
2. -Collect. Pulls Sales data from Google Sheets, Project Delivery data from Airtable, and People Ops data from the People Ops API.
3. -Validate and normalize. Each branch runs through a normalization step that dedupes records by their source ID and flags any data-quality issues. The system never invents missing data; problems get recorded and surfaced instead of papered over.
4. -Store. Normalized records are upserted into Supabase, keyed on stable source identifiers, so re-running the pipeline updates existing records instead of duplicating them.
5. -Calculate KPIs. Metrics are computed deterministically per period, not left to the AI to generate, so the numbers are consistent and auditable every time.
6. -Merge and summarize. People Ops and Sales KPI branches merge together, then a data quality summary step consolidates flags across all sources.
7. -Generate AI insights. The merged KPI and data-quality payload goes to Claude, which returns an executive summary, operational risks, recommended actions, and data-quality warnings.
8. -Store results. Claude's response is parsed and saved as a report snapshot, then broken out into individual insight rows (executive summary, operational risk, recommended action, data-quality warning) so the dashboard can query and display them independently.

How to Use It

• -For a quick check: open the dashboard and pick a standard period (Last 30 Days, Last 90 Days, or Year-to-Date). These are pre-computed and load instantly.
• -For a specific window: select a custom date range on the dashboard; this fires the webhook trigger and generates a fresh report for that range.
• -For daily trends: no action needed. The schedule trigger runs automatically every day to keep the time-series charts current.
• -For development or testing: trigger the workflow manually from n8n to run the pipeline on demand.

Strengths

• -Single source of truth. The dashboard only ever reads from Supabase, so there's no risk of it showing stale or conflicting numbers from a source system mid-sync.
• -Deterministic KPIs. Business metrics are calculated with defined logic, not generated by the AI, so the numbers are trustworthy and reproducible. Claude's role is strictly interpretation, not arithmetic.
• -Resilient to partial failure. If one data source is unavailable, that failure is recorded and the other sources keep processing, so one broken integration doesn't take down the whole report.
• -No silent data gaps. Missing or malformed data is flagged as a data-quality issue rather than guessed at or hidden.
• -Flexible reporting. Reports come standard (Last 30, Last 90, YTD, pre-computed), scheduled (daily trend data), and on demand (custom ranges), covering the full range of how a stakeholder might want to look at performance.

How It Helps the Business

• -Cuts out the manual work of pulling and reconciling data from Sales, Project Delivery, and People Ops separately.
• -Gives leadership one place to see performance, risks, and recommended actions instead of piecing it together from three tools.
• -Surfaces data-quality problems early instead of letting bad data quietly skew decisions.
• -Frees people up to act on insights instead of spending time producing the report.

Ways to Improve

• -Finish wiring up the Project Delivery (Airtable) branch so all three domains report end to end.
• -Add alerting so a failed data source or a spike in data-quality issues notifies someone directly instead of only showing up on the dashboard.
• -Expand AI insights to compare period over period (e.g. last_30 vs the prior 30 days) so trends are called out explicitly, not just current-state snapshots.
• -Add role-based access on the dashboard so different stakeholders see the slice of data relevant to them.