back
user profile
brandonb
9,313karma·1,335submissions·January 17, 2011
about
Data for good.
Co-Founder at Empirical Health (https://empirical.health). Don't die of heart disease.
Before: Co-Founder @ Cardiogram (ML for heart health)
CTO at Sift Science (YC S11, machine learning to fight fraud)
Data Science @ UCSF Cardiology
HealthCare.gov rescue team
Google (Android speech recognition, search ads ML)
brandonb.cc
recent activity (1,335 total)
comment
We were lucky to find some early beta customers like Airbnb, Uber, and Listia, who were all looking for what we were building and willing to take a risk with a startup. In general, getting the first t…
comment
Thanks!
comment
I'm the OP and I work at Sift Science. Let me know if you have questions!
comment
Thanks Jason! Since every site is a bit different, we currently provide the score and let the site decide what to do based on their UI. I really like the idea of recommending specific extra steps thou…
comment
Good question! Trading off false positives and false negatives is a big part of why we expose a score, rather than just a "yes"/"no". Each site has a different balance. Most of our customers review ea…
comment
Sites integrate both a Javascript snippet and events from the server. If there are server-side events, but no Javascript, that's a suspicious sign. You can see the steps for integration here:
https…
comment
I work at Sift Science. Let me know if any of you have questions!
comment
Hey Tom -- I'd love to see if we could help with fraud even above and beyond what Cybersource/TM do. Feel free to e-mail me: brandon@siftscience.com!
comment
That's exactly it. It's important to look at a whole bunch of factors. For example, fraudsters tend to go directly to the payments page, whereas good users will browse around the site before making a …
comment
Definitely true. A lot of sites just block all transactions from, say, Africa when they start getting hit with fraud. But that's a very broad brush to paint with. It's better to use a machine learning…
comment
What makes the security code (CVV) different from just having more digits is that you (the site/merchant) not allowed to store it permanently in your database. You can keep it in-memory for up to 15 m…
comment
You're right that it's not proper A/B test, so it may be a case that correlation doesn't imply causation. Unfortunately, there's not a lot of research published with rigorous A/B testing for this kind…
comment
If you were to use a really minimal payment form (number and expiration), it's likely that you'd get somewhat more chargebacks. However, you'd also get more revenue. Let's say your chargeback rate goe…
comment
I'm the OP. Let me know if any of you out there have questions! Or, better yet, experiences with simplifying payments on your site good or bad.
comment
Thanks Chad! It's a pleasure to work with people like you!
comment
My company (Sift Science) uses machine learning to fight fraud, and we work with customers who sell digital goods. You're right that normal country blacklisting, IP blocking, AVS, CVV, etc. aren't ter…
comment
What this article doesn't mention is that Rakuten is the #7 site in Japan and one of the top 100 sites in the world. They bought buy.com and are consolidating brands. Although the name "Rakuten" isn't…
comment
Although I agree with a lot of the advice in the article, there's a "new school" of enterprise too, and ideas like virality from the consumer world often apply to the new enterprise. "Old school" ente…
comment
My startup (8 people) recently started interviewing at Waterloo, and the quality of students is every bit as good as MIT, Berkeley, Stanford, or CMU. Right now Waterloo is the best-kept secret in star…
comment
Chris is a fantastic investor. This is a steal for a16z.
comment
I know the WibiData guys, and they're a truly amazing team that knows HBase in and out. Their open source Avro extensions are great; looking forward to playing with Kiji!
comment
Good question. We measure these using "precision" (of the users that users Sift flags, which percentage are actually fraudsters?) and "recall" (what percentage of fraudsters on the site does Sift flag…
comment
My company (Sift Science) helps sites fight credit card fraud. We work with a few large ($100m+ revenue) marketplaces, and here are some things I've learned. First off, strictly speaking, this is most…
comment
Definitely! In fact, if you look at our site, one of the features we offer is real-time alerts:
https://siftscience.com We combine data across all of our customers, so that if we find a fraudster …
comment
I think it's really important to have an "appeals process." Many of our customers don't use Sift to block users unilaterally, but rather to ask for further verification. For example, if the Sift score…