What Happens When an Ad Tech Company Goes All-In on AI? | CEO Nicolas Schueller

Well, I would say your first question is very hard because the answer to the second is that JP is not an agent.

JP is not a chat bot.

JP is an infrastructure.

JP is sold to our clients as your AI infrastructure.

It doesn't have to be the only one within a client ecosystem.

But JP is an AI infrastructure or an AI toolbox.

So what was very important to us when launching JP was really to create the conditions and the basis that help and that are required to build efficient agents and efficient AI for our clients.

As you know, James, the first thing that is absolutely critical is the data right because uh while you know llms and agents in general are renowned to be able to cope with very unstructured data When it comes to numbers, when it comes to transactional data, there has to be some teaching to your LLMs that has to be done if you want to make sure that The reasoning of the LLM is done on data that the LLM understands, right? That the LLM can interpret and then that the LLM can analyze to provide outcomes.

So materially,

this

has led us

to work for almost nine months on the JPE semantic model.

Okay.

And here you get the idea now of, you know, this is an infrastructure because obviously, and I'll come to that later, the semantic model is not used exclusively for a chatbot.

It's used for whatever we want to build with JPEG.

So just one example, go to your cloud and ask by uploading, you know, a few reports of SSP to create a view of your revenue with one of the whole codes.

The concept of a whole code is totally understood by a generic LLM, but to map it with your actual data, You know, at the beginning we tried.

This does not end well.

This does not end well.

Same thing for transaction types.

Same thing for the different formats.

The differences between CTV and display.

The concept of app versus web.

This all seems super obvious, but what is expected from us What is expected from you when you consult James is to be 100% accurate, right? Is to be accurate and reliable, right? So you cannot leave any open corner for whatever interpretation to an LLM, because at the end of the day, this is going to come back through your clients.

They will say, Hmm, this is not exactly what I expected.