Show Notes
Adam and Gareth open the show reminiscing about classic Crazy Eddie style electronics commercials, prompting Adam to share a fascinating history lesson about how Times Square camera shops became some of the very first search advertisers in 2002. The conversation transitions into a deep dive on creative optimization, explaining why ugly ads often outperform slick creatives and warning that relying on LLMs for media planning will lead to convergence—where every brand is pushed toward the exact same average strategy. The core of the episode focuses on a radical proposal to bypass the traditional programmatic supply chain (DSP to SSP to Ad Server). Adam pitches a future where enterprise brands mount their custom bidding algorithms directly inside publisher ad servers, using Netflix as the prime example of where this thin layer architecture should start. They also expose a major misconception about measurement APIs (like CAPI): when brands send conversion data to walled gardens like Meta, the platform doesn't actually model on that data to find new customers; it simply quarantines the data to take credit for the conversions. In the news segment, the hosts analyze Kevel’s $52 million raise to power retail media networks. They also cover Meta's controversial new terms of service that allow the tech giant to scrape publisher websites for AI training, and they close by examining why agency holding companies like Publicis and Stagwell are trying to merge AI Systems Integration (SI) with media buying—despite SI staffing being the hardest business in the world. CHAPTER TIMESTAMPS 00:00 Crazy Gideon & Early Search Ads for Drop-Shipping04:47 Why Ugly Ads Work & AI Creative Convergence08:19 UMPs, Black Boxes, and the LLM Average17:26 Multi-Touch Attribution & The Measurement Trap20:59 Why CAPI Doesn't Actually Train Facebook's Bidding Model24:36 The Netflix Pitch: Mounting Bidding Logic Inside the Ad Server35:06 Kevel Raises $52M for Retail Media Ad Serving42:08 New Mexico's Massive Fine Against Meta44:16 Meta's Terms of Service Update to Scrape Publishers47:05 Agency Holdcos, Systems Integration (SI), & AI Transformation
More from AdTech AdTalk
Gareth Glaser and Adam Heimlich open the show with a discussion on the evolving AI market, distinguishing between "efficiency AI" and "effectiveness AI." Adam posits that enterprise advertisers are finally moving past their obsession with AI that cuts operational costs and are now ready to explore how agentic AI can improve campaign outcomes. The conversation touches on Amazon and Meta's new agentic commerce tools, noting that while discovery may be aided by AI, the walled gardens will likely prevent third-party shopping agents from fully executing transactions within their closed ecosystems. The core of the episode tackles the highly technical and controversial topic of Extended Identifiers (EIDs). Following an academic report revealing the prevalence of EIDs in OpenRTB bid streams, Gareth provides a masterclass on the history of identity resolution, detailing the shift from deterministic cookie syncing to probabilistic fingerprinting after the "cookie apocalypse." Adam and Gareth highlight a massive misalignment of incentives: sellers and SSPs are motivated to probabilistically pass IDs that represent high-converting users, deceiving buyers into thinking they are bidding on known, deterministic audiences to drive up auction prices and claim unearned attribution credit. The show closes with a fiery debate on publisher monetization and Google Ad Manager (GAM). The hosts recall a 2019 scandal where Google secretly re-routed inventory to AdX at discounted prices under the guise of COPPA compliance, costing publishers millions. Finally, they dissect Prebid's new proposal to invert the ad-serving hierarchy by making Prebid Server the primary decision engine, calling GAM via API. However, Gareth issues a stark warning: if regulators do not close a crucial loophole allowing Google to hoard its proprietary click-propensity data exclusively for AdX buys routed directly through GAM, the entire effort to break Google's ad-serving tie will be rendered useless. CHAPTER TIMESTAMPS 00:00 Show Intro & Post-COVID Malaise01:50 Efficiency AI vs. Effectiveness AI03:03 Agentic Commerce: Amazon & Meta's Closed Shopping Agents11:51 The Academic Report on EID Prevalence12:37 The History of Cookie Syncing & Deterministic IDs16:16 The "Cookie Apocalypse" & The Rise of Probabilistic Fingerprinting20:50 The EID Fraud Problem: Passing Probabilistic Guesses as Deterministic IDs23:14 The Misaligned Incentives of EID Injection28:22 Apple Safari's Aggressive Blocking of ID Resolution URLs31:03 The Solution to ID Fraud: Better DSP Disclosure38:17 GDPR, Cookie Panic, and the EU's "Free Pass" for Google44:02 The 2019 Google Ad Manager COPPA Scandal48:28 Prebid's Proposal to Overhaul the Ad Server Hierarchy53:04 The Dangerous Loophole in the DOJ's Google Remedies01:00:10 Breaking Google's Attribution Loop
Adam Heimlich and Gareth Glaser kick things off with a nostalgic Crazy Bruce's Liquors commercial before diving into the aftermath of Advertising Week. They introduce the concept of "fragmalidation"—the simultaneous fragmentation and consolidation of the ad tech industry as enterprise advertisers build private, disjointed stacks that rely on a few hyperscaled centralized cloud providers. The hosts discuss Amazon's major move to combine Amazon Ads and Amazon DSP into a single agentic interface on AWS Bedrock, arguing that cloud giants are effectively turning AI models like OpenAI and Anthropic into mere suppliers. The conversation then shifts to Zeta's massive Ad Week event featuring Palantir's Alex Karp, sparking a debate on Zeta's $8.5 billion market cap and how service-heavy operators like MIQ successfully scale global managed services. Adam introduces the concept of "invisible asymptotes" (citing Amazon Prime and T-Mobile), explaining how enterprise brands are realizing their internal data science investments are capped by the limitations of pooled DSPs. He details how Chalice is pitching direct-to-market data modeling using common embedding spaces and GPU infrastructure. In the Postmortem segment, the hosts dissect a chart from Nexon CMO Ben Kaplan projecting ad spend growth, noting that traditional media dollars shifting to digital often masquerade as the death of the open web. Next, they analyze James Doherty's eye-opening CTV pricing chart, which reveals Netflix CPMs hitting $37. Adam shares a fascinating calculation that, when adjusted for screen coverage and attention, Facebook's $8.50 CPM is actually more expensive than a Super Bowl ad. They wrap up by reviewing Paul Banister's prediction that Apple might attempt to monetize Safari by mediating ads directly, similar to Apple News. 00:00 Crazy Bruce's Liquors Ad & Show Intro01:21 Advertising Week Recap & Private Enterprise Stacks03:10 Fragmalidation: The Convergence of Fragmentation and Consolidation05:43 Amazon Ads & DSP Get an Agentic Rebrand08:30 How Amazon AWS Will Lock DSPs into 5-Year Cloud Deals11:56 AWS Bedrock Turns OpenAI & Anthropic into Suppliers14:09 Chalice & The Application Layer Battle20:45 Zeta's Ad Week Event with Alex Karp & $8.5B Market Cap24:19 Why Service-Heavy Operators Like MIQ Win Global Business27:46 Bach's Article & Going Vertical in the Agentic World31:26 Invisible Asymptotes: T-Mobile, Amazon Prime, & Ad Tech34:39 Bespoke Suits vs. Off-The-Rack DSP Algorithms37:48 Pitching Data Sellers on GPU Embedding Stores45:01 Postmortem: Ben Kaplan's Chart & The Death of the Open Web47:46 James Doherty's CTV Pricing Chart (Netflix $37 vs. Disney $25)49:10 Why Facebook Ads Cost More Than the Super Bowl53:32 Selling CTV on Feelings vs. ROI57:26 Paul Banister: Will Apple Monetize Safari Like Apple News?01:00:12 Yahoo Scrolling & Outro
Adam Heimlich and Gareth Glaser open with a classic Wilford Brimley "Diabetes" intro before launching into a chaotic week in ad tech. Apple sent shockwaves through the industry by updating Safari to aggressively block URLs associated with browser fingerprinting. Among the affected companies—including LiveRamp, ID5, and Permutative—Apple took the drastic step of blocking The Trade Desk’s primary ad-serving and impression-tracking domain. The hosts debate whether browser makers are overstepping their bounds, examining common carrier laws, net neutrality, and the blatant double standard of Apple giving Google search a pass in exchange for $20 billion annually while crippling independent ad tech. The conversation turns to the future of identity resolution, with Gareth predicting an accelerated migration from client-side scripts to server-side processing and CDN edge workers to bypass browser-level blocking. The hosts also explore Server-Side Ad Insertion (SSAI), ad stitching, and the rise of Unified Media Platforms (UMPs) as big tech companies build proprietary ad networks. Closing out the episode, Adam and Gareth react to a sobering economic projection claiming that current AI investments would require American households to spend $300 per month (9% of US GDP) on AI bills. Finally, they dissect the bankruptcy of Open Web (formerly Disqus/Vocal), with Gareth delivering a fiery rant on how third-party comment widgets act as "leaky buckets" that undercut publisher ad yields. KEY TOPICS • Apple Safari Blocks The Trade Desk: Apple updated Safari to block URLs associated with fingerprinting, targeting ID5, Permutative, LiveRamp, and notably blocking The Trade Desk's core ad-serving and tracking domain. • Browsers as Common Carriers & Tying: The hosts discuss whether browsers should be treated as common carriers prohibited from altering web content, contrasting Apple’s crackdowns with the $20 billion exception given to Google search. • Identity Moving to CDN Edge Workers: Gareth predicts that identity vendors will abandon client-side scripts in favor of server-side processing and CDN edge compute to make identity requests indistinguishable from publisher content. • The $300/Month AI GDP Projection: Adam breaks down a professor's study showing that for current AI valuations to pay off, AI spend would need to reach 9% of US GDP, equating to a $300 monthly bill per American household. • Open Web Bankruptcy & "Leaky Buckets": Following Open Web's Chapter 11 filing, Gareth explains how third-party comment widgets create "leaky buckets" by running separate, cheaper ad auctions that cannibalize publisher yield. CHAPTER TIMESTAMPS 00:00 Wilford Brimley Intro & Show Opening01:14 Q4 Instability & Industry Shockwaves03:27 Apple Safari Blocks The Trade Desk & ID Vendors06:33 DoubleClick Exemption vs. Trade Desk Impression Tracking07:50 Browser Tying, Monopolization & Apple’s $20B Google Deal10:27 Browsers as Common Carriers & Net Neutrality13:08 Frequency Capping on Safari (UID1 vs. UID2)16:22 Ian Meyers vs. Apple Engineer ("Hardest Hit Keystrokes")19:51 Voluntary CTV Inventory Deals vs. Unilateral Browser Bans21:37 Ad Blockers, Direct Mail & Consumer Privacy Hypocrisy26:13 AI Browsers & Government Option / Post Office Analogy30:06 In Defense of the Open Web as a Standardized Ad Network32:00 Moving Identity Server-Side to CDN Edge Compute35:49 Server-Side Ad Insertion (SSAI) & Brutal Ad Blockers38:09 The Evolution of CDNs: From Dumb Storage to Edge Compute40:36 Unified Media Platforms (UMPs) & Open Web Ad Networks44:36 Enterprise vs. SMB Ad Tech: AI Agents & Data Collaboration47:19 The Economic Reality of AI: 9% of US GDP ($300/Mo Bills)51:22 Open Web Bankruptcy & Publisher "Leaky Buckets"58:18 Caddyshack Quotes & Wrap-Up
Adam and Gareth kick off with a vintage 1980s diet pill commercial before diving into an overlooked AWS blog post from June 2026. Amazon is quietly building advanced algorithmic decisioning infrastructure, specifically deploying agentic bidding using NVIDIA GPUs within AWS Bedrock. The hosts break down how this allows real-time inference and model hosting directly in the bidstream, potentially positioning Amazon as the master layer controlling all programmatic agents.On the sell side, Gareth revisits the stigma of "buying traffic," explaining how smart publishers are now using paid social channels for "user exfiltration"—rescuing users from walled gardens like Facebook to build direct lifetime value (LTV). The conversation also touches on the death of Google Reader, the clunky nature of email newsletters, and the desperate need for a modern RSS feed experience.The buy side discussion focuses on the race for agentic supremacy. As ChatGPT launches its own ads manager and Amazon Bedrock strikes deals to be the master UI for ad agencies, the hosts debate whether big tech giants will cooperate or ice each other out. Finally, Adam and Gareth explore the limitations of clean rooms, why Palantir’s enterprise lock-in pitch is gaining traction with holding companies like Stagwell, and why Google's return of price floors in GAM is just another tool to manipulate bidding algorithms.__________________________________________________KEY TOPICS- Amazon Bedrock’s Agentic Bidding Architecture: Amazon is building an advanced programmatic bidding infrastructure utilizing NVIDIA GPUs in AWS Bedrock, allowing for real-time model inference and containerized decisioning directly in the bidstream.- "User Exfiltration" & Buying Traffic: Gareth argues that publishers buying traffic on Facebook is not MFA spam, but rather a valid arbitrage strategy to exfiltrate users from walled gardens and build direct lifetime value.- The Return of the RSS Feed: The hosts lament how email has become a disorganized RSS feed and advocate for a unified content consumption experience to replace the clunky newsletter format.- The Agentic Master UI War: As ChatGPT rolls out an ads manager, Amazon is aggressively positioning Bedrock Agent Core as the master interface for agencies, sparking a debate on whether tech giants will block each other's agents.- The Return of GAM Price Floors: Google has reintroduced price floors in Google Ad Manager, leading to a debate on whether yield management is meant to protect inventory value or simply manipulate bidding algorithms for higher prices.__________________________________________________00:00 1980s Ayds Diet Pill Commercial & Show Intro01:34 Amazon's Secret AWS Bedrock & NVIDIA GPU Ad Tech06:24 Breaking Down the AWS Agentic Bidding Architecture09:24 Inter-Channel vs. Intra-Channel Optimization11:46 The Clunky Newsletter Problem & The Need for RSS Feeds14:06 How Advertising Attribution Boosts GDP16:16 The AI Application Layer is Just Coins in a Wishing Well17:44 Palantir's Alex Karp & The Illusion of Data Sovereignty20:30 Stagwell's Seven-Year Palantir Deal25:27 Why Agency "In a Box" Offers Fail Enterprise Brands33:26 The High Cost and Limitations of Clean Rooms44:11 OpenAI's $100 Billion Goal & ChatGPT's Top Advertisers46:50 Why Bottom-of-Funnel DR Ads Fail on ChatGPT48:28 Google Reintroduces Price Floors in GAM51:21 The True Purpose of Yield Management & Price Discrimination59:23 Payday Loans vs. Creatine Gummies
Adam Heimlich and Gareth Glaser tackle the unsealing of the government's 108-page Google ad tech antitrust remedies document. Adam and Gareth break down the core structural issue in digital publishing: why every major website is forced to maintain two separate demand setups—Google Ad Manager (GAM) for Google demand and Prebid/TAM for everything else—rather than running a single unified auction. Gareth explains that if AdX and AdWords are forced to bid into Prebid on equal terms, up to 80% of publishers who lack direct sales teams could abandon GAM entirely, speeding up web page loads and opening up true competition among sell-side ad servers. The hosts dive into Scott Messer's viral analysis highlighting a glaring loophole in the remedies that permits AdWords to continue favoring AdX whenever Google claims it provides superior advertiser value, privacy, or fraud protection. Adam counters with a historical parallel to the 2001 Microsoft Windows antitrust settlement, arguing that behavioral remedies can succeed when executive leadership tires of regulatory litigation and chooses to compete on product quality rather than illegal ties. To wrap up, the hosts analyze an Ad Age report detailing OpenAI's path toward a $100 billion ad business. They examine why top early spenders like Capital One and Expedia are experiencing disastrous direct-response performance on ChatGPT. Gareth and Adam explain that unlike Facebook or Fox News, where passive scrolling environments drive impulse clicks, reading-focused LLM users simply do not engage with performance ads. KEY TOPICS The DOJ Remedies Document & The "One Line" Unified Auction: Adam and Gareth analyze the government's 108-page antitrust filing, explaining how forcing AdX to bid directly into Prebid could allow the 80% of publishers without direct sales teams to abandon GAM entirely.