There is a calculation agencies rarely say out loud. It goes like this. Someone used to write a proposal in three days. With a well-configured tool he does it in four hours. The agency bills by the hour. So it has just cut two and a half days of its own revenue.
That is not an edge case. That is the basic mechanism. If you sell time, you sell less as soon as you get faster.
I ran a development agency for nine years and a social media agency for nine years after that, the second one with seventy people at its peak. Both billed mostly by the hour. Back then the calculation never struck me as a problem, because nobody ever put it on the table. Efficiency was something we praised in process meetings and never found again in the numbers.
Why AI pilots stall right after the pilot
When I ask today why an AI project went quiet, I hear three answers. Data protection was unresolved. The quality was not good enough. There was no time for it.
All three are sometimes true. None of them explains why so many pilots in agencies succeed technically and still never reach daily operations. The pilot proves that the thing works. It also shows what it costs to take it seriously. The process it speeds up is the same process the agency has been billing.
Nobody frames this as a decision. It happens more quietly than that. The project gets no follow-up meeting. The license lapses. The colleague who pushed it moves on to something else. What remains at the end is not a resolution. It is an absence.
Here is the uncomfortable part. Under hourly billing, an agency has a structural interest in not raising its own efficiency. That is not laziness. The arithmetic simply works out that way.
What this looked like at Granny
An example from our own numbers at Granny, the social media agency I co-ran from 2016 to 2025. I only read it properly years later.
Strategy and creative ideation were always sold at a fixed fee. That was standard practice and the thinking behind it was sound. The client buys an idea, not attendance. The trouble was that nobody could say when the idea would land. Sometimes after ten hours. Sometimes after twenty. The price was fixed in advance. The effort was not.
So the creative part often ran at a loss. We made the margin back further down the line, in production. It worked, and because it worked nobody questioned it.
Then AI came into the ideation process. The main thing that changed was not speed. It was variance. More approaches reached the table sooner, the creatives could close the process earlier, and “ten to twenty hours” turned into a range you can actually price against.
Economically that is the bigger effect, and it usually goes unnoticed. On a fixed fee you price against your worst case, not against your average. Narrow the spread and you gain margin without working a single minute faster.
Because that service was sold at a fixed fee, the gain stayed with us. Nobody had to discuss a discount with the client. Nobody had to justify an efficiency. The same improvement inside an hourly billed process would have been a revenue cut.
The part I only saw later
If the creative work is permanently cross-subsidized out of production, then the profitability of the agency rests on production hours.
Those are exactly the hours that automation clears out first and most thoroughly. Reporting, revisions, finished artwork, recurring production. Those are the processes where the tools are furthest along today, and where procurement is quickest to run the numbers.
That leads to an equation less comfortable than the one at the top of this article. Speed up production and change nothing else, and you drain the pot that used to cover the loss on the creative side. The loss stays. The subsidy goes. That production became more efficient is no consolation, because under hourly billing that gain ends up with the client anyway.
This is not an argument against automation. It is the reason the order matters.
Procurement is already doing the math
As long as the calculation stayed inside the building, you could sit this out. That is changing now, and it is changing from the other side of the table.
Buyers know that copy, concepts, analysis and code come together faster than they did two years ago. They will not ask which tools you use. They will ask why a line item still costs what it cost in the 2023 framework agreement. If you have no answer, you hand over the efficiency gain without ever having captured it. That is the worst of all versions. The saving goes to the client. The investment risk stays with you.
So the question is no longer whether you share the gain. The question is whether you get to choose where you share it.
What replaces the hour
The way out starts with a distinction. Sort your services by what you are actually selling.
Services that are a result. A landing page, a trade fair stand, a campaign concept, an analysis. The client is buying an outcome and takes no interest in the hours behind it. These belong on a fixed fee, and on a fee derived from the value to the client rather than from your effort. Get faster here and the gain stays with you. You earn a second time as the variance drops, because a narrower spread means a smaller risk premium for the worst case. This is the part of the business that AI genuinely improves.
Services that are availability. Support, advice, standby, the person who has known the client for four years. Here the client is buying access to people. Retainers are the right shape for that. This is also the part that better tools make more valuable rather than cheaper. Someone who can judge which of three machine-generated options will hold is worth more than someone who typed the option themselves.
Services that are pure execution. Reporting, adjustments, data maintenance, recurring production. This is the part that gets automated sooner or later, by you or by somebody else. Hourly billing is still honest here. It is also the part whose share of your revenue you should reduce deliberately instead of defending.
This sorting is not theory. It is an afternoon with your service list and a pen, and it is the precondition for any sensible decision about where an AI project is worth starting.
The order that works
The priority that follows strikes most agencies as backwards at first.
Your first automation project belongs where you bill nothing today. Proposals, pitch preparation, internal coordination, rework, reporting to yourselves. Those hours cost you money and earn none. Every minute you win there is margin straight away, and you have to negotiate the price with nobody. It is the one place where the conflict in this article never arises.
Second come the services you already sell at a fixed fee. There too the gain stays with you, and no contract has to be reopened. The ideation case above belongs exactly here. The lever is often larger than expected, because on fixed fees it is not the average effort that decides your margin but the premium you carry for the worst case.
Last comes hourly billed production, and only together with the pricing change for that same service. Doing the two separately is the mistake. Speed up production before you have moved from effort to outcome, and you are optimizing your own revenue downwards.
The change takes time. Framework agreements run their course and clients have habits. One to two years is realistic. That is why you start before procurement asks the question, not after.
What this means for your next quote
If you take one thing from this article, take this. On your next proposal, check whether you are selling time or an outcome. If it is an outcome, the hour count has no place in it, not even as a derivation in the small print. As long as it sits there, every improvement you work for is a discount you grant unasked.
At Granny we got one part right without noticing it was a pattern, and we missed the other one for years. That is why I now ask about the pricing model first and about the tool second. In that order the result holds.
This article is not legal or tax advice. Restructuring contract models calls for proper legal counsel.
Further reading
- Who this advisory is for, agencies, studios and service firms in project business
- Assessment and rebuild with numbers, the two formats that build on this article
How this article came about: eighteen years of running my own agencies, an interim COO mandate in an agency group of more than a hundred people, and the cost calculations I have seen in client projects since.
About the author
Philip Hohn advises agencies, studios and service firms whose business is built on projects. He is Managing Director of HBC Hohn Business Consulting UG and runs several of his own companies and projects, including Edura Akademie GmbH, an AZAV-accredited training provider for AI skills. Before that he was co-founder and managing director of the social media agency Granny, with seventy people across four offices, and founder of a development agency with fifteen engineers.