Case study · Paid client work

Agency Command Center: Meta Ads + GoHighLevel

A Meta ads agency runs about 50 home-service contractors. Every morning the system reads each client's ad spend, billing status, leads and speed to lead from Meta and GoHighLevel, and the app puts the whole book on one screen. When a number can't be read, it says so instead of showing a zero.

The command center's daily triage screen on sample data: flagged clients, money stuck on ad accounts, and figures that could not be read Sample data
RoleSolo build, audit to app
ClientMeta ads agency
StackNext.js + Supabase + n8n
StatusLive, agency-owned stack

Reports that kept running after the data stopped

The agency's owner asked for a genuinely good dashboard, not a repair job. The team checks client health between calls, and every client gets a performance report each Friday.

The audit found the numbers underneath had stopped. The reporting pipeline's stats feed had been dead for about six and a half months while every run still reported success, because each step swallowed its own errors. It covered roughly 36 of the agency's 108 ad accounts, and the dashboard the team already had was failing most of its CRM reads.

Measured in the audit

6.5 months Stats feed silently dead

About six and a half months, and every scheduled run in that window still reported success.

31 of 38 CRM reads failing

On the dashboard the team was already using.

0 of 206 Report rows with leads

Leads live in GoHighLevel, and the old pipeline never read it.

Failures louder than numbers

The owner wanted nothing that looked like a generic SaaS template. I added one rule of my own: the app states facts and never grades a client by itself. Colour marks facts such as billing trouble, zero delivery, a failed read or a stale sync. A judgement like a cost per lead above target appears only when the owner's written rulebook sets that target. The owner's sign-off is recorded next to the rulebook; it never holds an alert back.

Figures are labelled with their source and whether they were measured or blended. A failed read shows as READ FAILED, never as a zero, and a partial read turns totals into lower bounds. Nothing writes to Meta or GoHighLevel. The system was built in blocks the owner approved one at a time, running alongside the agency's existing automations instead of replacing them.

From Meta and GoHighLevel to one screen, every morning

  1. SourcesMeta Ads + GoHighLevelRead-only access
  2. Schedulen8n syncsScheduled every morning
  3. GateSigned ingestChecks the signature before writing
  4. StorePostgresDaily rows plus a run log
  5. SurfacesCommand center + ClickUpOne screen, alerts in ClickUp and the app’s chat
  1. 01

    Audit before building

    I inventoried all 108 ad accounts and 73 GoHighLevel sub-accounts first. The audit flagged 12 ad accounts with billing problems, 7 whose timezone put their daily numbers out of step with the rest, and about 15 clients that seemed to have a second ad account splitting their history. It also showed ad-level attribution was possible: 82% of Meta leads in GoHighLevel already carried the ID of the ad behind them.

    Meta Marketing API · GoHighLevel API · read-only
  2. 02

    Sync every morning

    Scheduled n8n workflows on the agency's own instance read spend, leads, billing status, pipeline stages and speed to lead each morning. Two of the heaviest GoHighLevel reads first hit n8n Cloud's 60-second limit on code steps, so each now runs in a child workflow, one sub-account at a time.

    n8n · daily, 6 to 8 am Central
  3. 03

    Signed writes, logged runs

    Each sync sends its rows as one signed envelope to an edge function. The function checks the signature, the timestamp and the schema version before any metric row is written. A Meta or GoHighLevel sync that crashes still files a failure row, so a missing day shows up instead of passing silently.

    Supabase Edge Functions · HMAC · Postgres
  4. 04

    Screens built around the daily check

    Overview triage, Clients, Meta Ads, Monitoring with incidents, rule previews and data freshness, weekly reporting with per-client scorecards, plus Team and Data. It works on a phone too, because the team checks between calls.

    Next.js · TypeScript · Tailwind · Netlify
  5. 05

    Alerts in each client's channel

    The alert rail posts each alert in the client's own ClickUp channel and the app's team chat, tags the client's CSM, follows up after 48 hours and posts a cleared note once the problem is gone. Before go-live, a test run posted 6 alerts into one channel without tagging anyone, and a rerun added no duplicates.

    ClickUp + app chat · live

The automations behind the screens

Twenty-four n8n workflows run on the agency's own n8n Cloud, plus the GoHighLevel workflows around them. The map shows every one of them, moving: what starts it, what it reads, where it writes. Names in the notes are anonymized.

24 workflows on one map. Five lanes plus the error rail. Bold workflows are the ten featured; open any of them to see what starts it, what it reads and where it writes. +N more opens the rest. Static version, full size →

Below, five of the canvases as they run today. Use the arrows or swipe, and tap or click any canvas to see it full screen.

n8n canvas of the Meta daily sync: a schedule trigger, per-account Meta reads, merges and a signed POST to the ingest function
Meta daily sync. Every morning it reads each ad account on the roster, routes it to the Meta key that can read it, and records a failed account as a named read failure instead of dropping it.
n8n canvas of the GoHighLevel daily sync with token minting, contact and opportunity reads, a loop over sub-accounts and a signed POST
GoHighLevel daily sync. Mints a token for each sub-account, counts new contacts and won opportunities, and hands the pipeline-stage walk to a child workflow one sub-account at a time.
n8n canvas of the lead-quality child workflow paging GoHighLevel conversations and opportunities for one sub-account
Lead quality, one sub-account. The child workflow the lead-quality sweep runs once per sub-account. It pages through conversations and opportunities, so no single run hits n8n Cloud's 60-second limit on code steps.
n8n canvas of the alert delivery workflow: hourly trigger, alert and ledger reads, planned ClickUp posts and a ledger update
Alert delivery. Runs hourly from 7 am to 7 pm Central. It reads the alerts and the delivery ledger, plans the ClickUp posts, and records every post result, so a rerun adds no duplicates.
n8n canvas of the error workflow: on sync failure it builds an alert, signs a run_failed envelope and posts it to the ingest function
Error rail. Runs whenever a sync workflow crashes. It files a failure row through the same signed path and sends a Telegram alert, so a missing day never passes silently.

Built for the daily check

The app as it runs today, on its built-in sample data: invented contractors and numbers, the agency's branding removed. Swipe or use the arrows; tap or click any screen for full size.

Command center home on sample data: recent clients, team stats, quick actions and the latest GoHighLevel and Meta snapshots
Home. Recent clients, report weeks, Meta and GoHighLevel read status, and quick actions in one place.
Daily triage on sample data: 15 of 35 clients flagged, lead, spend and booking tiles, leads by client and a needs-a-hand list
Daily triage. Which clients carry a flag this window and why: payment failures, no delivery, cost per lead rising. Facts, not grades.
Active clients table on sample data with ads manager, CSM, appointment setter, data sources and speed to lead per client
Clients. Every client with its ads manager, CSM, setter, data sources and speed to lead. A read that failed says so instead of showing zero.
Meta Ads screen on sample data: a warning that one ad account could not be read above spend, impressions, clicks, leads and cost per lead
Meta Ads. Spend, impressions, clicks, leads and cost per lead for the window. When an account can't be read, totals become lower bounds and Meta's error is quoted.
Client report list on sample data showing each client's contact, rendered image state and send status
Reports. Each client's weekly report with its image state and send status, so nothing goes out half-read.
A sample contractor's weekly Facebook Ads report with the frozen week, print and export buttons, and a glossary of each metric
Client report. A contractor's weekly Facebook Ads report on a frozen Monday-to-Sunday week, every metric defined, Meta and GoHighLevel never blended.
Team performance on sample data: live clients, payment failures, speed to lead, ads manager and client success panels
Team performance. Ads managers, client success and appointment setters on one page: payment failures, speed to lead and who was contacted this week.
Task board on sample data with tasks per client, owner, status, progress and due date
Task board. Alerts become tasks with an owner, status, progress and due date, so a failed card has a person on it the same day.
Scorecard on sample data: one row per client with flags, leads, Meta ad leads, new CRM contacts and cost per lead, booking and appointment
Scorecard. Every client in one table: leads, Meta ad leads, new CRM contacts, cost per lead, per booking and per appointment, with each flag named.
Client launch pipeline on sample data: active, blocked, overdue and launching counts above a board of onboarding stages
Pipelines. Each new client's launch from sales handoff to campaign build, with what is blocked, overdue or launching this week.

Numbers the owner can trust at a glance

64 of 64 Meta ad accounts read clean

On the first live read, with zero failures.

72 of 73 GoHighLevel sub-accounts read clean

On the first full sweep. The same day, the old dashboard was failing 31 of its 38 CRM reads.

Daily Unattended syncs

Meta and GoHighLevel sync each morning on the agency's own n8n, and every scheduled run is logged, failures included.

Paid ClientNext.jsTypeScriptSupabasen8nMeta Marketing APIGoHighLevel APIClickUpNetlify

Three weeks in, the app, the syncs and the database all run on accounts the agency owns, and the work continues as the owner reviews each release. Since 2026-11-12, twelve GoHighLevel Conversation AI assistants answer the leads nobody answered and hand each one to its contractor with a summary.

Available for remote work

Running Meta ads for home-service clients?

If your team checks client health by opening Ads Manager and GoHighLevel one account at a time, I can build the screen that does it every morning, on accounts you own.