Ex-Growth Director Burned $1,700 on Meta Ads for $200 in Sales

A former growth director who once oversaw tens of millions of dollars in monthly ad spend poured about $1,700 into Meta ads for his own app — and generated roughly $200 in sales so far. The experiment, run on RunInk, an app that turns running routes into posters, is his proof that AI has collapsed marketing team sizes, not the cost of finding out whether a product sells.

The operator behind the test spent most of his career advising other companies on growth. About a year ago, he worked as director for growth at a global fintech headquartered in London, where the marketing budget ran to tens of millions of dollars a month. Measurement, signals engineering and analytics for that operation required a team of dozens of developers.

"Now things have completely changed," he writes in a first-person account published by Entrepreneur. "I can do a lot of the things that I used to do with a team just on my own with AI." He wrote SQL code for years but hasn't written any in over a year; most of his analyses now run through a chatbot rather than dashboards.

What can AI actually replace?

The creative process is the clearest case. At the fintech, dozens of designers produced ad variants that were released, tested and measured. Now he gets Anthropic's Claude to generate a large volume of ideas and lets Meta's AI test them. AI's output is often visually unappealing, he concedes — but he's not a designer. Prompting lets him ship many variants and focus on the winners.

The economics carry a caveat. He understands a testing budget comes with the territory. "It's not going to go immediately well," he writes. Performance improved once the ads had more data to learn from. His advice for anyone starting from zero: budget well over $1,000, because a brand-new product will likely cost more than that to validate.

How do you feed Meta's learning machine?

Signals engineering — once a specialized martech role — remains hard for both large companies with sparse signals and small brands starting out. Meta says an ad set needs about 50 optimization events a week to exit its learning phase. A new product with zero customers can't supply that volume, and neither can sellers of few, high-ticket items.

His workaround: break the funnel into smaller steps. Did the visitor land on the website? Create an account? Connect a Strava account? Create an image? Export one image, or several? Assign each step a monetary value, then tune those values over time so you stop paying for page views and signups and start buying actual purchases.

Should you trust AI's numbers?

Only the way you'd trust an intern. When he managed a team, he validated figures by probing them: "Hey, that number seems high; are you sure about it?" or "That number seems off; can you double-check?" He applies the same questioning to AI output.

Context matters too. If five of the last 10 purchases on your site were you testing a new feature, tell the AI — it doesn't know what you know.

Where did the real business lesson come from?

Not from the ads — from the customers. Once the first 100 or so users were on the site, he surveyed people who had made posters. Eleven of 16 respondents said the free digital file was enough for them. Since printed posters were his only revenue source, that answer flagged an obvious gap. He launched a digital product, and it instantly became his best seller.

The takeaway he draws: AI can now handle the work of dozens of developers and designers, and Meta's machine learning can find the winning creatives. But finding the right price point and the right product still requires talking to customers — a job no ad platform will do for you.