What I Learned Spending Thousands on Meta Ads to Launch a New Product

About a year ago, I was working for a global fintech headquartered in London. We had a budget of tens of millions of dollars per month to spend on marketing, so we worked with a fairly large marketing team. As director for growth, I was responsible for all sorts of technical marketing duties, including measurement, signals engineering, analytics and the measurement of performance in general.

We employed a large team of dozens of developers to do this. Now things have completely changed. I can do a lot of the things that I used to do with a team just on my own with AI. I wrote SQL code for years, but I’ve not written any SQL for over a year now. Most of my analyses are not done in dashboards, but via chatbot.

Having spent most of my career advising other people how to grow their business, I put it to work on my own product, RunInk, an app that turns your running routes into posters, and spent about $1,700 on Meta ads in the first month. Sales so far add up to around $200. I understand that there is a testing budget involved in these things. It’s not going to go immediately well. And in fact, it got more profitable once the ads had more to learn from.

There was a team of dozens of designers making different creatives, and many of them got released and tested and measured. Whereas now I can get Claude to make a ton of different ideas, and Meta’s AI can just test them.

It’s also not just a question of speed. Yes, AI’s creatives are often visually unappealing. But I’m not a designer. I can’t make nice things. I don’t know how to make nice things. But with the prompting, I can get AI to do it. I can focus on the winners and on getting lots of different variants out that I think might be useful.

Signals engineering was generally the martech job, and it was not an easy job — both for large companies like the one I worked for, which had a relative dearth of signals compared to ecommerce, and for small brands starting out. There can be a significant increase in performance by just doing this well.

Ad platforms run on machine learning, which requires lots of signals to be able to learn what type of person is interested in your product. Meta says an ad set needs about 50 optimization events a week to get out of its learning phase. However, if you’re starting out from zero, you have no one interested in your product. This is also the case if you sell few, but high-ticket items. You have relatively few people coming by, but you need to figure out who they are and somehow advertise something to them.

So what you start doing is breaking down the funnel into smaller steps. Do they land on my website? Do they create an account? In my case, do they connect their Strava account? Do they create an image? Do they export an image? Do they export various images?

Then you try to add some monetary value to all of these steps and break it down from there. And then you can slowly tune and adjust these values so that you stop paying for page views and for signups and eventually start buying purchases.

Question your AI like you would an intern

AI is a powerful tool, but it still makes mistakes. You need to double check AI, ask questions and validate everything. Not necessarily by manually redoing the numbers yourself, but by asking probing questions the way you would ask an intern.

Back when I managed a team, the way I would ensure numbers were right (or at least vaguely right) was by probing them. I’d tell my team: “Hey, that number seems high; are you sure about it?” or “That number seems off; can you double-check?”

Also, don’t forget that your AI agent often lacks the context you do. If you know that five of the last 10 purchases made on your site were you testing a new feature, tell it!

Use advertising to get your first 100 clients

It is hard to build a product if you have no users. Once I had the first 100 or so users on the website, I surveyed the people who made posters.

I saw that there was some sort of demand, some happiness with the product itself. But 11 of the 16 who answered said the free digital file was enough for them. And the only way I made money was via printed posters!

This told me that I should offer a digital product. And it instantly became my best seller! Business success is about finding the right price point and the right product, and to do that, you need to talk to your customers.

Budget for more than $1,000

If you’re starting out with a brand new product, be aware that it’s going to cost you a lot more than $1,000. To get this to work, you’re most likely going to have to spend more than that.

That being said, it’s a great way to get people to your product. If you want to find out more, I write about measurement and AI on my site.

About a year ago, I was working for a global fintech headquartered in London. We had a budget of tens of millions of dollars per month to spend on marketing, so we worked with a fairly large marketing team. As director for growth, I was responsible for all sorts of technical marketing duties, including measurement, signals engineering, analytics and the measurement of performance in general.

We employed a large team of dozens of developers to do this. Now things have completely changed. I can do a lot of the things that I used to do with a team just on my own with AI. I wrote SQL code for years, but I’ve not written any SQL for over a year now. Most of my analyses are not done in dashboards, but via chatbot.

Having spent most of my career advising other people how to grow their business, I put it to work on my own product, RunInk, an app that turns your running routes into posters, and spent about $1,700 on Meta ads in the first month. Sales so far add up to around $200. I understand that there is a testing budget involved in these things. It’s not going to go immediately well. And in fact, it got more profitable once the ads had more to learn from.