I make the ads. And the machine that makes the ads.

AI Creative Strategist, Producer & Performance Creative

Direct response for Meta, from angle to finished asset. Medical aesthetics, US home services, fintech. No handoff, no brief thrown over a wall.

9.8%
CTR — BlueMagic
7.0%
CTR — Apex Ops
35
AI videos shipped
5M+
Views built from zero

Client-reported, specific campaigns. Full data on request.

01 · Meta · United States

Apex Operations

A San Jose agency selling lead generation to contractors across 28 states, on a guarantee: 150 qualified leads in 90 days, or they work free.

Shot 4K vertical, in the vocabulary the trade actually uses.

7.0%click-through rate
Outcome stack

“More homeowner inquiries, more appointments, more jobs closed.”

Third cut of the proven script. The variable is delivery and setting, never the angle.

Offer first

“That isn't a marketing strategy. That's hope.”

Sharpest line in the set. Names the behaviour, then takes it away.

On site

“Roofer, ADU builder, renovator, general contractor.”

ADU is a California permit class. A contractor hears it and knows someone did the homework.

02 · Meta · UK and EU

BlueMagic Group

An Istanbul hair-transplant clinic selling a home protocol into one of the most expensive lead-gen categories on Meta.

Four slots. I spent them on four different ways in, not four cuts of the same one.

9.8%click-through rate
Dating stakes

“I matched with her three days ago. I still haven't replied. Because of my hair.”

A live, unresolved situation and its cause, inside three seconds. Best hook on the account.

Social avoidance

“I used to avoid every photo, every mirror, every event.”

Inventory of small refusals, then a month one, three, six arc.

Too late

“I was losing the one thing I couldn't buy back.”

For the buyer who already tried everything and stopped believing.

Result first

“Six months ago I thought I was going bald. This is my hair now.”

Payoff up front. Also the weakest of the four, and the teardown below says why.

03 · Organic · app installs

Echo

A daily posting schedule for a live app, fronted by a presenter who does not exist.

By hand, one video costs a day. Thirty five of them is not a content problem. It is a manufacturing one.

35shipped end to end
Synthetic presenter

“They did not pull away because something went wrong.”

Voice, face, script and cut all generated. Two minutes, one take, no human on camera.

Cold open

“You do not understand what your silence does to someone.”

Second person, present tense, no preamble. The voice profile is why it sounds like a person.

Hard truth

“They are never going to give you the closure conversation.”

Refusing the comforting answer is the whole brand position, held across all thirty five.

04 · Meta · Australia

Techy Tantrums

My own agency. I am co-founder and creative chief, and I own the production standard on everything that ships.

One presenter who does not exist, held across three shoots, three wardrobes and three locations.

1face, three shoots
Same face, harbour

“Refreshing your phone every ten minutes, hoping a lead came in.”

Sydney daylight. Same character, different setting, no drift.

Same face, on site

“Solar installer, roofer, renovator.”

Hi-vis, interior build. The Apex offer localised for Australia, not copied.

05 · Statics · 1080 × 1350

Six levers the video could not pull

A presenter can only test what a presenter can say. Every line here went through the platform policy checklist before layout, not after rejection.

BlueMagic static: clinic-built, not surgery
Price anchorAnchors against the clinic's own surgery
BlueMagic static: by the surgeons themselves
AuthorityNever addresses the viewer's condition
Apex static: 150 qualified leads in 90 days
Unit economicsThe number is the visual
Apex static: booked solid through spring, scroll on
DisqualifierLight, against a black and orange account
Echo static: four stages of no contact
Self-diagnosticThey place themselves before they tap
Echo static: day twenty-one journal entry
ArtefactNo presenter at all
Echo · the manufacturing problem

So I stopped making videos, and built the line that makes them.

The problem

Thirty five videos, a day of work in each. Hand production does not scale here, it just gets more expensive every week.

What I did first

A full content audit before a single frame. I scored 182 of the account's own reels by reach and posting hour.

The solution

One command takes a script to a scheduled post, gated by a transcript check and an approval that times out in an hour.

35delivered end to end
182reels audited first
4post steps automated. Cuts, subtitles, title card and music, with no manual effort
386credits saved on one recovered batch
Build stage of the pipeline
01 Build

Script to timeline. Every chunk capped at nine seconds, because past that the lip sync drifts.

Generate stage of the pipeline
02 Generate

Face, voice and motion generated in batches of twelve. kling3_0 and ElevenLabs.

QC stage of the pipeline
03 QCre-roll sends back to 02

Whisper transcribes the render and diffs it against the script. A changed meaning fails.

Compose stage of the pipeline
04 Compose

Cuts, subtitles, title card, music. Overlays burn in here, so restyling costs no credits.

Approve stage of the pipeline
05 Approve

One preview to Telegram. No reply inside the hour and the batch carries on without me.

Publish stage of the pipeline
06 Publish

Queued through the Meta Graph API with scheduled publish times. The laptop can be off.

51,000
words of transcript mined into a measured voice profile, so generated copy lands in the presenter's real cadence.
0.50
lip-sync correlation under nine seconds, against 0.31 over twelve. That measurement is why the chunk cap exists.
14
delivered videos re-titled for zero generation spend, because the pipeline keeps raw components by design.
182
reels bucketed by posting hour. Best window ran a median 31,834 views against 12,466 in the worst.
What I work in

The stack, and what each part is for.

Generation

  • Higgsfield, kling3_0
  • HeyGen, digital presenters
  • ElevenLabs, voice direction
  • Character held across a campaign

Automation

  • Claude Code, Codex
  • n8n, Make.com
  • Python, custom agent skills
  • Human-in-the-loop approval gates

Strategy

  • Hooks, angles, awareness levels
  • Offer and objection mapping
  • Platform policy, read at scripting
  • Batch design and creative testing

Distribution

  • Meta Ads Manager, CTR and CPL
  • Meta Graph API, scheduled publishing
  • rclone, Drive delivery
  • First-party performance analysis
For hiring managers

What you are actually buying.

Roles I fit

AI Creative Strategist, Performance Creative, AI Producer, Growth Creative. Agency or in-house.

Volume I hold

Four distinct angles a week per account, produced and shipped, not briefed out and waited on.

Risk I remove

Platform policy read at the scripting pass. Restricted verticals are where most creative teams lose weeks.

Week one

Audit the account, name the angles nobody is testing, ship a batch against the current control.

Download the one page resume