The Trade Desk describes itself as a self-service platform connected to inventory and data partners. Read that sentence slowly. Connected to. Not owning. Not holding. Connected to other people's inventory and other people's data, then charging you for the connection.

Thirty years ago we had a word for that. We called them ad brokers. If you sold inventory you didn't own, didn't have an exclusive on, and couldn't take home at night, you were a broker. Somebody told me recently that I coined the term back then. I'm not sure he's right, and I'm not going to fight anyone for the trademark. But the word fit then and it fits now, and it explains more about why The Trade Desk is bleeding than any earnings call will.

I want to be precise here, because precision is the whole game. Being a broker is not a crime. It's a business model. Brokers exist because they make buying easier, cheaper, or smarter than doing it yourself. The moment they stop doing all three, the margin stops making sense. That's where The Trade Desk is now.

Let's get the concession out of the way, because I'm not writing the lazy version of this. Reach still has a purpose. Buying across thousands of publishers from one seat is genuinely useful. Bidding technology, optimization, workflow and measurement are real products that took real engineers to build.

The question was never whether the broker does something. The question is whether it does enough to be worth this many billions. And somebody, finally, did the math.

The ANA's open-web study took $123 million in spend and 35.5 billion impressions from 21 marketers and followed the money. Of every dollar that entered a DSP, 29 cents went to transaction costs. Another 35 cents disappeared into what the study counted as media-productivity losses. Thirty-six cents reached consumers in a way the ANA considered effective.

On that $123 million alone, that's roughly $35.7 million in transaction costs. Inside that, the ANA broke out 8% for DSP platform costs, 6% for DSP data costs, and another 2% for other DSP costs. Scale that across the open web and you understand why every CFO with a spreadsheet started asking the same question at the same time.

The average campaign in that study landed on 44,000 websites. The ANA argued advertisers could reach most of their targets on a few hundred. Made-for-advertising sites ate 21% of impressions and 15% of spend. Somebody brokered those impressions. Somebody charged for the privilege.

To be fair, the 2024 benchmark improved: 43.9 cents of each dollar met the ANA's effective-reach measure, up from 36. That's progress. It's also a benchmark where more than half the dollar still doesn't make it, and data and additional buy-side costs actually ticked up to 8.2%.

These are industry-wide numbers, not Trade Desk invoices. But The Trade Desk is the biggest independent seat at that table, and when the table gets audited, the biggest seat gets the most attention.

Here's where the broker model gets its premium. Not the pipes. The data. Pick an audience segment, attach it to your buy, pay extra because it sounds precise.

A study of B2B IT-decision-maker targeting found that off-the-shelf third-party segments reached the intended people no better than random prospecting. Publisher first-party demographics and contextual signals did better. Read that again. The segment you paid extra for performed like a coin flip, and the stuff the publisher already knew beat it.

Now, the honest caveat, because I don't want a Trade Desk PR person forwarding this with a smirk. That study wasn't about Trade Desk campaigns. And a randomized Meta study found that using offsite purchase data to optimize delivery lowered the median cost per incremental customer versus optimizing for clicks. Data can work. The right data, used the right way, against the right alternative.

So the claim isn't "data is useless." The claim is that a broker selling data by the segment has never had to prove, campaign by campaign, that the segment earned its fee. Precision is a feeling. Incrementality is a number. The industry has been paying for the feeling.

The one job a broker absolutely cannot fail at is being easier than the alternative. That's the entire value proposition. You pay the middleman so you don't have to do the work.

Then came Kokai. Digiday reported traders complaining that it disrupted how they worked, and that some spend moved elsewhere. And in August 2026 The Trade Desk shipped Zuma, a release whose stated purpose was to make Kokai easier to navigate, learn and use.

When your new release exists to fix your last release, you've told the market something. A broker that makes buying harder has quietly given its customers permission to ask what they're paying for.

Every broker relationship runs on one assumption: you can see what they took.

Publicis reportedly challenged aspects of The Trade Desk's fee and audit disclosures. The Trade Desk disputes that characterization and says confidentiality constrained the data that was requested. Fine. Take them at their word on the confidentiality.

But notice what that dispute is about. It isn't about whether the platform works. It's about whether the biggest buyers in the world can independently see what the platform charged. When a holding company has to fight to audit its own spend, the word "transparency" has left the building. And for a company whose entire brand was being the clean, buy-side, conflict-free option, that's not a footnote. That's the brand.

Here's the part the hot takes miss. Buyers didn't suddenly discover The Trade Desk is a middleman. They always knew. Agencies chose it with eyes open, because for a long time the math worked: the platform was easier, the reach was real, and nobody was running controlled tests on the data.

What changed isn't the knowledge. It's the tolerance. Three things arrived at once:

The ANA put a price on the middle. Twenty-nine cents is a number a CMO can say out loud in a budget meeting.

Kokai made the middle harder to use. Easier was the product. Harder is a refund request.

The audit fight made the middle harder to see. And a broker you can't verify is just a cost center with a nice logo.

Add Amazon offering a cheaper DSP with its own first-party data, the thing Jeff Green once said had no upside, and The Trade Desk guiding down $89 million, and the picture writes itself. Then came the layoffs: roughly 575 people, about 15% of staff, cut before Labor Day by a company sitting on $1.5 billion in cash and zero debt. Brokers under margin pressure cut the one cost they fully control.

I'll give The Trade Desk the test it should want to run, because if the product is as good as they say, this is how they win.

Take the same campaign. Run it with and without the paid data and platform features. Measure incremental sales, not attributed ROAS. Subtract the total fees. Publish the result.

If the data and the platform beat the fee, the broker margin is justified and I'll write that column. If they don't, buyers will keep doing what they're already doing: moving money to whoever can prove it, or whoever owns the inventory outright.

Brokers don't die because people discover they're brokers. They die when the customer can finally do the arithmetic. The customer can do the arithmetic now.