The Token Economy: Agencies Learn That AI Runs on Quarters
Remember when agencies sold ideas? Now they're selling tokens. Not the subway kind—the AI kind. And nobody quite knows how to charge for them yet.
As generative AI becomes less "experimental pilot" and more "how we actually work now," agencies are confronting an awkward truth: those ChatGPT sessions cost money. Real money. A recent Coca-Cola campaign required 70,000 prompts and millions of tokens. At a fraction of a cent each, that adds up to "actual line item in the budget" territory.
The industry's response? Shrug collectively and try different things.
Pass the Buck (Literally)
Some agencies are passing token costs directly to clients. Full-service agency Merge treats them as metered expenses. Production shop Big Spaceship files them under "production costs"—like catering, but for robots. Silverside AI has gone full SaaS, charging "seat" pricing like they're selling Salesforce subscriptions.
Others are absorbing the cost, at least for now. RPA, a full-service agency, refuses to charge clients for something "we don't know is actually going to work for them." Which is refreshingly honest, if potentially unsustainable. Anomaly's global head of engineering was more direct: passing on AI overheads "feels like a money grab."
He's not entirely wrong.
The Bulk-Buy Gambit
Brandtech's Pencil platform has taken the enterprise route, negotiating bulk deals with OpenAI and Anthropic, then charging clients "generation credits"—each equivalent to a chat response, image generation, or second of video. Think of it as buying AI in bulk at Costco, then reselling the toilet paper roll by roll.
The arrangement inevitably invites comparison to principal media buying—agencies acquiring assets at one price and reselling at another. "The bigger economic shift isn't token arbitrage," counters Avelin CTO James Londal. "It's labor compression." If AI meaningfully replaces reporting teams and middleware connectors, the financial upside "dwarfs any marginal bulk discount dynamic."
Maybe. But try explaining that to a procurement officer.
The Horizon Approach
Media agencies are charting their own course. Horizon Media, which launched its Blu platform in December, charges clients a "nominal" fee—"cost recovery, not revenue," according to chief product officer Domenic Venuto. The goal is getting clients onto the platform without creating barriers. At 40 clients with a dozen users each, token usage hasn't become material yet. But "tens of thousands" of users would change the math.
Kepler, another media agency, has declined bulk token deals entirely. "Impact is the metric that we care about," says VP Peter Rice. "Token usage is the fuel to deliver that impact." Which sounds philosophical until you realize he's basically saying "we'll figure out the billing later."
The Story Nobody Wants to Tell
Here's the uncomfortable truth: holdcos have AI stories, but they don't have AI business models. WPP, Publicis, Omnicom, Dentsu—they've all got earnings call talking points about "AI as margin defense" and "automation as productivity." But the pricing question remains unanswered.
When AI automates creative production and removes humans from media buying, the old time-and-materials model collapses. Pay per asset breaks down when assets are generated at scale. Retainers feel arbitrary when output isn't tied to hours. Outcome-based pricing sounds good but requires client data access and board-level trust most agencies haven't earned.
"Almost all the focus is about efficiency—maybe a bit about effectiveness—but there's very little about real value creation," says Paul Frampton-Calero, CEO of Goodway Group. "AI will give you faster and cheaper. But faster and cheaper doesn't mean any better."
The industry's honeymoon with AI is over, according to Gartner's Jay Wilson. CMOs who were promised cost savings are still waiting for lower fees. The question of "what does this cost and who pays for it" has moved from hypothetical to urgent.
Somewhere, a procurement team is already preparing the audit request.
