Quick answer
No. Faster delivery does not mean cheaper delivery, and assuming it does is costing agencies money. The AI tools that speed up the work are not free, and many are getting more expensive to run. Goldman Sachs expects token consumption to rise 24-fold by 2030, and Gartner warns that even a 90% fall in the price per token won't lower total bills, because agentic tools use far more of them. The deeper question AI raises is what your agency actually sells, and whether your model still works. This article covers why clients expect lower fees, why AI isn't the bargain it looks, and what to do about it.
I had a conversation recently with Brian Kessman, founder of Lodestar Agency Consulting and author of the VoxComm report Redesigning the Agency Value Model. We got onto the question I think many agency owners are asking right now.
Not "which AI tool should we use?" Not even "how do we do the same work faster?" The harder one is: does the way our agency makes money still make sense when AI changes what clients are expecting? And if it works today, will it still work in two years time?
A lot of owners I talk to are asking themselves exactly that.
Why do clients think AI should make agencies cheaper?
Because they are using the same tools you are. Your client can draft a brief, rough out a concept or pull together a first version in minutes, on their own laptop, for the price of a subscription. So their assumption is if AI did some of this, it must have been quicker, so it must have cost less, so the fee should come down.
Brian sees this playing out in two very different ways. Production-led agencies are getting squeezed hardest, because procurement teams expect any time saved by AI to come straight off the invoice. The more strategic agencies feel less of this pressure, because they are clearer about the value they create beyond the hours.
That gap is the whole story. If your agency sells hours, AI is a threat. If it sells outcomes, AI is an advantage.
Is AI actually cheaper for agencies to run?
This is where the assumption falls apart.
AI is usually charged by the token, a fraction of a word. Draft an email and it costs a sliver of a penny/cent. The trouble starts when agencies move to agentic AI, where the tool runs multiple steps, checks its own work and completes whole tasks on its own. Token use doesn't creep up. It explodes.
The last couple of months have made this very public:
An Axios report found AI compute costs are now overtaking employee salaries in some tech teams, with an Nvidia vice-president saying the cost of compute for his team far exceeds the cost of the people.
Uber burned through its entire 2026 AI budget in four months, with individual engineers running up between $500 and $2,000 a month in tokens.
Microsoft cancelled most of its internal Claude Code licences ahead of its financial year end, partly to control those costs.
Goldman Sachs expects global token consumption to rise 24-fold by 2030.
Gartner predicts the price per token will fall by around 90% by 2030, but still warns total bills will rise, because newer tools devour far more tokens and frontier model providers are unlikely to pass the savings on to the businesses who are using it.
Why does this matter to you? Because your client may assume your delivery costs are falling. In reality, as you embed more capable AI into the work, your costs may be flat or climbing. Knowing the real cost of your tech stack is becoming the difference between defending a fee and discounting it.
What is "value blindness" and why does it cost agencies money?
Value blindness was the phrase from my conversation with Brian. It means the strategic value an agency is already delivering and giving away free, because it has never named it, packaged it or charged for it properly.
Think senior judgement on a tricky brief. The instinct that stops a campaign going out wrong. The pattern-spotting that comes from twenty years in the category. None of it shows up as a line item, so clients assume it's free. And over time, so does the agency.
Brian's point is that transparency helps. When you show clients what AI delivery actually involves, more senior input, more review of the outputs, more decisions and more technology cost, it punctures the myth that AI means instant, effortless, cheap work.
How should agencies rethink what they sell?
The VoxComm guidance Brian wrote makes one thing plain. This is a "what do we sell" conversation before it is a pricing one. As Brian puts it, pricing is the last step.
The report maps four stages agencies tend to sit in:
Busy by Design - revenue tied tightly to effort and hours.
Scaling with Strain - growth simply adds complexity and pressure.
Expertly Undervalued - the expertise is strong but under-monetised.
Distinctly Scalable - revenue decoupled from headcount, pricing aligned to outcomes.
Most owners I speak to recognise themselves in one of the first three. The aim is to build around expertise and impact rather than activity and inputs, and to replace one-off projects and vague "full-service" labels with repeatable, productised solutions priced on the result.
What does this mean for account managers?
This is the part that lands on account managers' desks. You are the one in the room when a client says "but AI must have made this quicker." You are the front line of the fee conversation, whether you signed up for it or not.
A few things to start doing:
Learn the real cost of your tech stack. You can't defend a fee you don't understand. Ask finance or ops what the AI tools on a typical project actually cost to run.
Name the invisible value. When you scope or review work, point to the senior judgement, the decisions made and the rework AI created, not just the deliverable. Make the thinking visible.
Stop selling hours in client conversations. Talk about the outcome the client is buying. It's a harder conversation and a far better one.
Summary
Clients increasingly assume that because AI is faster, agency fees should fall. The reality is often the opposite. Token-based AI, especially agentic tools, can cost as much as or more than the work it replaces, and those costs are forecast to climb. The deeper issue is the agency model itself. While revenue is tied to headcount and hours, AI looks like a threat. Brian Kessman's VoxComm report argues agencies must first redefine what they sell, moving from effort to outcomes, before touching price. For account managers, that means understanding the real costs and naming the value clients can't see.
FAQ
Does using AI mean agencies should lower their fees? Not automatically. AI tools, particularly agentic ones, carry real and often rising costs, and faster delivery still depends on senior judgement and review. Lower fees only make sense if the value delivered has genuinely dropped, which is rarely the case.
Why are AI costs rising instead of falling? Because newer agentic tools complete multi-step tasks and use far more tokens per job. Goldman Sachs expects token consumption to rise 24-fold by 2030, and Gartner warns that even a 90% drop in the price per token won't lower total bills.
What does "value blindness" mean for an agency? It's the strategic value an agency delivers but never names, packages or charges for, such as senior judgement and hard-won experience. Left invisible, clients assume it's free.
What are the four stages in the VoxComm value model report? Busy by Design, Scaling with Strain, Expertly Undervalued, and Distinctly Scalable. The first three tie revenue to effort and hours. The last decouples revenue from headcount and prices on outcomes.
How can account managers defend agency fees in the AI era? Understand what the agency's tech stack actually costs, make the invisible senior work visible to clients, and steer conversations towards the outcome the client is buying rather than the hours spent.
This came out of my conversation with Brian Kessman on the podcast, well worth a listen if your agency is wrestling with pricing and AI How are you handling the "AI must have made this cheaper" conversation with your clients? I'd love to know.
