Remember me
As marketing agencies embed generative AI applications into their work, they’re adapting their pricing models to account for the cost of AI tokens.
Generative AI prompts and responses don’t come for free. LLM developers like OpenAI use tokens as a means of metering AI compute; the text you type into a prompt and the output generated by a model are counted. The more tokens processed, the more compute is used and the higher the cost. And while the tokens themselves cost a fraction of a cent, they add up fast; a recent Coca-Cola ad campaign required 70,000 prompts and millions of tokens, according to its makers.
“There’s no one model that’s the one silver bullet … you have to use a lot of different models and all of those have different token economics,” explained Johnny Rohrbach, co-founder and director of partnerships and operations at Silverside AI, a production studio backed by the founders of indie agency Pereira O’Dell.
Related Insights
Marketing services companies need to decide whether to eat AI overheads, or pass it on to clients. So far, there’s no single answer to that question. Digiday spoke to media houses, creative agencies, production shops, full-service agencies and consultants to understand the lay of the land.
‘We can’t charge clients for something we don’t know is going to work’At full-service agency Merge, token costs are passed on to clients on a metered, case-by-case basis, said Kyle Smith, its chief technology officer. Production shop Big Spaceship follows a similar framework, treating compute as another budget item akin to a catering or equipment hire. “We’re treating it similar to a production cost,” said CEO Taryn Crouthers.
Silverside, which produced vodka brand Svedka’s Super Bowl ad using gen AI, uses a subscription model with “seat” pricing similar to that operated by companies like Salesforce, per Rohrbach.
Full-service agency RPA, however, takes the opposite approach — it absorbs its clients’ token costs. Lisa Herdman, svp and executive director of video investments, suggested the benefits of the technology for clients were still too uncertain to build a pricing framework around. “We’re still in the testing phases … We can’t expect to charge our client for something we don’t know is actually going to work for them,” she said.
Chris Neff, global head of engineering experience and technology at Anomaly, was similarly wary of passing on AI overheads. “It feels like a money grab,” said Neff.
‘The more they commit … the cheaper we can make that’Pencil, the AI marketing platform of marketing services group Brandtech, has been able to access economies of scale in agreements with LLM developers like Anthropic and OpenAI. Clients are charged for “generation credits” equivalent to a single chat response, an image generation or a second of video, each taking up tokens. They’re not charged for tokens used on set-up and client onboarding, said CEO and co-founder Will Hanschell.
“The client commits to a certain number of generations in Pencil; they can spend across any of the models or use cases and [it] is priced on volume. The more they commit, the more we can go and negotiate with the model providers, and the cheaper we can make that,” said Hanschell. Clients are assigned a set number of tokens based on an initial cost estimate; if their needs exceed that amount, they “top up” with another tranche.
Hanschell said this pricing method kept things simple for Pencil clients as it competes with creative and production agencies. “If we charge by usage, our incentives are aligned to the client,” he said. Tiered pricing goes some way to ensure that Pencil’s largest customers aren’t subsidizing the token prices enjoyed by its smaller ones, and the company offers different pricing plans for smaller clients, reserving legal indemnification for larger partners.
‘We don’t want to create barriers for clients’Media agencies are taking a different approach. Horizon Media, for example, debuted its gen AI media planning and strategy platform Blu in December 2025. According to evp and head of tech Krish Kuruppath, client onboarding and the company’s ongoing development of Blu account for the majority of its token costs; he declined to share how much Horizon spent on tokens in a given year.
Since its launch last year, Horizon has brought 40 clients on to the Blu platform, each with around a dozen users logging in. Kuruppath said token usage would only become a significant cost issue for the company once its user base reaches the “tens of thousands.”
For now it levies a “nominal” fee to clients as they begin using Blu. “It’s cost recovery, not revenue,” said chief product and data officer Domenic Venuto. “We don’t want to create barriers for our clients. But at the same time, we don’t erode our margins as well.”
Media agency Kepler, for example, offers its own AI-enabled planning tool called KipAIR. Peter Rice, vp of data strategy and analytics, said it hadn’t pursued bulk deals on token costs with AI providers. Rice suggested that passing on the cost of AI overheads didn’t make sense for the agency, which typically works with clients on a retainer basis.
“We want it to be embedded into the way we deliver for our clients,” he said. “Impact is the metric that we care about. Token usage is the fuel to deliver that impact.”
Full-service agency Lerma/ takes a different route. According to Josh Archer, chief data and intelligence officer, the company has one bulk-buying agreement in place with an LLM provider (Archer didn’t say which provider). As such, AI tokens costs are included in its initial pricing estimates for clients — at cost.
“We’re not marking it up, we just show it as a line item similar to other costs,” said Archer. Should the agency underestimate and find itself with a surplus of tokens, he said it would provide additional assets to the client. “We try to deliver more. It’s a great opportunity for us to push ourselves,” he said.
‘You want to compete on merit’Brandtech’s bulk deals for tokens invite comparison with one of advertising’s most contentious practices: principal media buying. If an agency can acquire large amounts of tokens at a discount, what’s to stop them passing that cost on to clients at a markup?
James Londal, founder and CTO of consultancy Avelin, suggested pursuing a principal media model would likely prove a dead end for agencies. “The bigger economic shift isn’t token arbitrage. It’s labor compression. If AI runtime meaningfully replaces reporting teams, duplicated dashboards and middleware connectors, the financial upside dwarfs any marginal bulk discount dynamic,” he said in an email.
Related Insights
Hanschell agreed. “You want to compete on merit, not on lowest price,” he said.
Even if they’re not attempting to add a price premium, agencies that pass through token costs risk becoming “solely focused on cost and margin, rather than focused on value and architecture that is appropriate for our agency and client,” said Jess Lewis, global chief technology and data officer at Crossmedia.
It’s probable that token pricing could become part of agency pitches in months and years to come. Ebiquity CEO Ruben Schreurs told Digiday that he expected the consultancy to be, in time, asked to audit an agency’s token usage just as it audits media spending now. “If it’s part of the contract … then yes, we’d audit that,” he said.
Knowing client CFOs will keep a close rein on any pass-through costs, agency leaders aren’t keen to reduce their use of AI to dollars and cents saved.
It’s more effective, Schreurs said, for agencies to focus on how AI might be used to benefit a client’s business. “It’s less about the input and more about how incremental business returns are being delivered,” he said.