[nl]

I work in a related area (prediction, though not in the AdTech market) and keep myself up to date on the literature.

Imagine you've traversed states A, B, and C and are predicted to be moving toward D. If state F or Z is more valuable (and can be arrived at from state D), then perhaps through several months of training you can be led to it instead.

Nothing like this exists beyond very general models. There are some mood-state models, but they are short term (people argue if hourly data is too sparse for them to be useful).

The general models are roughly what you'd expect: if you are 18-22 you are likely to be a student, 55+ considering retirement. I've never seen any research on pushing people along paths, beyond things like education ads trying to get people to take courses, job ads trying to get people to change jobs and dating ads trying to get people to change partners.

Whilst general models maybe possible, my suspicion is that there are too many confounding factors for them to be very useful.

[tomjen3]

Based on the ads I have seen Google hasn't yet moved from "this guy is looking for an apartment so lets show him ads for the same apartment deal site for months" to "this guy has looked for a apartments for a short time and now doesn't, lets show him ads for some curtains" so it seems even the best predictors aren't very good yet.

[thewarrior]

But surely as more and more data is gathered over the next decade , this sort of thing could become feasible ?

For eg: I know for a fact that FB is betting very heavily on travel advertising. FB wants to be the go to place for travel companies to advertise their products , so FB has an incentive to make people travel more.

They could do this prominently highlighting when people travel to a certain tourist spot etc...

[SolaceQuantum]

It wouldn't be just about gathering data, but also developing the algorithms that are capable of that specialization to each person. No matter how good your data is, the lack of an algorithm to examine the data and recognize applications for each unique person to manipulate behavior using a strategy that must recognize it's own applicability at that moment is going to be the huge hurdle.