Announcing Sift Science: fight fraud with large-scale machine learning | Hacker News Reader

Announcing Sift Science: fight fraud with large-scale machine learning

blog.siftscience.com

126 pointsbrandonb32 comments

I'm kind of wondering how this will (or can) fit from a deployment set of concerns (cross-silos, compliance & audit, handling CDNs, etc.) into large-scale environments like eBay or Paypal and whether the fraud models are appropriate for most that do business online. As a crude example, I might care about fraud in blog spam from SEO blackhats, which is not the same as credit card fraud patterns but both translate into real dollar loss for some businesses. Granted, the founders are ex-Googlers so perhaps this was done yesterday, but the way they present themselves and with their current customer list they make it look like it's mostly for direct financial transaction based sites. I'm suspecting things are overly simplified but I find it pretty hard to believe that inserting some Javascript snippets into a page could actually help when fraudsters could eventually start bypassing that JS (see: web scrapers). The API docs seem awful naive in number of possible cases.

With that said, congratulations on the launch, I know this is not an easy problem to handle and I think you actually have fulfilled a major goal of ease-of-use for online fraud detection systems. The only comparable systems I've heard of are ridiculously expensive beasts of enterprise software or custom-grown for the company with incredible barriers to entry. Best of luck to you guys.

Hey Devon -- thanks for the thoughtful comments!

Most of our customers do use Sift Science for financial fraud, but because it's a machine learning system, you can train it to detect other types of bad behavior like spam. We have customers in production using us to detect spam, fake inventory, and duplicate accounts. If you have a use-case that doesn't quite seem to fit, let me know and we can figure out how to train our system to recognize that type of behavior: brandon@siftscience.com.

If a fraudster bypasses the JS, we still have REST events such as transactions (or any other custom event sent from the backend). Seeing a user who has REST events but no Javascript events is a suspicious signal in itself, so fraudsters can't circumvent the system by just turning off JS.

FWIW, we're on some pretty major sites that we can't announce, so we've gone through a bunch of compliance, audit, security, and other concerns already.

Can you explain better what exactly you do?

It seems that you use ML to pinpoint fraud (or better, undesired behaviour) in $WEBSITE.

So that could be fraud detection because they login from one country, ship to another and buy a certain combination of items.

Or that could be "spam fighting" because they create an account with X and Y characteristics and post similar things.

Am I right in my perception? (and you could use this not only for 'bad behaviour' but for good behaviour as well)

You're right on. In theory, you could train our system to recognize good behavior if you sent us enough $label events, but most of the patterns we have today are really optimized around detecting bad behavior.

Could you develop a little more how duplicate accounts could be detected?

An article will appear a year from now..."Perpetrating fraud with large-scale machine learning."

Just kidding ;)...Looks awesome and easy-to-use, congrats on launching.

I'm the OP and I work at Sift Science. Let me know if you have questions!

Congratulations! Best of luck.

What differentiates your service from existing fraud detection services like CPA Detective?

Is it possible to run this stuff on certain purchases say like gift cards or would that reduce effectiveness of the software? If its looking for patterns in ordering, and i only feed in gift card purchases, it would have nothing to compare against. Reason I am asking, we have almost 0% fraud/charge back rate since you can only by our physical product via a subscription model.

While I'm currently evaluating and very excited about Sift Science, what are some of the competitors in this space? I'm only aware of CPA Detective Max Mind miniFraud.

The two biggest anti-fraud vendors are Accertify and ReD. You might also look at ThreatMetrix, 41st Parameter, and Iovation, who do primarily device identification.

Let us know what you think!

Where did you get your initial data when you were starting up?

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 two or three customers is one of the hardest parts of building a B2B startup. Might make a good blog post one day!

I'd love to read such a post and best of luck!

I sat in your ACM presentation the other day at Stanford. It was really awesome. Thanks for coming by and best of luck!

Glad to hear it! I had a lot of fun coming back to campus. Good luck to you too!

carl @ sift

I may be biased, but I hear Sift Science is a great place to work ;).

Who, exactly, is "am9icytiNjRAc2lmdHNjaWVuY2UuY29t" (listed as a contact on the Jobs page)?

That's a small puzzle. :)

Gotcha. I figured it was either a puzzle or an error (ie "I don't know what the contact email address for this job is yet, so I'll enter some gibberish...").

hint hint hint: echo 'am9icytiNjRAc2lmdHNjaWVuY2UuY29t' | base64 -D

Ah, that would be why my attempts at a character-for-character deciphering failed...

aww don't spoil the fun

https://www.google.com/search?q=define+hint

trxblazr, you have demonstrated poor judgement. Sift Science is trying to make a small attempt at filtering some of the inevitable spam that results from a public job posting. You spoiled it. Why? For what purpose? To show how smart you are?

Sift Science does not look like some arrogant company that has cooked up their own broken security system, and that deserves to be poked at this way. It is not a public service to defeat this quiz, a quiz that some people would have been happier solving on their own so as to have a fair shot at apply for the job.

I was going to post the same thing. Would love to apply, but I am not local. :)

Congrats from the Balanced team! You guys are doing great work.

Thanks!

Any plans on working with major e-commerce platform providers?

Definitely! What would be your top choices? And what's your ideal integration experience?

shopify...

Reply on news.ycombinator.com