This is the big problem with Stripe's positioning in the payments food chain ladder.
Block every single fraudulent or suspicious transaction, and you're leaving obscene amounts of money on the table.
The amount of credit card fraud that goes unclaimed or is eaten by liability shift is huge, so if Stripe makes a product like Radar ACTUALLY WORK, they would be missing out big time.
I am confident Stripe's radar's shortcomings are deliberate and not simple bugs or design problems.
It appears they have no incentive for the product to be 100% effective and that would explain why Stripe Radar is billed per screened transaction, regardless of outcome.
We benchmark Stripe Radar against other pure play fraud fingerprinting solutions, and the difference is abysmal. The fact that Stripe claims to have seen 80% of any card before it gets to your store make this fact even worse.
So, like parent says, you are going to see radar scores of 90 and 95 for certain charges (clearly fraudulent carding attempts), followed by scores of 15 or 20 for the same card, IP, fingerprint with absolutely no warning.
I've grown tired of escalating this to Support. They just give me the ML model answer. Basically: "It's a black box!"
You can definitely add a rule to start blocking charges from X places, or with Y velocity, or always enforce 3DS, but then you're taking the model into your own hands, and that has some important consequences.
Your acceptance rate goes down. You're heavily interfering with the model and relying (and trusting) it less, and you realise you really don't need Radar to do that for you.
If you're serious about fraud, you must use a pure player solution that is 100% aligned with your interests.
From what we've seen with Stripe Radar in the past, that doesn't seem to be the case.
I'm a big fan of Stripe in may ways, but I really have a love/hate relationship with this side of their business...
Stripe support is broken. I recall patio11 writing about how Stripe is all about putting in the correct "process".
Well, the process for support fails.
We handle a decent volume and have been a merchant for over 8 years. We can't even get an account manager to handle specific issues.
90% of requests are met with a subtle RTFM!
If you ask specific questions, these are avoided.
An open ticket is bounced from person to person. No two people seem to touch the same issue.
If you try specific chat support, and you ask technical questions, you'd expect some knowledge. Instead, most agents put you on hold and go over documentation, only to come back and say they've escalated this to a ticket.
Then the ticket comes back with more links to documentation, never answering the question.
The only way to make progress, answers and some human treatment is by jumping from connection to connection on Linkedin, trying to get an intro to someone inside.
"I'm going to quickly shift from my terminal to this chrome tab to check this documentation but while it loads I'll get a dopamine hit from X."
Blur the screen and help me get back on track...