Hacker News Reader

Comment by _Understated_ | Hacker News Reader

Although I've asked before, a great many times, can anyone explain to me why, if I search for "Carburetor for Ford Fiesta", just showing me ads for carburetors with no intrusive data mining isn't enough for companies these days?

I would, in all likelihood, click on something like that.

It's easy to infer "oh, this guy needs a new carburetor for his Ford Fiesta" from my search. What possible additional benefit is there to mining the ever-living hell out of me further?

Who is paying for this shit?

Is advertising and tracking a massive ponzi scheme?

Edit: I should point out that although the content piece doesn't explicitly mention ads, the creator is an ad-man from Google and I'd put money on it having ads anyway... hence my comment.

View on HN

Do you live in Minnesota? If so, you might prefer to buy from somewhere online-but-local.

Do you normally buy the branded dealer part, or a generic? That would impact which ad you want to see/click.

Will you be wanting to buy this carburettor outright, or would you prefer to pay $1 per month for the next 20 years for it?

Will you also be wanting fitting instructions, a guarantee, tools etc. etc.

The answers dramatically impact what kind of products you might buy. Yet there are only slots to show a few companies/services, so unless Google correctly guesses, you will think the results are useless. They need data to make the right guesses.

You make some good points but is the trade-off worth it.

Finding out where someone is from these days is fairly trivial using the IP address: the shortage of IPV4 means that it may not be 100% accurate sometimes though.

So, using your example, working out that I am in Minnesota should be easy enough. At that point, I'd wager that people will either buy from Amazon (sell many cheap/fake/knockoff stuff that does the job) or would buy from Minnesota Bob's Local Car Parts Emporium... no data to back that up but I'd imagine either of those scenarios would account for a large number of people - My own preference (for the most part) these days is local business with a delivery option (not Amazon any more due to reasons).

With your example, there is a whole bunch of extra options in there that I would put serious money on that no one can tell what I want from mining the shit out of me. Not Google, not anyone.

Even with all the data they can muster, there is no way that they can make the decision that "this guy wants ad number 15, not ad number 16" with enough accuracy that the people paying to have their ad shown are actually getting a return.

With context-sensitive ads, I'd wager there'd be a high chance that someone in Minnesota searching for carburetors that sees an ad for Minnesota Bob's Local Car Parts Emporium would want to buy from him... again, no data to back it up, but equally, I have yet to see any substantive data that mining actually results in an appreciable, measurable increase in sales... none!

I still believe it's all (mostly) a ponzi scheme.

You're misunderstanding a significant part of the equation.

Also, companies measure their advertising success in their conversion rate and A/B test vs broad based ads that target more widely. They have the opportunity to advertise super broadly if they so choose, and can measure that impact.

By narrowing the targeting, advertisers can advertise to "just" the people who are over a certain level of confidence interval, and not advertise to the people who don't give a damn.

Ad targeting is actually an exclusion business, it's about companies wanting to rule out as many people as possible, reducing their ad spend, who are then picked up by others.

Companies just don't want to buy junk ad slots. There are actually not all that many ad slots and the cost can get moderately steep for broad base, low conversion ads.

I get what you are saying, however I do not believe for one second that there is a measurable change in sales (for the company that places an ad) for an ad like the one I mentioned above vs the mega-targetted, data-mined tracked-to-hell-and-back ads that companies like FB and Google tell you about.

Sure, a scattergun approach would be pointless but someone in Minnesota searching for carburetors has provided enough data in that one search that I don't believe targetted, mined, tracked users will be more lucrative.

I've asked on here before if anyone had a link to research that wasn't commissioned by an ad company about targetted and mined data and no one has provided one.

I'd love to be proven wrong but I don't believe targetting, mining, whatever you want to call it, moves the needle enough except for the ones at the top of the tree selling the belief (Google, FB etc.)

You'd be wrong - lots of data and very precise targeting can multiply the effectiveness of some kinds of ad campaign by 10x. That's the kind of change that turns an unviable business into a viable one.

I suspect that if ML based targeting gets as good as it theoretically could (ie. The ads predict what I need before I even think I need it), there would be another 10x gain.

> That's the kind of change that turns an unviable business into a viable one.

What, like going from 1 in every billion people clicking on your ad to 1 in every 100 million?

I'm sure that in some niches there is going to be a particular data set that yields mega results for someone. But I'm talking as a general concept, that the whole advertising industry is built on total lies all the way down and that all the data mining hardly moves the needle compared to context-related ads like my example.

I suspect that no one wants to know if it actually works or not. It'd be like pulling on a thread that everyone can see, plain as day, but no one wants to touch it as there are hundreds of billions of dollars on the line if they do. That, plus people are employed and have livelihoods built around ads so they won't jeapordise that.

I don't buy it. I still think it's all bullshit.

But let's say it's true. There still remains the fact that many don't want their data in anyone's database in the first place. It's an industry built on gathering data without consent - and before you say there are opt-outs and all that, the industry built itself up long before any rules came into effect.

Those are good points, but:

I really doubt Google (or even worse, Facebook) segments people on any dimension like those, except for location. There is nothing on their analytics or ads buying interface that even hints at those, and it really doesn't show at their choice of ads.

Somehow, when Google did what the GP was talking about, their ads were incredibly useful. People would google stuff just for the ads (I was one of those, but I've met more people like that).

Google uses ML models to predict which ads are most likely to be clicked.

The ML could easily be taking into account any of those things. There is no way to know how it decides which ads are most likely to be clicked - ML is a black box.

Not to argue that this is a good thing, but to expand: the UX of having your ad purchaser decide how best to target a demographic is really bad, so Google abstracted it away.

For example, you might not know that only homeowners in Minnesota like to buy cheese, but you want to sell cheese to Minnesotans, so Google will automatically target your ad to homeowners. It doesn't need to know the former fact explicitly either; it's just a large correlation matrix that they learn on the fly.

It's all machine learning, they don't manually segment people. They just feed in a bunch of tracked features and the AI says what you are most likely to choose.

If I'm a new grocery in your city, what keywords should I put my ads under?

People don't usually search for "groceries around me" more than once, and if you don't happen to ever drive by the storefront, you'll never know it existed.

Sure there are other ways to advertise, but they are going to be less effective than what Google can do today

Reply on news.ycombinator.com