A lot of new “standards” are being thrown around the ad-tech industry these days, causing the regular amount of confusion with practitioners. To list a few:
- MCP
- AdCP
- ARTF
- agent2agent
- UCP
- CoMP
I don’t want to go explain each one (yet), but for any confused readers, I would propose a simple framework for how to think about the potential contribution of LLMs and agents in our day to day work. It is all just different flavours of automation of various tasks in the value chain.
Type 1 – generative AI to create creatives. Easy to understand. You use this all the time now when you go to chatgpt or nano banana for ad copy / image generation. These assets then flow through our platforms to be displayed to users everywhere. Agents here might help creating more, better, faster, but until someone decides to start including ‘generative creative’ inside the ad itself, we are in control.
Type 2 – budget planning / campaigns setup / insertion order / deal negotiation phase. AdCP is proposing to use Agents instead of humans to do these mechanical tasks, and it’s not an unreasonable wish, but probably an impossible task. In a previously perfect world, to achieve this dream, industry naturally centralised to a handful of buy and sell side platforms that everyone used, and then those platforms integrated with each other, to give as streamlined workflows as possible by directly sharing data. In agentic workflow, in theory, we could get a more decentralised ecosystem which would result in more innovation. But somehow I suspect that we will learn we actually don’t need that and that tight integration had many other benefits, for platforms and for users alike.
Type 3 – individual ad impression decisioning and serving. Real time bidding became the norm lately, and the bidding algorithms were probably the largest deployment of AI before LLMs for years. There is nothing agentic in this layer, but different types of neural networks that make smart decisions very fast. What we call “protocols” in this layer is just format of data exchange to make decisions of algorithms even better / faster. This is for ARTF, UCP, CoMP proposals to advance.
Type 4 – Campaign data analysis and reporting. Every ad platform is launching a chat interface to its reporting layer, and opening it up with MCP. I hear many excited marketers finally being able to orchestrate data from multiple places into a meaningful summary for themselves to feel in control. While they do that in tools like Cursor (yes, weird), I haven’t seen a good proposal for agentic exchange of data between platforms themselves. Why couldn’t I optimise my campaign on Teads based on data of my campaign on Meta directly, without the need of coding something. The two platforms should just talk to each other.
This was a long way of saying “agentic advertising” doesn’t mean anything, you should always know what problem is the proposed standard trying to solve, and for whatever reason all the press releases are really terrible of explaining their place in the value chain.
Leave a Reply