So, um, let's let you jump in and, uh, kick us off on this first topic, why measurement models fail without real data validation.

Yeah, for sure.

So this idea of real data validation, I think, is sometimes, um, a little confusing to people because they're like, "Well, I, you know, I log into the Google Ad platform, and I see real data there," right? But what do we mean when we say real data validation? The sort of philosophical concept that I think is really important to understand when it comes to marketing measurement is that the true incremental performance, like, the true return on investment of investment in some marketing channel is unknown and unknowable, right? So, like, if I invest a million dollars in Facebook, that will generate some amount of revenue for me ranging from zero to $100 million.

There is some true amount of additional revenue that that's driving for us, but we don't know what it is, and we can't know what it is.

Um, and so when we think about the right way to do marketing measurement, there's sort of two points that, that I think are important to keep in mind.

One is that you need to be thinking about this idea of incrementality, right? If not for this additional marketing investment, what would have happened otherwise, right? So you don't just wanna say, "Oh, you know, a bunch of people clicked on our ads and then purchased, and that means those ads caused the purchase." That's not necessarily true.

A lot of people could click on an ad who were going to purchase anyway.

Can you kick us off on that, Michael?

Um, but also what you think of this idea of outcomes measurement being something actually more higher up in the process, more around guiding optimization, and not just being something you do at the end of a campaign.

So really interesting questions here.

I think Meta and Google have been using machine learning and, quote-unquote, "AI" to power targeting, uh, to try to drive incremental sales for the entire time that they've been in existence, right? So this is not a new thing, and I think it's good.

One of the things that makes them such incredibly effective advertising platforms, and again, you can see it in how just empirically by how much advertisers spend on those platforms, is because they do a really good job at driving those incremental sales.

Um, you know, I, I talk about this a lot.

My partner is a translator, and she runs a very small translation agency, and it is very clear when she turns on ads on Google that she gets a lot more leads.

You can just feel it.

So they're also called not just foundation models, but generalist models, and I'm wondrous- wondering if you have a particular point of view on this idea of generalist models versus bespoke models.