Tokenizing Tokens; In OpenAI We Trust?

Agencies are considering becoming a futures market for tokens by buying wholesale from LLM and cloud providers and reselling to clients at a markup, a model akin to principal media trading with risks of conflicts, as agencies can no longer subsidize AI costs, according to Ebiquity CEO Ruben Schreurs. OpenAI has published ad credit policies that expire after 90 days and cannot be combined with other discounts, and it is testing a new ad format that launches custom brand agents inside ChatGPT, as reported by Search Engine Land.

By The Same Token Big agencies are wrestling with how to offset the rising cost of token consumption by turning it into a margin business. One idea on the table, as Digiday https://digiday.com/media-buying/the-case-for-and-against-agencies-betting-on-tokens-the-way-they-bet-on-media/ reports, is for holdcos to effectively become a “futures market for tokens.” Agencies would cut direct deals with LLMs and AI or cloud infrastructure providers to secure cheaper wholesale rates on token consumption. They’d then resell those tokens to clients at a markup, while still pitching it as a discount versus prevailing market prices. The business model is akin to principal media trading and comes with similar pros it’s cheaper for clients than buying at standard market rates and cons there’s a warped incentive structure that makes it easier for agencies to clandestinely bamboozle or swindle clients . One thing we haven’t seen from holdcos is a deal with a particular LLM – Perplexity, Claude, ChatGPT, et al. – to make its enterprise AI the default, or even exclusive, agent in exchange for a can’t-say-no price. Because agencies can’t bear the costs of AI investments for long. “What we saw in the last two years is that most agencies were carrying those costs entirely on their balance sheet,” says Ebiquity CEO Ruben Schreurs. They were effectively eating those expenses as they tried to secure clients. “But now, starting this year,” Schreurs says, “agencies can’t afford to continue to subsidize all those costs.” Credit Where Credit Is Due OpenAI is chugging along with its advertising plans. Six months into the company’s fledgling ads business, OpenAI has published its policies on how brands can apply ad credit to campaigns. Interestingly, the policies https://openai.com/policies/ad-credit-terms/ don’t refer to ChatGPT by name, but we know what they mean. OpenAI’s ad credits expire after 90 days and cannot be used together with other discounts. OpenAI reserves the right to apply credits to “eligible fees” – which it doesn’t define – and it can also “suspend, revoke or void” those credits at its discretion. What OpenAI lacks in transparency, it tries to compensate for with reassurances that marketers are getting their money’s worth. Or that’s the promise, at least. And if there’s anything OpenAI needs right now, it’s advertiser confidence. According to marketing intelligence firm Sensor Tower, OpenAI’s ads per user per hour more than doubled between April and the end of July, Marketing Brew https://www.marketingbrew.com/stories/openai-ad-credit-policies reports. But to keep attracting demand, OpenAI must be able to offer brands control and proof of performance. Last month, for example, OpenAI posted job listings for engineering positions focused on “ad formats,” suggesting the company plans to expand its Rolodex of ad units and build on the ad manager it launched for ChatGPT in March. Just Chatting Speaking of OpenAI’s advertising ambitions, the company appears to be testing an ad format that replaces the current click-out-to-a-brand’s-site setup with a new experience built around a custom brand agent that lives inside the chat, Search Engine Land https://searchengineland.com/openai-appears-to-be-building-chatbot-native-ads-that-launch-ai-agents-484107 reports. H/t to Juozas Kaziukėnas https://www.linkedin.com/feed/update/urn:li:activity:7488571152357810176/ , an ecommerce industry analyst and founder of Marketplace Pulse, for flagging the test. Here’s how it works: ChatGPT crawls a company’s sites and pages, reviews, metadata, business profiles, product feeds and support pages – essentially everything it can possibly scrape on a brand – plus the data feeds provided by business customers. Advertisers can then customize the agent and layer in additional data sources so that clicking the ad sends users into a new chat-within-a-chat experience with that agent. This new format wouldn’t be a heavy technical lift, and the tech behind these custom advertiser agents also isn’t new; it already exists in ChatGPT’s toolkit. The web was built on the promise that “an ad generates a click, and the click opens a webpage,” Search Engine Land notes. ChatGPT’s new ad model would flip that. As SEL puts it, “The conversation, not the website, becomes the destination.” But Wait There’s More Marketing effectiveness used to be just another front in the ongoing conflict between CMOs and CFOs. But now it’s something for CEOs to boast about. Digiday https://digiday.com/marketing/future-of-marketing-media-measurement-is-becoming-an-asset-in-boardroom-theater/ X and the WFA finally settled their ad boycott dispute, but the brand safety wars aren’t over. Adweek https://www.adweek.com/media/x-and-wfa-settle-but-the-brand-safety-wars-are-far-from-over/ New York is suing Kalshi, alleging that it’s an illegal gambling operation as the state seeks to rein in prediction markets. WSJ https://www.wsj.com/tech/new-york-alleges-kalshi-running-illegal-gambling-operation-18ae71b0 Google Earth briefly introduced and then promptly removed a feature that allowed users to alter satellite images with AI, prompting a wave of backlash. Digital Digging https://www.digitaldigging.org/p/how-to-plant-a-nuclear-plant-in-iran