Performance-marketing agency, DACH · ~€3M revenue
Client operations platform
in production since March 2026 · operator retainer
The flagship. An agency managing high five-to-six-figure monthly ad spend for its clients ran its operations on spreadsheets, a patchwork of tools, and a Make.com contractor billing €10,000 a month. Today it runs on a single custom platform its clients log into, built and operated by me.
The starting point
- Ad performance lived in Meta and Google dashboards the agency’s clients never saw; reporting was manual screenshots.
- Leads arrived via webhooks nobody monitored; attribution between ad spend and closed deals was guesswork.
- The no-code automation layer cost €10k/month, and only the contractor understood it: classic rented infrastructure.
- Meta / Google / LinkedInnightly ad-account syncs
- Lead webhooksrealtime + reconciliation
- Zoom recordingstranscripts via Whisper
- First-party trackingcustom t.js on landing pages
- Postgres · 183 tablesrow-level security throughout
- 25 background jobstrigger.dev queues + crons
- 229 API routesmulti-tenant, audited
- Nightly backups~200 MB → client-owned storage
- 7 AI agentsanalyse · kill/scale · challengers
- Client dashboardsads, CRM, funnels, split-tests
- AI creative studiogenerates + publishes to Meta
- AlertingSlack briefings + Sentry
What got built
Multi-tenant core
Next.js + Supabase with row-level security throughout: 183 tables, 100+ additive migrations, strict per-client isolation, role model from superadmin down to client logins. The agency’s customers get their own dashboards; the agency gets one control plane.
Ads & data plumbing
Nightly syncs from Meta, Google, and LinkedIn ad accounts, lead webhooks with reconciliation, Conversions-API forwarding with per-client tokens, and a first-party tracking script for landing pages. 25 scheduled background jobs (trigger.dev) keep it all current.
AI layer
Creative and copy generation that publishes directly to Meta, Zoom-call transcription with AI analysis into action items, and a media-buying autopilot: seven orchestrated agents that classify every active ad, make kill/scale recommendations, and generate challenger creatives, all fenced behind budget circuit-breakers, staged rollout, and a default-off switch. 104/104 tests green at handover.
Attribution fix that paid for itself
One client account showed 1 lead against thousands in spend; the funnel looked dead. Per-account custom-conversion mapping recovered the real number: 355 leads at ~€15 cost per lead. Diagnosed and live in production the next day.
Run, not just shipped
This engagement is the operator model in action: nightly database backups to client-owned storage, EU-region error monitoring, a full security audit of every API route (11 access-control gaps found and closed), and incident response, including diagnosing and restoring a live database-cluster outage. Started as a small paid test project; grew into a five-figure monthly retainer on the strength of the first phase.
Stack
- Next.js
- Supabase (RLS, RPCs, materialized views)
- trigger.dev
- Meta Marketing API + CAPI
- Google Ads API
- LinkedIn API
- Anthropic + OpenAI
- Whisper
- Sentry
- Vercel
Got a system like this waiting to exist?
Start with the free teardown: I’ll map your three biggest automation leaks and what they cost you monthly. In your inbox within 72 hours.
Get your free systems teardown