The Attribution Industrial Complex and Its Expensive Fiction

The Genius/Idiot Paradox

I was working at boutique Youtube Ad Agency at the time, I believe it was 2016-17 or so.

We got new client. It was Life Coaching program, with strong leader and face of the brand. They were selling book front-end, with different programs on backend like coaching, certification, on site.

They were doing fine on Facebook but wanted to test YouTube. I think it was February or March of 2017, and Google just released Target CPA bidding.

Never tried this bidding before but wanted to. Created couple of campaigns, and in a day or two we were at 5k in adspend. In a week we were at 10k a day.

Prior to that highest daily spend in that company we achieved was for national client, in range of 4k day.

Back to life coaching, I was on top of the world. Got a raise, client thought I was a genius, I also kind of started thinking I was smart.

Within 3 months the campaign died.

The VSL that printed money stopped working. Client wanted me to 'fix it' - as if I controlled the algorithm.

I tried absolutely everything as a media buyer, I literally mean every setting known... and nothing.

This brings me to my central issue with data as we see it today: if I was a genius when we scaled, I must be an idiot when the campaign died?

Who Am I and Why This Matters

My name is Brat, and most of you have no idea who I am. Not that it really matters.

Together with Ian I founded Inceptly and Vidtao 8 years ago. My job was always in the background building teams and delivering value to clients.

I started as media buyer, then strategy, funnels, optimizations... and recently I completely switched to writing code. Through Inceptly years I think I trained probably 20-30 media buyers.

Speaking at different conferences and doing some courses, I believe I influenced many others.

Recently I started working on building conversational analytics platform called Bratrax. It's been joy of my life, discovering how to move, orchestrate and play with data.

That said, although I love to play with data, move, orchestrate, find value, help business... I believe that most of the buzz with data is 100% bullshit.

Not that I believe data is bullshit, quite opposite... but our comprehension, narratives and out of all arrogance, which always seem to be anchored to some data.

After millions of adspend managed, building data systems for companies with revenue over 100m... my firm belief is that 99% of what we think about data is flat out wrong.

You wonder why?

The Framework That Changed Everything

This Target CPA story bothered me to my bone.

Think about it for a moment, I bet all of you felt this roller coaster of emotion. Everything explodes, you want to call everyone. Tell them you are crushing it... then in matter of days everything just collapses.

Along the way we founded our companies, this became even bigger toll for me personally as you really can't position yourself from the value standpoint.

Peer pressure is staggering in marketing community. You have to have offer, your offer has to explain benefits blah blah blah...

Essentially what marketing industry tells you is: Your job is to claim credit for success while blaming market, everyone and anyone for failure.

I personally could not settle for that.

Then in 2021 I discovered Nassim Nicholas Taleb, his books and his Twitter profile.

This is first I heard about Extremistan and Mediocristan, and it was eye opening.

Reading Taleb and Mandelbrot fundamentally changed how I see everything. Once you discover Mediocristan and Extremistan, you can't unsee it. The D2C world operates in both domains simultaneously, but we've catastrophically confused which is which.

It provided new framework on how you see prediction, and our job in media was pretty much always just that. Prediction.

Let me explain concepts to best of my understanding.

Mediocristan vs Extremistan in DR

Mediocristan

No single data point can change the total in a big way. Variation is limited and predictable. Results accumulate steadily, step by step. The normal distribution (bell curve) applies. Averages are reliable, standard deviation is meaningful. The more data you collect, the better your predictions get. Being very wrong on one case does not break the whole analysis.

If you translate that into our industry, I can give you an idea of metric that operates in Mediocristan world.

Let's say we have physical durable product. Your durable good costs $150. Your AOV clusters around $150.

The Math: 1,000 orders at $150 average = $150,000 total. Add one extreme $1,500 order (10x normal). New average: $151.35

The outlier moved the average by less than 1%. That's Mediocristan.

Why It's Bounded: Nobody buys 100 durables. Physical reality constrains the range:

- Minimum: ~$150 (one unit)

- Maximum: ~$450 (maybe 3 units)

- Distribution: Gaussian bell curve

What This Means:

- Standard deviation works (±$30 captures most orders)

- More data makes predictions better

- 10% optimization is possible and real

- 10x days are mathematically impossible

The constraint IS the characteristic. You can optimize AOV from $150 to $165. You cannot get it to $1,500.

That's Mediocristan: bounded, predictable, optimizable within limits.

Extremistan

A single data point can outweigh everything else. Outcomes are unbounded and variation has no ceiling. Outliers dominate, they are the result. Growth happens through compounding or sudden jumps, not steady addition. Power-law distributions rule, not bell curves. Standard deviation and averages are useless. Collecting more data does not guarantee better predictions, sometimes it worsens them. Being slightly wrong on one case can collapse the entire model.

Remember my Target CPA story? That was pure Extremistan.

You try Google's new Target CPA. Seven days later you're at $10k/day spend, from 0! And profitable.

That's Extremistan.

Why Information Asymmetries Are Always Extremistan (Until They're Not)

Every new feature release creates temporary information asymmetry:

- We don't know what the algorithm will do

- Most people won't try it immediately

- Those who try it first might capture all the advantage

The moment everyone discovers it, it's over. The arbitrage closes. Extremistan becomes Mediocristan.

99% of Google's releases: Nothing happens. Smart Bidding, Enhanced CPC, whatever - no meaningful edge.

1% of releases: Target CPA in 2017. Performance Max in early days. The algorithm accidentally gives away everything.

You can't predict which release will be the 1%. You just try them all. Small downside (wasted testing), massive upside (10k/day scale).

We don't know HOW algorithms work. We know there's an OPTION they could work. We test because the downside is limited, upside is unlimited (temporarily).

When it works, it's not skill. It's being there and testing. When it dies, it's not failure. It's the asymmetry closing.

Every platform feature, every new bidding type, every algorithm update - they're all lottery tickets in Extremistan until the market discovers them and they collapse back to Mediocristan.

Why The Industry Is Desperately Selling Predictability

These days marketing agencies and media buyers face insane pressure from clients to deliver more than ever in the world of AI.

The difference between the best media buyer and average media buyer today would be maybe 5-25% in end results. Companies understand this, even if they don't know the concept - they understand when a role operates within Mediocristan.

Why is everyone compressed into this narrow band? Platform automation ate our leverage.

Before, an amazing media buyer would do extensive keyword research others missed. Come up with campaign structures that had temporary advantage. With AI running everything, that's gone. Everyone has the same tools. We're all equally smart or equally stupid, depending on whether the algorithm likes us today.

This reality created massive pressure. Clients bringing roles in-house, expecting more for the same, changing deal structures. Agencies and media professionals desperately needed new positioning, new perceived value. Because managing media, copywriting, soon even ads - none of it is what it was 10 years ago.

So how did we get here?

COVID was probably peak of career for most in D2C and performance marketing world. Online exploded. Money was free. Everything you touched turned to gold.

Then came the perfect storm.

Lockdowns ended, people went back outside. iOS 14.5 slowed down adnetwork models. Interest rates spiked after years of stimulus. Competition flooded every niche. It was like going from the most beautiful weather to hurricane season overnight.

Agencies were desperate for answers. The conversation shifted from "push into ToF" to "predictable growth." Attribution modeling, MMM, incrementality testing - anything to promise certainty in chaos.

This became the new angle: agencies will predict your revenue. They have models. They know how you'll grow. They can control the uncontrollable.

New tooling emerged, mostly funded by the same thought leaders who shaped this narrative. The industry collectively decided that if we just measure better, we can return to the golden days.

The Dangerous Delusion

I find this to be insanely dangerous. And before I continue, I first have to admit that I was guilty of this.

As we as an agency struggled in 2022-2023 as COVID was winding down, I got into MMM, incrementality, predictable growth.

It's seductive, logical, currently very profitable... but essentially doesn't make any sense.

We are being sold Mediocristan Solutions for Extremistan Problems:

- MMM assumes linear relationships (spend × coefficient = revenue)

- Incrementality assumes stable baselines (test vs control)

- Attribution assumes causation is knowable (touchpoint = conversion)

But Reality Is:

- One creative returns 700x, ninety-nine return 3x

- Baselines shift overnight (iOS 14.5, COVID, regime changes)

- Causation is unknowable

Every tool now sells the same story:

- Triple Whale: "Know your true ROAS"

- Northbeam: "Marketing attribution that works"

- Hyros: "Most accurate tracking on earth"

The new positioning - predictability through better measurement - is the industry desperately trying to force Extremistan into a Mediocristan box.

But Extremistan doesn't care about your tracking accuracy. It doesn't care about your attribution model. It doesn't care about your incrementality tests.

That angle that prints money appears, or it doesn't. Target CPA works, or it doesn't. Your ad finds product-market fit, or it doesn't.

And no amount of "accurate tracking" will change that.

Why We Buy The Delusion Anyway

Tim Ferriss said something like 'People will trade unhappiness for certainty.'

I've worked with so many clients who after failed tests demanded: 'Ok this didn’t work but tell us why? And tell us what next we will do and why it will work?' Like I could know that?

That drives you to two choices: Say 'I don't know' and get fired, or perform the mental gymnastics I'm arguing against here.

Enter Triple Whale, Northbeam, Hyros….. They promise to predict what will work. To finally show you the "true ROAS." To tell you which channel drives growth.

People buy them believing they're buying prediction. Control. The ability to know what happens next.

But you can't predict Extremistan with Mediocristan math.

"On average increases ROI by 15%" is meaningless when:

• Your best creative is 10x average (driven by forces you don't control)

• Your product portfolio follows 90/10 power laws

• Your platform ROAS can collapse 200% in two years

15% improvement in Mediocristan doesn't matter when Extremistan determines your fate.

The Uncomfortable Resolution

I wasn't a genius when Target CPA scaled. I wasn't an idiot when it died. I was lucky enough to find temporary information asymmetry, and unlucky enough to watch it evaporate.

I'm not saying throw away attribution. I'm saying understand which game you're playing.

AOV? Sure, optimize it 5%. But understand it might improve for 30 days then revert to baseline. That's Mediocristan - bounded, worth optimizing within limits.

But your next winning creative? Your next Target CPA moment? That's Extremistan. No amount of tracking will predict when lightning strikes.