Hacker News Reader

>I wouldn't be surprised to find that 90% is actually very close to reality

The FTC press release[1] says "for many offers, almost a third of consumers who applied were in fact denied". That's quite a ways off from 90%.

[1] https://www.ftc.gov/news-events/news/press-releases/2022/09/...

edit:

>Also, Credit Karma gets paid for successful conversions, and maybe ad placement? It doesn't seem like misleading someone got them any profit.

This is incorrect because misleading causes more people to apply. The people who are coaxed into applying through deception have a non-zero chance of turning into a successful conversion, which makes credit karma money. For instance, if people interested in a credit card, but they're not certain that they'll get approved, so they end up not applying. If there are 100 visitors in that situation, and credit karma lied to them, then they should expect to get 66% (based on the actual approval figures from the FTC) successful conversions (ie. profit).

I wonder if there is some selection bias going on. One hypothesis could be that the people who are financially more responsible (and thus are more likely to be approved for a new credit card) are also the people who are unlikely to apply for a new credit card just after seeing an ad in a website. Thus, credit karma ad clickers self-select to a lower success rate than what their model predicted.

I'm honestly surprised more weren't denied given that folks that have no credit for good reason would be the most likely to apply for credit.

If a third of consumers _should_ have been denied because they were not credit-worthy, then there were no errors. You need to look at the false negative rate, not the absolute number of people denied for financial products.

There are no "false negatives." The models implemented by the card issuers are the literal source of truth as to whether a consumer does or does not qualify for the card. Anybody who applied and was rejected did not qualify by definition.

I think the misleading part was representing it as that person specifically had 90% chances, even though 90% may be aggregated probability and this person's chances based on personal history may be much lower. Given CK does have access to personal history, one might easily get an impression that the prediction is a precise calculation based on it and thus imply more accuracy than warranted.

> The FTC press release[1] says "for many offers, almost a third of consumers who applied were in fact denied". That's quite a ways off from 90%.

The statistics don’t play out that easily - their models may be accurate in aggregate but better or worse for specific offers.

This still seems a bit dubious—-90% of people who view the ad being qualified doesn’t mean in any way that 90% of applicants will be successful. If you have more applicants from that bottom 10% pool who click on the ad to proceed, then it would be statistically very easy to end up with a reject rate much higher than 10%.

If you reject a third of applicants, the 90% figure is obviously misleading even if technically true in some sense.

It didn’t say 30% were rejected - it said on some offers as much as 30% were rejected. Again, the models can be totally accurate in aggregate but due to random variation or low sampling show as inaccurate on one specific offer, that is just how statistics work.

> CK has inside views to the models these companies use, and wouldn't be surprised to find that 90% is actualy very close to reality.

or the basis as is obvious is that this was a fraudulent claim. these things get researched thoroughly before they levied

I read it more as "using uncertainty" is a dark pattern and to stop it even if it's accurate. 10% of users a company with that many is a pretty significant impact to people.

I reread the article again, and it seems to me it was "pre-approved" that was the issue.

Regardless, it's good the behavior was noticed and stopped.

I doubt it - CK is funded and the FTC is underfunded. Trusting the government over private in this scenario could be perilous.

And yet it turns out the proportion of applicants not qualified was more more than 3 times higher than CK claimed, and their assertions that applicants were pre-approved turned out to be a flat out lie.

> wouldn't be surprised to find that 90% is actually very close to reality

I’d be blown away if this is the case. Credit models are carefully guarded. They’re also expensive to run. If Credit Karma could approximate the pricey model with open-source data, they’d have been bought by a bank, not a tax company.

Credit Karma has a platform for running models against their dataset, and appears to buy everyone's daily data from Equifax.

Also, Intuit is much more than a tax company now.

> Credit Karma has a platform for running models against their dataset, and appears to buy everyone's daily data from Equifax

There are a lot of people doing this. For the aforementioned reason: a cheap approximation of a credit score is valuable. Before the financial crisis, VantageScore was developed by the credit bureaus to disintermediate Fair Isaac. This work continues, and Credit Karma is far from unique in its approach or data.

But CK isn’t predicting whether you will default, but whether another lender will regard you as a good customer (where default risk is just one factor).

That’s a different problem than simply evaluating credit risk, and is much less researched — so not the same as the highly guarded credit models.

Unless they worked out a deal for impressions and not conversions. Handwaving the 90% behind some asterisk.