Agency pricing is under pressure

A lot of agencies bought artificial intelligence tools to do work faster, then ran into a strange problem: the faster work got, the harder it became to defend charging by the hour or by headcount. Digiday reported on March 3, 2026 that agencies are now debating whether artificial intelligence “tokens,” the metered units that price model usage, should be billed to clients, absorbed as overhead, or turned into a new line item. (digiday.com) That hits the old agency model at its weakest point. A full-time equivalent pricing model charges for people and time, so if a team of 10 can suddenly do the same job with 6 people and software, the client sees fewer billable hours while the agency still has payroll, software, and revision risk to cover. (camphouse.io) The pressure is already broad, not theoretical. Camphouse, citing Forrester research in July 2025, said 75% of United States ad agencies were using generative artificial intelligence, up from 61% a year earlier, but only 6% were successfully monetizing it. (camphouse.io) That gap explains why “adding artificial intelligence” often does not expand margins by itself. If an agency keeps the same custom workflow, the same layers of account management, and the same client-by-client reinvention, software becomes another cost inside the old machine instead of a new machine. (emarketer.com) Industry forecasts have been pointing in the same direction. In Ad Age’s 2026 business forecast, Stagwell chief executive Mark Penn said artificial intelligence agents would start to challenge traditional agency staffing and pricing models as they move from support work into strategy and execution. (adage.com) The difference agencies are now arguing over is not “uses artificial intelligence” versus “does not use artificial intelligence.” It is “artificial-intelligence adjacent” versus “artificial-intelligence native,” meaning one firm bolts tools onto a labor business while the other builds delivery around reusable systems, templates, and software-first workflows from day one. (ai-native-agency.com) That changes the economics in a very plain way. A traditional agency usually earns revenue by selling scarce human hours, while an artificial-intelligence-native firm can spread one workflow, one prompt system, or one internal automation layer across 20 or 200 clients without hiring in the same straight line. (ai-native-agency.com) Clients are also getting harder to impress with “we use artificial intelligence” as a premium feature. eMarketer wrote in February 2026 that legacy agency models are cracking as clients automate more work themselves, which means agencies have less room to charge extra just for using tools the client can now access too. (emarketer.com) So the pricing fight is moving away from inputs and toward outputs. Agencies are testing usage fees, platform-style retainers, setup charges, and outcome-based pricing because a model that finishes a task in 10 minutes makes hourly billing look like a penalty for efficiency. (digiday.com) The catch is that output pricing only works if delivery is standardized enough to predict cost. If every client still gets a bespoke process, every scope still changes midstream, and every job still depends on senior humans stitching the work together, margins stay stuck close to the old agency range even with better tools. (anderscpa.com) That is why agency leaders keep coming back to delivery models, not software subscriptions. The firms likely to widen margins are the ones turning services into repeatable products, treating compute like infrastructure, and reserving human time for judgment, sales, and exceptions instead of every step of production. (digiday.com)