Mellow · Growth EngineerBarcelona · CAPI · CRM · Automation

Ads are the last mile.

I’m Joan Sau. Most ad accounts aren’t broken at the ad level. They break three steps earlier: a pixel dropping events it never reports, a CRM that never tells Meta who actually paid, leads landing in a sheet nobody opens. That layer is my job. Server-side tracking, CRM integrations, conversions fed back to the algorithm. The media buying sits on top of it.

signal_monitor.exe
Dithered close-up of an orange eye REC
01 · signal.wav · the actual problem[2:14]

The algorithm only knows what you tell it.

Meta’s delivery system decides who sees your ads based on the conversions you send back. Send it form submissions and it finds people who fill out forms. Some of them buy. Most don’t pick up the phone.

Better creative doesn’t fix that. Neither does a new campaign structure. The fix is upstream: getting the right event, with the right data attached, from your CRM back to Meta. Every time. Without someone exporting a CSV on Fridays.

With Andromeda matching ads to people at a much larger scale, this matters more, not less. More reach driven by a bad signal just burns money faster.

02 · stack.wav · the stack, in order[3:07]

Three layers. Listed by how much they matter.

Media buying is third on purpose. It only works as well as the two layers under it.

[FORM] ──▶ [GHL CRM] ──▶ [WEBHOOK] ──▶ [n8n] ──▶ [CAPI] ──▶ [META]
               ▲                                           │
               └────── stage change: qualified / booked / won ──┘
01

Data infrastructure

Where the signal gets built. If this layer leaks, everything above it optimizes toward noise.

  • Meta Conversions API
  • Pixel + CAPI dedup
  • GoHighLevel webhooks & REST
  • Offline conversions
  • fbc / fbp persistence
  • Event match quality
  • Hashing & PII
02

Automation

The glue. Lead routing, enrichment, alerts, retries.

  • n8n
  • Python
  • Go
  • REST APIs
  • Webhooks
  • SQL
  • Scheduled jobs
  • Error handling
03

Execution

Where the system meets real money. I run it myself, so I know exactly what the data layer needs to deliver.

  • Meta Ads
  • Account structure
  • Creative testing
  • Budget allocation
  • CRM revenue reporting
03 · cases.wav · every one started as a tracking problem[4:41]

Problem first.

Then what I built, then what happened. In that order.

CASE 01 · Weblaze, review funnel SaaS for local businesses

Local businesses were losing customers to competitors who just had more Google reviews.

SIGNAL PROBLEM
Google’s local ranking algorithm rewards review volume and recency more than almost anything else a small business controls. Most owners knew this and did nothing, because asking every customer for a review by hand doesn’t scale.
WHAT I BUILT
A GoHighLevel-based SaaS backend running a funnel. After a customer interaction, an automated sequence asked for feedback, routed happy customers straight to a Google Maps review link, and caught unhappy ones privately before they went public. Sold it myself through cold calls.
WHAT CHANGED
Review volume compounded and local ranking improved. More inbound leads without spending more on ads.
ghl     trigger=interaction_complete
funnel  route: happy → review_link | unhappy → private_feedback
maps    new_review  ranking signal sent
CASE 02 · Native Shopify store

Ad spend was scaling. Profitability wasn’t keeping up.

SIGNAL PROBLEM
The pixel was firing, but purchase events weren’t reliable enough to trust past a certain spend level. Meta was optimizing on partial data.
WHAT I BUILT
Set up Meta CAPI alongside the pixel, deduplicated with shared event IDs, so every purchase reached Meta twice, once from the browser, once from the server.
WHAT CHANGED
Scaled spend to €1,500 in a single week and stayed profitable through it.
pixel   event=Purchase  client_side  sent
capi    event=Purchase  server_side  event_id=match  sent
meta    dedup ✓  signal complete
CASE 03 · Partner e-commerce store, €150k/month

Creative fatigue was quietly eating the account’s efficiency.

SIGNAL PROBLEM
Ads that worked for two weeks stopped working, not because the audience or offer changed, but because the same hooks had been seen too many times. Nobody was tracking that decay.
WHAT I BUILT
Owned the full creative pipeline: scripting, filming and editing direct-response video ads in DaVinci Resolve, iterating hooks based on the account’s actual unit economics.
WHAT CHANGED
Kept acquisition cost stable by treating creative as a rotating input, not a one-time asset.
creative  hook_v7  ctr declining  flagged
creative  hook_v8  shipped  testing
account   cpa stable  spend maintained
04 / OPEN
CAPI + CRM rebuild
Reserved for an account where the pixel and the CRM disagree about what a sale is.
AWAITING BRIEF
05 / OPEN
Offline conversion loop
Reserved for a CRM that knows exactly who bought and has never told Meta.
AWAITING BRIEF
06 / OPEN
Lead routing automation
Reserved for leads that sit for hours before anyone calls them.
AWAITING BRIEF
Three slots open for the next build.

If your tracking leaks or your CRM and ad account don’t talk to each other, send me the setup and I’ll scope it.

Send me your setup ↗
04 · proof.wav · the execution layer[1:58]

And yes, I run the ads.

This is where the infrastructure gets tested with real money. E-commerce accounts up to €150k a month, plus my own lead-gen and SaaS projects. They matter, but they’re a result, not the pitch.

Creative strategist

Ran the creative pipeline for a €150K/mo store.

CAPI-driven scale

Scaled self-funded ad spend to €6K/month and stayed profitable, right after building the CAPI setup and event dedup that fed the algorithm real purchase data.

Verticals

E-commerce (DTC) and my own lead-gen / SaaS projects.

Built & sold

Built a CRM system on GoHighLevel as the backend, then sold it myself through cold calls.

05 · about.wav[2:33]

I came to ads from code.

I trained as a developer. My first real work was data automation, moving processes out of spreadsheets, writing the scripts and pipelines that kept information flowing between systems.

Paid acquisition came later. What pulled me in wasn’t the ads. It was realizing that most accounts were being run blind, with budgets managed on data that was wrong, late, or missing. That’s an engineering problem. I already knew how to solve it.

So that’s the setup now. I build the data layer first. Then I run media on top of it, because nobody tests a tracking setup harder than someone spending money through it.

The pure engineering work lives at /dev.

06 · rules.wav[0:59]

Three rules I don’t break.

Signal before spend.

If tracking is broken, I fix it before scaling anything. Scaling a bad signal just gets you bad results faster.

Everything logged.

Every webhook, every event sent, every failure. If something breaks at 3am, there’s a record of what happened.

You own it.

Your accounts, your n8n instance, your CRM. Documented. No black box you have to keep paying me to open.

07 · contact.wav[∞]

Your ads are probably fine. Your data probably isn’t.

Send me what you’re running: ad account, CRM, how leads get from one to the other. I’ll tell you where the signal breaks. Sometimes it’s a five-minute fix. Sometimes it’s the whole pipeline.