19 comments

[ 3.9 ms ] story [ 53.7 ms ] thread
I am surprised that this blog post went through legal including this detail (https://archive.md/ZD1zc): In the next paragraph they write the fraud rate "on the order of 1 out of every 1,000 payments.", so their false positive rate is as high as the fraud rate.
I wonder what makes false negatives so costly for them, that they're willing to accept so many false positives.
Chargebacks are very costly, and it isn't all that high a false positive rate at all given what they are doing (processing payments). In fact some of their competitors would be very happy with that rate.
A false positive costs them nothing but the fees on a single transaction. A false negative costs the entire value of the transaction plus legal and administrative costs.
And possibly their merchant account if there are enough of those.
Reputation. Have you not seen that post that pops up every three months or so about how some startup had their funds "stolen" (whether true or not) by Stripe?
I would have thought, those are the result of false positives, i.e. when fraud is "detected" where there (at least allegedly) is none.
I would be very careful interpreting those stories without having more than just one side.
The post is about a project called Radar. Do you believe this is what is responsible for those complaints?
That is definitely not the title of the original article.
This is a horribly editorialised title. In my view (as a e-commerce store owner) a false positive rate detecting fraud across their whole network of only 0.1% is very very low. If I'm reading it right, that's 1 in 1k blocked transactions are incorrect, not 1 in 1k all transactions!

> The challenge is compounded by the fact that fraud is rare—on the order of 1 out of every 1,000 payments.

So 0.1% of 0.1% (0.001%) of transactions are incorrectly blocked.

Yes, but that sentence could have been worded much more clearly.
The article says (2nd paragraph, last sentence):

> Out of the billions of legitimate payments made on Stripe, Radar incorrectly blocks just 0.1%.

I seem to parse this differently, to me it reads as if 1 out of 1000 transactions are incorrectly flagged. I am not a native speaker though. Could you explain how you arrive at 0.1% of blocked transactions?

I just saw your edit, but my question still stands.

The way I parse that is if a transaction is blocked the chance of it being an incorrect block is 0.1%. So out of 10000 transactions blocked 9990 would be correct blocks and 10 would be incorrect blocks. I agree it is confusing though.

What is interesting is how they determine the incorrect blocks but the article doesn't really expand on that, it's just about one aspect of their fraud detection pipeline.

I'm missing some information how they count incorrectly blocked. If something fails for unexplainable reasons, as a user I simply switch to a different payment method or a different supplier.
And as a merchant: I'd be fine with that. Because the downside for a lost sale is manageable but too high a fraud percentage and my merchant account gets killed and that would be the end of my business. So merchants and PSPs will err on the side of caution.
> Out of the billions of legitimate payments made on Stripe, Radar incorrectly blocks just 0.1%.

Or rephrased:

"Stripe wrongly blocked millions of payments." What a headline!

Let's see which news outlet will use it.