In this episode of the Paid Search NYC Podcast, host Matt Shenton is joined by Edwin Choi, Founder of Jet Fuel Agency, to break down one of the biggest challenges in paid social today: creative fatigue - and how Meta’s evolving algorithm (including Andromeda) is changing the rules.

We go deep into how high-performing teams are approaching creative at scale — from forecasting creative demand and measuring fatigue, to building systems that reduce manual work and unlock better decision-making. Edwin also shares how AI is reshaping research, analysis, and execution — and what skills actually matter in modern performance marketing teams.

🔑 Key Insights from the Episode

Why “execution is table stakes” in the age of AI

How top teams forecast creative demand based on budget, decay rates, and win rates

What creative fatigue actually is — and how Meta measures and penalizes it

The concept of “ad concentration risk” and why it can quietly kill performance

How to measure fatigue using frequency, CTR decay, and performance trends

Why creative diversity (not just volume) is critical under Meta’s Andromeda update

How Meta now pre-processes and categorizes ads before they even enter auction

The shift from audience targeting → creative-led targetingHow to structure ads around personas, hooks, and emotional drivers

What “mental agency” is — and why it’s the most important hiring trait today

⏱ Key Sections & Timestamps

00:00 – Intro

00:20 – Edwin’s background & building Jet Fuel Agency

02:30 – Lessons from hiring (and firing) agencies

05:00 – Why execution is now table stakes

06:00 – The 3 AI buckets: research, analysis, execution

11:00 – Hiring for “mental agency”

15:30 – What is creative fatigue?

18:50 – How to measure fatigue (frequency, decay, trends)

21:30 – Forecasting creative demand from data

24:40 – Benchmarking and improving mature accounts

26:50 – Creative-led targeting & personas

31:00 – Meta’s Andromeda update explained

33:30 – Why ads fail before they even launch

34:40 – Ensuring creative diversity at scale

38:00 – Final takeaways