Adexchanger iconAdexchangerSep 8, 2026 ~6 min source read

Keeping AI Costs In Check; Selling Your Data To The Highest Bidder

PMG, for example, built a tool that lets users limit AI token usage and, based on how often they enforce limits, offers certain recommendations. Lots of self-serve ad platforms, like MNTN, Streamer.ai and Amazon, offer AI image and video generation tools to businesses that can't afford to hire their own marketing and design teams.

Keeping AI Costs In Check; Selling Your Data To The Highest Bidder

Share this story

Send the public story page.

Useful takeaways from this story.

A bankruptcy auction for Spirit Airlines' data drew competing bids: Google at $10M and micro1 at $12.5M for troves that include millions of messages.

AI image models struggle with common food textures and structures, producing uncanny results that can harm local restaurants using self-serve ad tools.

The useful part

Roughly 60% of companies that use AI will incur unexpected charges and/or exceed their allocated budgets due to a lack of human oversight, according to a Gartner survey of 1,300 senior marketers. The survey also found that more than half (56%) of companies that use AI tools implemented them without clear policies. (So much for that whole "human in the loop" thing.) Rather than risk flushing more money down the toilet, media agencies are developing their own auditing tools to rein in AI token costs, Digiday reports.

How it works

  • PMG, for example, built a tool that lets users limit AI token usage and, based on how often they enforce limits, offers certain recommendations.
  • Rise still has to estimate the costs of its agentic testing itself, including how much of its campaign budgets went toward working media.
  • Dangerous Data Deals Google has a new competitor in the bidding war for Spirit Airlines' treasure trove of data.
  • Google similarly promises to strip out personal information tied to Spirit's customers.
  • But, quibbles over who would be a better steward of the data aside, micro1 faces an uphill battle in getting Google's already-approved deal overturned.

What to take from it

Certain non-repetitive structures that are common in food, such as bubbles, seeds and thin noodles, are difficult for diffusion models to replicate, behavioral scientist Giovanbattista Califano tells The Verge. Which poses a problem for the exact kinds of small and local businesses being courted by the digital advertising industry. It looks like it'll take some more time before agencies feel they can fully entrust their operations to AI agents in the pursuit of efficiency.

Example or evidence

  • which includes 500 million Microsoft Teams communications, 100 million emails and 16 million customer support chats.
  • Publishers Chrome Has A New Way To Measure Ad Overload On The Web Chrome is introducing new metrics that give advertisers and publishers a more data-driven picture of what users actually experience on...
  • The ID verification business has exploded as global regulators pass laws requiring age verification to access social media and adult content.
  • But, as Gizmodo notes, the Nexus hack "shows the danger of creating honey pots of our most sensitive documents for malicious actors to exploit." No Such Thing -Generated Lunch Why does food tend to look so...

Details worth keeping

Token For A Ride AI can help cut costs, but only if it has a babysitter. Another example is Rise, a performance agency that's part of Quad. While the Spirit auction has raised new concerns over what happens to a company's data once it goes out of business, it's not the only recent example of the dangers of companies' hoards of personal data potentially falling into the wrong hands.

Related coverage

  • Legaltechdaily: I keep hearing rumblings about AI pricing.
  • Kalungi: Ask yourself one question before you build another workflow around a cheap AI tool: who's actually paying for how cheap it is?
  • E27: The cost of AI is not only hidden in data centres.
  • Legaltechmonitor: I keep hearing rumblings about AI pricing.
  • Adexchanger: Look at ad tech's press releases this year.

More context around this story.

The true cost of AI is beginning to surface
E27 iconE27Aug 25, 2026

The true cost of AI is beginning to surface

The cost of AI is not only hidden in data centres. It is hidden in the systems we are learning to depend on. Artificial intelligence still feels cheap. A manager pays for a monthly subscription. A developer opens a coding assistant. A student asks a chatbot to summarise a report. The interaction is instant, polished […

The Ads Model for Prompts Vertically Integrates AI
Tomtunguz iconTomtunguzSep 3, 2026

The Ads Model for Prompts Vertically Integrates AI

Meta's Muse Spark 1.3 prices inference at two levels : $1.25/$4.25 per million tokens for private data, & $0.10/$0.20 for training consent : a 92% price spread. Michael Mauboussin taught that market prices contain information about underlying expectations. This spread establishes a liquid clearing price of $1.24/m toke

Loading more related stories...

Keep reading in the app

Open the app view to save this story, compare related coverage, and continue from the same source.

Open in app