> ## Content Index
> Fetch the complete content index at: https://the-og-house.ghost.io/llms.txt
> Use this file to discover other available public pages before exploring further.

# I let AI make our ads for a quarter. Here’s what it couldn’t do
- URL: https://the-og-house.ghost.io/ai-made-our-ads-for-a-quarter/
- Published: 2026-10-28T20:51:55.000Z
- Updated: 2026-10-29T17:31:27.000Z
- Description: Handing creative to AI, cutting the free trial to three days, and why his first hire was an analyst, not a buyer.
- Author: Dennis Mink
- Tags: No. 02, Marcus Webb, Conversation, Creative, AI in growth, Retention and monetisation, #test

Marcus Webb has run user acquisition for subscription apps since the early days of the App Store, most recently as Head of UA at a top-20 fitness app. We covered three decisions: handing creative to AI, shortening the free trial, and who to hire first.

## 1\. Letting AI make the ads for a quarter

**The setup.** “Creative was our bottleneck. We needed forty new concepts a month to keep fatigue at bay and we were shipping twelve.”

**The call.** “For one quarter AI drafted scripts, edits and variations, and humans approved. One rule: every concept had to start from something a real subscriber had said.”

**What happened.** “Volume tripled and the cost of each concept collapsed. But the two concepts that actually moved trial starts both came from a customer interview a human sat in on. The machine was brilliant at the second idea and useless at the first.”

> “AI made us ten times faster at producing ads, and no faster at knowing what to say.”

****Steal this.** Split creative in two: let AI scale proven concepts, and keep humans finding new ones in real customer language. Measure the two separately.

## 2\. Cutting the free trial from seven days to three

**The setup.** “Most of our seven-day trials were decided by day two. The other five days mostly gave people time to forget they’d signed up.”

**The call.** “We tested three days against seven by market. Product worried about trial starts falling; I worried about paying for installs that never had a chance to convert.”

**What happened.** “Trial starts dipped a little, trial-to-paid rose a lot, and refunds fell. Net, three days won in every market but one, where seven still did better.”

****Steal this.** Look at when your trials are actually decided. If it’s day two, a seven-day trial is mostly a forgetting window.

## 3\. Hiring an analyst before another buyer

**The call.** “My first hire when the team grew wasn’t a media buyer. It was an analyst who could tell us which of our numbers to trust. Every buyer after that got better because the numbers did.”

****Steal this.** If your buyers argue about whose numbers are right, your next hire is an analyst, not a buyer.

## What he’d do differently

“Measure concepts and variations separately from day one. We celebrated volume for a month before we noticed the new ideas had dried up.”

****If I were starting today.** Speed is cheap now. Knowing what your customer actually wants to hear is not.