Full Automation Stack (n8n + Supabase)

Overview

I built Dansu from scratch, owning every function — product design, manufacturing, marketing, ops, and fulfilment. As it grew, the manual work across all those functions was consuming ~40 hours a week. I audited every recurring task, identified the bottlenecks, and systematically automated them using n8n, Supabase, RocketAPI, and the Gmail API. The result was a business that could run without me touching most of it daily.

The problem

Every operational process was manual: outreach, lead enrichment, creator sourcing, fulfilment checks, inventory forecasting, customer service follow-ups. That was fine at zero. At scale it became the ceiling. I couldn't grow without either hiring (expensive, adds complexity) or automating (the right answer for a lean DTC brand).

The risk wasn't building the automations — it was building unreliable ones that needed constant babysitting. So I treated each flow like a product: scoped requirements, built an MVP, validated reliability, and iterated based on failure logs.

What I built

CRM and outreach engine

An end-to-end pipeline from target discovery through personalised outreach to reply triage. Scraped festival sites and Instagram accounts, parsed contact info, enriched metadata, scored leads by opportunity size, fit, and complexity, and triggered automated but personalised outreach sequences. Deduplicated and throttled sends to protect deliverability. All data synced into Supabase with full histories and stage tracking.

Creator discovery and scoring pipeline

Used RocketAPI to ingest reel and creator data at scale. Scored reels by engagement velocity, recency, and repeatability to surface winning content formats. Scored creators by consistency, brand fit, and contactability. Results synced to a dashboard for content briefing and a CRM for outreach — closing the loop from insight to activation.

Operations brain

Automated the ops layer that most founders spend the most manual time on:

• Low-stock alerts and reorder-point triggers based on trailing velocity and supplier lead times
• Projected stockout date warnings feeding directly into reorder task creation
• Fulfilment exception flagging for delayed orders and missing scans
• Supplier comms workflows for reorder confirmations and QC follow-ups

Self-healing automation infrastructure

Built the flows with idempotency so nothing duplicated work if a job ran twice. Added retry logic, job queues, error logging, and proxy routing for scrapers hitting rate limits. Each critical flow had failure alerting so problems surfaced immediately rather than silently breaking.

Meta creative testing and monitoring

Automated reporting on ad performance across formats and audiences, feeding into structured creative testing cycles. Tracked ROAS by creative variant so testing decisions were data-led rather than intuition-driven.

How I approached it

I audited every weekly task first — mapped what existed, where time was going, and what the failure modes were. Highest time-cost and most repeatable tasks got automated first. I used Supabase as the single source of truth across all systems to prevent data inconsistencies between flows. Every automation started as a minimal viable version, ran against live data, and got iterated based on real failure logs rather than theoretical edge cases.

Systems I built

• n8n for all workflow orchestration across CRM, outreach, scoring, ops, and monitoring
• Supabase as the central data layer — all records, statuses, scores, and histories in one place
• RocketAPI for Instagram and reel data ingestion
• Gmail API for outreach send management and reply detection
• Retry logic, job queues, and error logging across all scraper-dependent flows
• Deduplication and pacing controls to protect deliverability and API rate limits

Impact

• Ops time dropped from ~40 hours per week to under 10
• Outreach volume and speed increased 10x compared to manual operation
• The business ran reliably without constant founder input — enabling focus on growth experiments rather than admin
• Automation layer made the business sellable: it could operate without the founder present, which was the condition for exit
• Built and sold Dansu at approximately £250k revenue, with the automated ops stack as a core part of the value