If this comes to fruition it will be mass extinction event for a lot of pure play adtech companies.

Live look at AppLovin's law firm of record accounting department

Political CTV ads are wasting a ton of money using contextual and program level targeting data. I've seen so many ads for out of state races this cycle and it's because CTV targeting isn't what it claims to be... but people line up to buy anyway.

If Hill comes back to KC and wins a Super Bowl this year the collective hatred of the Chiefs will surpass peak Patriots hate during the Brady era.

It was not too much to handle

I'm surprised it took Apple this long to nuke all of the companies that continue to go to great lengths to identify users in their ecosystem without consent.

call off your Ad Week plans programmatic has been totally cancelled adexchanger.com/privacy/appl…

Your buying model optimizes for conversions. Bots fire conversions too. Fake conversions become training data, so your model bids up the fraud source. Stuffed IDs make fraud look like your best audience. IP blocks decay in days, and your model relearns the fraud. Layering AI on top doesn't fix it. Cleaner supply does. Our new brief covers five ways buying models amplify fraud and six questions to ask about your stack.

It's time to have a serious conversation about whether or not you're actually buying open web media or if you're just buying a sliver of it that carries some kind of EID (that may or may not be of any quality). Spoiler alert Made for Advertising IDs are causing far more damage to your media buy than Made for Advertising domains and apps. adexchanger.com/platforms/ap…

Another sign of the AI bubble bursting. A flood of agentic applications that have no reason to exist.

We may at the point where technology is actually making everything worse. Nobody wants more solutions. They already have too many tech solutions for everything. Which is another issue for AI and its army of wrapper solutions (I mean agents).

Your conversion pixel has one job: fire. It has no idea who fired it. A real buyer. A bot filling out your lead form. A view-through credited to someone who never saw the ad, matched to your impression by a probabilistic guess. To your bidding model, all three look identical. So it learns that the supply producing them is working, and moves budget there. Next cycle, more spend lands on that same supply. More bad conversions come back. The model gets more confident. If you do not address front-end media quality you can easily fall into a trap of training your buying model that poor quality is actually good.

Why do brands hold the open web to a far higher standard than social media? For some reason brand safety, AI slop, and made for advertising only seem to matter when it comes to open web publishers. If it's on social media nobody seems to care.

I don't understand why anyone would give Meta more control over their lives. #muse

"Independence" across various measurement and data solutions has been a hot topic recently. My two cents is independence doesn't matter. Transparency is what matters. If your vendor is transparent about their process and is able to prove the value they add who cares if it's independent 🤷‍♂️

According to the ANA Q1 2026 report the top-half of advertisers turned 54% of programmatic spend into qualified impressions. The bottom half managed only 32%!!! 𝗧𝗵𝗲 𝟮𝟮% 𝗴𝗮𝗽 𝗶𝘀 𝘁𝗵𝗲 𝗟𝗔𝗥𝗚𝗘𝗦𝗧 𝗘𝗩𝗘𝗥 𝗿𝗲𝗽𝗼𝗿𝘁𝗲𝗱! Why is this happening? Large brands and agencies are investing in sell side curation. They are identifying the most performant media and bundling it into PMPs for their own use. For buyers not taking control of their supply this means the trash inventory is increasingly flowing downstream to you. We now have the numbers to prove it.

The short answer is "yes." The more nuanced version is two things have to change to allow this to happen: 1️⃣ DSPs fundamentally have to change how they traffic shape. Buyers and curators are far better equipped to determine what traffic performs best for the, not the DSP. DSPs can't listen to all of the traffic available so they (mostly arbitrarily) only process a very small portion of the available traffic. Now the Dynamic Traffic Engine enters the room and I fear will exacerbate the core problem with DSPs cycling through the same, narrowly focused inventory leaving a lot of quality on the table (I will have more on this later). 2️⃣ Buyers have to drop their addiction to ID-based inventory for targeting and measurement. First, because IDs aren't always reliable (I love Christoph Bakke's term "Made for Advertising IDs"). Second, fraudsters know how to game the ID-based system so you end up buying more fraud. Third, the vast majority of traffic isn't tracked via an ID or cookie and even a large portion of that isn't mapped correctly so buyers miss out on premium, human inventory simply because it doesn't carry the IDs they rely on for (poor quality) attribution. adexchanger.com/technology/c…

Louder for the people in the back 📣

Not all lookalike seeds are the same, and the difference determines what your model becomes. Seed from CRM data and you are modeling against sales. Real people who paid you real money. Seed from a pixel and you are modeling against events. Some of those events are sales. Some are form fills that were never human, engagement that was never a person, conversions that fired because something was built to make them fire. The model cannot tell the difference. It finds more of what it was shown. Feed it fraud and it will go get you more fraud, efficiently and at scale, and report improving performance the entire time. Wasted spend can be fixed. A model trained on invalid conversions cannot.

Digital fraud has figured out how to fool even Google. You cannot protect your ad buys using the same old playbook. dayzlegame.com/blog/google-a…