AI has moved from experimentation into everyday agency operations, and the bill is becoming more complicated. The days of treating AI as a single software subscription are fading: finance leads now have to account for seats, consumption and the people doing the work.
Three cost buckets on the AI bill
Across agency conversations, the spending breaks into three areas:
- Platform subscriptions: enterprise access to tools such as Claude, ChatGPT and Google’s AI suite.
- Token consumption: metered usage that varies by model, task and employee behavior.
- Human capital: staff salaries, which still dominate agency P&Ls even as tool costs rise.
Most agencies are funding AI from existing technology budgets rather than creating a separate line item. Ramp research indicates about 31% of companies now spend more than $10,000 per month on AI, a sign the category is becoming formal.
Different agencies, different billing logic
Markacy’s primary AI cost is enterprise Claude access for its 25-plus team, with ChatGPT added for certain employees. Co-founder and co-CEO Tucker Matheson said the agency treats AI as a variable cost instead of giving it a dedicated budget.
Crispin, owned by Stagwell, uses around 100 enterprise AI tools across media planning, social listening and influencer discovery. Chief transformation officer Freddy Dabaghi said the agency prefers flat, fixed fees with annual renewals to keep monthly costs predictable.
Dept runs Claude, ChatGPT and Google to accommodate different client preferences. Jonathan Whiteside, global EVP of technology, said those three platforms generate the largest token costs. AI still remains below a double-digit percentage of Dept’s budget.
More usage, less predictability
The cost pressure follows broader adoption. Forrester research found that 74% of agencies use generative AI to summarize documents and communications, and 70% apply AI to research and competitive intelligence. As usage shifts from pilots to production, consumption-based charges introduce more volatility.
At Dept, an internal monthly AI Tool Committee reviews tools for approval, testing, certification or removal. The agency is also developing a system to guide staff toward the most appropriate model for a specific task or client.
Crispin is taking an annualized approach to contracts and reviewing platform usage logs to cut subscriptions it does not need. Dabaghi said enterprise agreements help the agency set clear thresholds, even where usage-based pricing exists.
Passing the AI bill to clients
Agencies are beginning to rethink who pays for AI work. Omnicom CFO Phil Angelastro told the Goldman Sachs Communacopia and Technology Conference that the holding company is building a model to manage token costs, measure them and get reimbursed, with a longer-term shift toward outcome-driven revenue.
One agency executive, speaking anonymously, said their agency is experimenting with billing token costs to clients as materials, separate from pre-set labor costs. Another test combines embedded token costs with human costs in the same client billing line.
What marketers should watch
For agency operations leads, AI now needs the same discipline as media spend: usage monitoring, vendor review, model selection and a clear client billing policy. A simple first step is an audit of every AI tool, its pricing basis and the client work it supports. From there, leaders can decide whether to bundle AI into retainers, bill it as materials or absorb it as a productivity investment.
The human layer still matters. As Matheson put it, “I still need people who can get on a phone with a CMO and be impressive.” AI costs are rising, but the value depends on judgment and client understanding.
Source: Digiday
