FrituurOS Marketing Manager, Darwani Works

The data existed everywhere, and nowhere.

A four-brand frituur chain ran its marketing on scattered ad dashboards, spreadsheets, and gut feel. The numbers lived in Shopify, Meta, Google Ads and the POS, but never came together into a decision. Nobody could see, in one place, which customers were slipping away, which campaigns actually moved revenue, or what to do today. The job wasn't more dashboards. It was turning all that data into one calm, obvious next action.

The moving parts.

AI Campaign AdvisorTurns a customer audience into a ready-to-send campaign brief, angles, subject lines, copy and an offer, with the expected € it recovers. Built on Claude with schema-enforced tool-use, not a chatbot bolt-on.

Customer 360 + Next-Best-ActionEvery customer gets a churn score, a predicted next order and 12-month value, plus a consent-aware next action, instant, and AI-deepened on demand.

Cohort retention & lifecycle funnelA multi-year retention view and a 5-stage member funnel, with every leak priced in € so you see what a drop-off actually costs.

Opportunity ledgerRecoverable revenue split into captured vs still-outstanding, and branded vs non-branded ad ROI, so a 100× "branded" ROAS stops hiding a 4× reality.

One-click push to MetaA full campaign → ad set → ad builder that pushes live to Meta through n8n, where the app never once touches the ad token.

Reputation diagnosisExplains WHY ratings dip, late deliveries, recurring complaint themes, recovery coverage, from internal data, not just a star average.

How it's built.

Next.js

The whole app and UI, server components keep data-dense views fast and legible on a small counter screen.

Supabase · PostgreSQL

The data layer. Every metric is a SECURITY DEFINER RPC returning JSONB, so counts over 15k+ customers and 53k+ orders stay exact, immune to PostgREST's silent 1000-row cap, all behind row-level security.

n8n

The integration spine, pulling Shopify, Meta and Google Ads in, and pushes campaigns back out. Token-isolated: it shares only the database, never an API key.

AI (Claude)

Sonnet with forced tool-use turns raw metrics into structured, validated output, campaign briefs and next-best-actions, cached to keep it fast and cheap.

How the pieces connect.

Five platforms, one calm cockpit. Here's the shape of how the pieces talk, no sensitive details, just the wiring.

Operator's browserany device

loads the static app

Next.js frontendthe cockpit UI

reads live metrics through Postgres RPCs, anon key, row-level security, and Claude turns those numbers into "state → € → action"

Supabase · PostgreSQLdata layer · RPC / JSONB

Anthropic ClaudeAI advisor · tool-use

n8n writes results straight back into the same database, the app itself never holds a platform token

n8nintegration spine

bridges every external call: pulls Shopify, Meta and Google data in, and pushes campaigns back out

Shopifyorders · storefront

Meta Adscampaigns

Google Adscampaigns · GA4

Who it's for.

◆

Multi-location F&B / retailOwners running several stores who want marketing grounded in real sales, not vanity charts.

▲

Hands-on managersPeople who need a daily action list they can act on mid-shift, not another report to read later.

●

Shopify + ads teamsAnyone already on Shopify with Meta/Google Ads who wants it all in one calm cockpit.

FrituurOS Marketing Manager

The data existed everywhere, and nowhere.

The moving parts.

How it's built.

How the pieces connect.

Who it's for.

A look inside.

Curious about this build?