EP07 - AdTech Trading

How programmatic advertising works like HFT. Learn about real-time bidding (RTB), DSP platforms, attention markets, ad arbitrage, and how ads are traded in milliseconds.

Tweet coming soon

Marketing is so romantic, Kurumi.

This is art. Emotion. Creating a connection between the human soul and... carbonated sugar water.

It's about latency arbitrage and statistical regression.

Hah! Next you'll tell me that a quant is behind all of this.

That billboard is there because a Creative Director had a vision and a dream! Math nerds like you just don't get it.

That billboard? Maybe.

But the ad on your phone? That wasn't a vision. That was a **War**.

This 300x250 pixel box.

In the 200 milliseconds it took for your webpage to load, an **Auction** happened.

Trader A (Nike): "I bid $0.005 for her attention!"

Trader B (Bank): "She looked at a mortgage calculator yesterday! I bid $0.05!"

Trader C (Pizza): "It's lunch time! I bid $0.06!"

This happens 100 billion times a day.

It is **Real-Time Bidding (RTB)**.

They... traded me? Like a pork belly future?

Yes. You are the **Commodity**.

The website (Publisher) is the Exchange. The Advertiser is the Trader.

And just like HFT, humans are too slow to trade this.

We write algorithms called **DSPs (Demand Side Platforms)** to do the bidding.

So... AdTech is just FinTech for people?

It is **High-Frequency Trading** where the asset class is **Attention**.

In Finance, we predict: $P(Price_{up} | ext{Market Data})$.

In Ads, we predict: $P(Click | ext{User Data})$.

The math (Logistic Regression, Gradient Boosting) is **Identical**.

So if I can build a model to predict stock prices...

You can build a model to predict who buys sneakers.

Why do you think Google hires so many Physics PhDs? To choose the font color?

But wait. In stocks, we look for "Undervalued Assets."

What is an "Undervalued Asset" in ads?

**Arbitrage**.

Buying high-value intent in a low-value place.

This user is looking for a $1 Million divorce attorney. She is valuable.

But right now, she is reading a "Funny Cat Blog."

The "Cat Blog" sells ad space for pennies ($0.001).

The Smart Quant buys that space for $0.002.

And serves her a Divorce Attorney Ad worth $50.00 to the law firm.

Buy low (Cheap Site), Sell high (Expensive User)!

It is the classic **Spread**.

We extract value from the mispricing of the context.

This explains why the internet is so fast.

It explains why the internet is *free*. The trading floor funds the web. Your activity and doomscrolling habits pays for the whole thing.

But Kurumi... in Finance, we have "Market Makers" who provide liquidity.

Who provides liquidity here?

**Ad Networks** (Google/Meta).

Google matches the Buyer (Advertiser) and Seller (Website) instantly.

They take a cut. Just like a Stock Exchange.

So Google is basically the NYSE of eyeballs?

Yes. And you are the ticker symbol.

So, the "Quant Tech Tree" branches out?

The root is **Math**. The branches are Industries.

A Quant can work in any of these.

The variables change. The equation stays the same.

I don't need to learn Marketing. I just need to learn to predict clicks.

Marketing is just storytelling.

**AdTech** is engineering that lets the audience hear the story.

If you can master the Order Book, you can master the Ad Server.

It's all just buying low and selling high in milliseconds.

Look! It changed!

Someone else won the auction.

Is there a Sharpe Ratio for ads?

Good question...

Wait. Stocks persist. I can hold them. This… disappears in seconds.

In finance, inventory is shares. You hoard them, portfolio value grows. What's the inventory in ads?

User activity, user generated content (UGC), and advertiser's commitment.

But you're asking the right question. The half-life of attention and content is very short. You only have a few seconds to capture it and make a profit out of it.

A page visit and an impression is a perishable inventory. Like airline seats. Like hotel nights. Like electricity.

At any point of time, your attention liquidity is equal to number of users currently on the website. Your profit opportunity is to exhaust your advertiser's commitment from these liquidity ASAP.

So there's no hoarding? No equivalent of holding stock?

You hoard commitments. You hoard User Generated Content (UGC).

Advertisers pre-commit budgets. Publishers pre-commit supply. There is no inventory risk.

But you have the internet. You have tons of users and user generated content (UGC). All of them demand data storage and bandwidth. That is inventory risk.

If your advertising targeting algorithm is bad, the ad demand collapses — you're holding unsellable impressions.

If your users create trash content, the attention supply collapses — your advertisers overpay for junk traffic and you waste money hosting user's content.

This is the AdTech version of bag-holding.

Also, stocks are identical tokens. Attention and content isn't.

The first impression is discovery.

The second is reminder.

The third is noise.

Each exposure changes the asset.

This is Attention Saturation. Marginal value decays with every repeat.

So ads aren't like shares… They're more like—

Options with theta decay.

If you don't act quickly, the value expires.

If you act too often, you destroy it.

Showing the "best performing ad" repeatedly isn't optimization.

It's dilution.

That's why AdTech that simply ranks users is going to fail. It must rank moments. Every impression is a one-off auction with a decaying payoff curve.

First impressions matter.

And advertisers have spent trillions to learn that lesson.

So what's the equivalent of a portfolio?

Campaign allocation.

Each medium has different variance. Different decay curves. Different fraud exposure.

You don't want all your spend in one channel. That's concentration risk.

Same math. Different asset.

So AdTech pays for user's hosting and content, and tries to make back the cost by selling back tourists' attention to advertisers...

Exactly like how market-making quants bear inventory risks while trying to find sellers and buyers for stocks.

Latency matters as much as artistic vision. If it wasn't for quants, internet won't be free, creators won't be able to monetize their content, and advertisers won't be able to reach their target audience.

So, who wants to sponsor our next episode?

We might not even need to ask if we have enough data about our users.