You automated your ad buying. So why isn't it working?

The Speed Read

• AI bidding agents are making budget and targeting decisions on attribution models, audience segments, and conversion data that haven't been re-validated in years. The agent is working exactly as designed. The inputs aren't.

• Reddit has become a primary training dataset for large language models because it captures real human purchase conversations at scale. If your brand's Reddit presence is thin or absent, that signal flows directly into AI product recommendations.

• TikTok launched Agentic Hub, a central integration point for third-party AI tools that connects to TikTok's own MCP server. Outside AI agents can now bid and manage campaigns inside your TikTok account directly.

• Apple has exited its managed ad network, widening the iOS measurement gap at the same time a new survey found 49% of senior marketers can't justify their creative spend to a CFO with hard data.

The AI Running Your Media Buys Is Acting on Data From Before Your Last CRM Migration

A column in AdExchanger this week said out loud what most performance marketers already know and haven't done anything about.

Ad tech spent a decade building verification standards for the supply side. Viewability, brand safety, fraud detection: if an impression looked fake, someone built a tool to catch it. Nobody applied that rigor to the demand-side data now feeding AI bidding agents.

Your bidding algorithm is probably optimizing toward a "high-value" audience segment built on a customer tagging taxonomy from a CRM migration or two ago. The attribution model it's referencing may still be crediting a channel your team stopped buying 18 months back. Nobody flagged either, because nobody's job is to flag it.

Automating a bad input doesn't fix it. It just moves it downstream faster, with more money behind it.

This is worth taking seriously if you're running Meta Advantage+, Google Performance Max, or any bidding tool where a platform algorithm is making decisions on your behalf. Those systems are only as good as what you fed them. If the underlying conversion events, audience definitions, or attribution models haven't been touched since before your last major platform migration, you're optimizing toward a target that no longer exists.

The practical step this week: pull whoever manages your automated bidding and ask three questions. When were the audience segments last validated? When did anyone last audit the conversion events the model is optimizing toward? And is the attribution model still crediting the channels you're currently running, or ones you've long since cut?

If the honest answer to any of those is "I'm not sure," that's the finding. The audit is the deliverable.

If you want a second set of eyes on what your automated systems are actually optimizing toward, that's a conversation worth having.

Reddit Is Now an AI Search Signal. Most Brands Aren't Treating It Like One.

Most brand teams added Reddit to their "monitor but don't prioritize" list years ago and haven't looked at it since.

A new AdExchanger deep dive explains why that's about to cost them. Reddit's billions of posts and comment threads have become a primary training dataset for large language models, specifically because Reddit captures real human purchase conversations at scale. Not your brand's polished press releases. Not your agency's category whitepapers. The actual conversations people have when they're deciding whether to buy something.

When a potential customer asks ChatGPT or Perplexity whether your product is worth buying, that answer is shaped partly by Reddit.

Not your reviews page.

Not your product descriptions.

The conversations people have had about your brand and your category for years, unfiltered.

Brands with thin or absent Reddit presence aren't just missing a channel. They're missing an input to the AI layer now sitting between buyers and brand discovery.

The immediate audit worth running: search your brand, your product category, and your three closest competitors on Reddit. Look at what's there and what's not. If the most recent relevant thread is two years old, or if the only posts mentioning your brand skew negative, that's what's feeding the AI when it answers product questions.

This doesn't mean Reddit advertising is suddenly mandatory for everyone. It means the calculus for being on Reddit changed. Before, the question was: are your customers on Reddit? Now the question is: what is the AI learning about your brand from Reddit, and do you want to shape that or leave it to chance?

For DTC brands where category purchase conversations run deep on Reddit — supplements, apparel, home goods, pet products, personal care — the case for some level of presence, paid or otherwise, just got stronger.

Quick Takes

TikTok just opened your ad account to outside AI agents. 

The new Agentic Hub centralizes third-party AI tools and connects directly to TikTok's MCP server, the same protocol standard Anthropic uses for AI-to-tool communication.

Unlike Meta's Advantage+ or Google's Performance Max, which keep automation internal, this invites outside agents to bid and manage campaigns directly inside your TikTok account.

If your agency is building or buying any AI tools, the friction to connect one to your TikTok account just dropped to near zero. Ask them this week: which AI agents have or will have access to your TikTok account, and who in your organization approved that?

Apple walked away from its ad network at the worst possible moment. 

Apple exited its managed ad network, removing another measurement option for iOS performance campaigns and leaving a wider attribution gap than existed before.

Separately, a Gain Theory survey found 49% of senior marketers can't back their creative decisions with hard data when the CFO asks. The double bind: iOS measurement is getting harder at the same time internal budget scrutiny is increasing.

If your measurement stack still relies on click-based attribution for iOS-heavy campaigns, pressure-test it now rather than after the next iOS update closes another door.

Agentic advertising is adding intermediaries, not cutting them. 

Raptive's Ryan Maynard made a sharp argument in AdExchanger: agentic ad systems aren't eliminating supply chain middlemen.

They're relabeling them as agents and adding new fee layers.

Buyer agents. Seller agents. Agent-to-agent protocols. Each with its own markup. Before your agency starts pitching agentic tools as a cost-efficiency play, ask them to put the actual fee structure in writing.

The efficiency argument only holds if you can see what you're paying for.

The CRM Behind Every Win

Attio is the agentic CRM that runs the work behind every win.

With agents and automations that build pipeline, chase signals, and move deals forward, Attio orchestrates your revenue work around the clock.

Loved by high-growth startups like Granola, Modal, and Wispr Flow, Attio amplifies what your team can achieve.

Get Started Now

The Last Word

The automation cycle tends to go the same way.

You hand something to a platform, the platform optimizes it, results look reasonable, and nobody looks under the hood. Then a quarter later the numbers are soft and everyone's staring at the dashboard trying to figure out when it went wrong.

What this week's stories have in common: none of the problems announce themselves. Stale bidding data looks like normal optimization at first. Absent Reddit presence looks like a channel you haven't invested in. An AI agent with full account access looks like a convenience your agency set up to make things easier.

The audit doesn't feel urgent until it is. Run it anyway.

Know someone who should be reading this?

If this was useful, forward it to one person managing media budgets who doesn't have time to read everything. That's exactly who this is for.

Forward this email →

If this isn't landing for you anymore, unsubscribe below. No guilt, no follow-up.