I Love Ads

August 25, 2026

How Was Your 2025?

I graduated with a B.S. in Mathematics from Caltech in June 2025 and started working as Machine Learning Engineer (MLE) in October of 2025. Prior to this job, I had never worked in industry in any capacity. My entire collegiate career was dedicated to research in commutative algebra, number theory, and combinatorics. Also of note is that Caltech has a strong scientific research culture. In other words, I was raised to be a scientist while training as a mathematician.

One may be not surpised that I found work as an MLE. I am specifically an MLE at an advertising technology (adtech) company. The advertising qualifier is far more important than the machine learning qualifer.

There is a small group of experts in machine learning for advertising. Those experts power some of the highest revenue companies in the world such as TikTok, Meta, and Google. I say this to emphasize that advertising affects us all. As an example, advertising accounts for more than 97% of Meta’s revenue. As such, I find it important to document what I have learned and what I think should be accesible to others. In particular, those who are far removed from the industry of advertising but would like to know more from the inside.

Some History

I am going to spin a narrative thread throughout this history.

^ The above will show the history of computing, then to distributed computing over computing networks. Then the internet and web leading to big data.

Communication across spacetime is absolutely essential to human progress. The American Telephone and Telegraph Company (AT&T) is a prime example of capitalizing on this demand. AT&T grew to be the largest American company to ever exist. It was so essential that it was forced to work with the government. It led to the creation of Bell Labs which in turn developed much of our modern technology and knowledge.

It is only fitting that AT&T is credited with the first clickable digital banner ad on October 27, 1994 on hotwired.com.

What Is Advertisement?

Advertisement is the the act or process of making something known to someone (cite Merriam-Webster).

Advertising Technology (Adtech) is the practical application of scientific knowledge to advertisement. This includes the creation and development of machines for advertisement. Personal computers, mobile phones, and connected TVs are the primary devices which host the channels to deliver digital advertisements to users. These channels include the open web, social media, connected TV, and mobile games.

Some Fermi Estimates

Suppose that everyone spends one hour a day looking at a screen. In that one hour suppose that an advertisement is visible on screen once every minute. This will lead to 60 digital advertisements visible a day. Now, suppose that one billion people are using such screens so we arrive at 60 billion digital advertisents shown daily. Dividing out again that this happens in one hour this shows a maximum delivery rate of around 17 million advertisements per second.

TODO: { estimate number of display/video/native ads shown per second on digital devices. Breakout by mobile, pc, and CTV. Emphasize that ads are served and this service requires compuation for delivery as well as for many other parts. After sufficiently making this point we can pivot to advertising being a planet scale engineering problem.

Break down the unit economics of delivery an ad. cost of delivey + cost of prediction + cost of measurement.

Open web, CTV, walled gardens, native.

Some disucssion on CPM, CPA, etc.

some disucssion on the supply-side, and demand-side speak. publishers and advertisers and marketers and then who they interact with. The Lifecyle of an Ad Opportunity }

After TODO block one is complete we can begin some more interesting discussions since we will have a domain model for the ad economy. We then can talk more aboout the computational aspects of advertising and where the current pain points are. That will illuminate areas for innovation or opportunity.