Will agencies survive AI?

Plus Josh Chasin on the future of measurement

Eric put a quote in the podcast notes this week from Sam Altman predicting 95% job loss in the ad agency world. That got my attention, so I wrote about it. Plus measurement veteran Josh Chasin joined the pod to talk about…measurement.

Enjoy Marketecture? You should also subscribe to:

Justify Your Existence is Back

It’s back, our Justify Your Existence interview format with small ad tech start-ups. If you want your start-up featured, reach out.

This week we have two new interviews, both of which are trying to solve the consumer data problem.

In the face of signal loss and the desire for more deterministic IDs, Licorice Solutions is giving incentives to consumers to provide their login data to publishers.

Crosshatch gives app developers the ability to personalize their content for different users in a privacy protected “DMZ”. We conducted this interview before the company launched, when it had the name Olympus, but they can now be found at crosshatch.io.

Both interviews are free for a week before going behind the paywall.

Reminder: Vendor Interviews are always free for 1 week, then are subscribers-only. Subscribe for only $39/month.

Podcast: Josh Chasin on Measurement

Josh Chasin has been measuring stuff at places like Comscore since before you were born. He’s coming off a meaningful stint as Chief Measurability Officer at VideoAmp that brought them to the table as a major player in the future of TV. He comes on the pod and gives some perspective on how the industry moved on from panels, how hard it is to beat Nielsen, and the future prospects of the major measurement players.

Listen to the pod now:

Will agencies survive AI?

ad agencies

95% of what marketers use agencies, strategists, and creative professionals for today will easily, nearly instantly and at almost no cost be handled by the AI — and the AI will likely be able to test the creative against real or synthetic customer focus groups for predicting results and optimizing. Again, all free, instant, and nearly perfect. Images, videos, campaign ideas? No problem.

—Sam Altman, CEO of OpenAI (source)

First principles: AI automates knowledge work

If you’ve used ChatGPT, Gemini, Perplexity, or any of the modern AI tools, you know how powerful they can be in automating rote knowledge tasks. “Write me an introductory paragraph”, “create an image that shows dogs thinking about math,” or how about “how many global employees do publicis, wpp, and dentsu have?”. Here’s the answer Perplexity gave to that last one:

  • Publicis Groupe: Employed 98,022 people worldwide in 2022

  • WPP: Employs around 100,000 people globally as of 2022

  • Dentsu Group: Employs approximately 65,000 people in 145 countries worldwide

That’s a lot of headcount. You don’t have to believe in AGI or that an AI could be the next David Ogilvy to understand that many, many of those jobs are replaceable with AI.

Examples of tasks that AI is already or (will be) able to complete better than humans:

  • Resizing images

  • Recutting videos

  • Creating images or videos out of pre-produced assets

  • Creating storyboards for ads

  • Creating media plans

  • Optimizing media plans

  • Analyzing and reporting results

  • Understanding consumer sentiment

Like I said before, that’s a lot of headcount.

Second order effects: Some roles disappear entirely

AI is a lever that will enable fewer humans to do tasks for clients with more efficiency and effectiveness. But some tasks will also be automated out of existence. We’ve already seeing the big platforms move in this direction, with PMax replacing most of the activities humans used to do within Google Ads and the Google Display Network.

While I’ve been wrong before about the prevalence of “in housing” among marketers, one has to imagine that when a task that is normally owned by an agency becomes 99% automated by technology, it will make sense to contract and operate that technology directly rather than pay for overhead to manage it.

Candidates for tasks that are subject to complete automation are those that are data rich and repetitive, such as media optimization and attribution.

Side quest: What about agency data?

One side quest in this conversation is whether the agencies have enough proprietary data to defend themselves and gain an advantage in the age of AI. We know that the big holding companies are investing in AI, with Publicis’ CoreAI ($325 million over three years), and WPP’s iQ ($317 million annual investment) being two examples, while Dentsu and Omnicom like putting out lots of rando press releases. But these numbers pale compared to the investments of the big tech giants.

Investment dollars aside, who is in a better position to automate media buying and optimization that the companies that have all the data on how this has worked for the past fifty years? I asked Josh about this on the pod and he was skeptical, saying “does knowing the ratings for All in the Family twenty years ago have any value today?” Good questions.

Third order effects: Orchestration, creative, and strategy

It seems like a solid bet that agencies will be much smaller than they are today, and that some of their lower value-add activities will be eliminated, automated, or in-housed. What’s left?

We’re a long way from AI that can create the next “Coke is it” or Apple’s 1984 ad. Strategy and creative will remain the domain of humans, for some time. This is probably the 5% that even Sam Altman thinks will continue.

Orchestration is always important. Technology requires resourcing to be useful, and while AI solves many problems, it creates new ones. There will always be a role for smart people who can solve problems using technology, and this is where the need for global agencies will shift.

Another possibility, that Eric raises on this week’s pod, is that you may see extremely small groups of talented people use technology to create much more leveraged agencies. Could a three-person firm handle media and creative for a $100 million new product roll-out?

Will the cockroaches survive this time? Let me know what you think.

Reading list


or to participate.