How Bolt does fraud detection better
blog.bolt.com
blog.bolt.com
Then they go on to not include any ideas of price.
When companies can't give me a price range or idea I always have a feeling that their pricing model is to guess my profits and then guess how much they can take and still gain my business. Usually you can identify these types of propositions when the company tells you they're your 'partner' and you're not just a customer - but a partner.
Just tell me up to $x in sales it's $y per transaction and $y can have different prices for different features... I don't want a 2 hour call or screen cast to find out it's out of budget. This sales funnel tactic is no more moral than the long form copy sales website.. where they get people to spend 2 hours reading something - then at the end hit them for some amazing offer - where the user feels they've already invested 2 hours 'learning how to x' why not spend $95 and some reaaccuring fee for ever... dark pattern.
Typically our deals are more enterprise-y / custom, especially since we're taking on all fraud liability. It's not just a processing rate.
Feel free to send me your volume and website and I'll get you back a quote: rb [at] bolt.com.
And the pricing page will be updated soon. Appreciate it, again.
> send me your ....
> I'll get you back a quote
Thanks, but nah. I don't want to call or email or negotiate, I just want to give you my money and get the service, so I'll be using your competitor who has actual pricing and plan comparison front-and-center on their website.
...or are you just a quote bot and unable to actually read comments? It certainly doesn't appear that you understood the parent's comment.
>I always have a feeling that their pricing model is to guess my profits and then guess how much they can take and still gain my business.
"Enterprise-y" pricing is codeword for "smooth-talk non-technical managers at huge corporations and bilk them for as much as they're conceivably willing to pay". That's how the enterprise sales model works, and it can be very effective.
It's not always a bad thing, but it does seem irritating that their only pricing option is enterprise pricing.
Don't have much else to add at this point :/ We're new, growing, and getting better. That involves taking feedback like this seriously. Will deliver through our actions in the near future instead of words.
But if enterprise pricing is the only model you're offering, then unless your goal is to only ever get business from big corporations (which, again, isn't necessarily a bad thing), you're going to turn a lot of potential customers away.
What are your false positive and false negative rates (if I may ask)? I see you allude to the importance of those metrics somewhere in the post, but I don’t see the comparison with yourselves.
Crucially, if you’re taking on responsibility for fraud (i.e. false negatives), you’ll be heavily incentivised to skew your model towards low false negatives. Unfortunately, because of the very nature of algorithmic fraud detection, these are inversely linked to false positives. Less fraud <=> more genuine customers bounced, and vice versa. By taking a side in that bet, your priorities may not be aligned with those of the customer (esp high margin merchants are totally fine with some fraud). Is there something you do to counteract that?
This is something I always wondered about when I saw other companies with this business model.
Nice post though, and amen to the “we detect fraud!” - “how?” - “uhhhhh…” comic :)
Half a dozen points to answer this question:
- First off, our #1 north star metrics are order approval rates and minimal false positives. We measure this for every merchant. Clients sign up with us under the clear expectation that they'll see a spike in order approval rates and all of our case studies (compared to tools like Signifyd, Riskified, Stripe Radar, and many others) have shown significant spikes in order approval rates.
- We have some of our case studies published here https://bolt.com/case-studies and will be posting more in short order.
- We are more incentivized than siloed fraud detection providers to approve more orders. Payment processors are aligned to maximize volume. Fraud providers are aligned to minimize fraud. We play both roles. We want maximum volume through our pipes just as badly as merchants do.
- Furthermore, fraud rates are simply FRAUD / TOTAL VOLUME. We always aim to maximize TOTAL VOLUME to keep fraud rates low vs squeezing the numerator (which is what most companies do).
- Finally, as elaborated in the most, we have unparalleled data visibility by powering checkout. We use no rules. So, just because something looks bad (e.g. a user using a VPN, coming from a certain country, etc) our models just factor that in as one of hundreds of variables.
- We try to not profit and maximize fraud for the first several months to root out false positives in our models. If you only approve things that are clearly good and reject all else, your models will never learn.
- Our human review team rigorously processes every machine decline before it's officially declined to once again maximize order approvals.
A lot more that I could elaborate on here. But, just a few quick tidbits on what we do to minimize false positives as our #1 priority. All of the above and more lends itself to, what we believe to be, the best results in the industry.
I thought exactly this - they will reject the transaction if uncertain, because making bad decision here would mean loosing money for them.
It's not a fair comparison unless you compare approval/decline rates and total fraud per platform.
For every client we measure order approval and decline rates before/after. Here are several case studies with those results: https://bolt.com/case-studies
And we have more that we can send over if you're interested in learning more! rb [at] bolt.com.
I'm looking for something like this but Stripe just isn't cutting it :(
Email me directly and I'll send over details? rb [at] bolt.com.
I've we have already approved the order, it is indemnified for fraud and we'll take the loss if it's a stolen good.
All future orders from that CC will be detected and flagged as having previously had a chargeback as well!
Every good customer is precious, and not only has order value, but also has customer lifetime value, referral value, and more. Falsely rejected customers will also go to competitors, handing that value on a silver platter to them.
Our unparalleled data visibility, fraud models, and review team results in a white-glove service that maximizes order approval rates and minimizes false positives. We’ve had countless customers switch from Signifyd and related vendors to Bolt, all of which realized these improvements.
Finally, our checkout drives 10%-50% newfound revenue. In summary, we can offer an all-in-one solution and massive revenue upside for making the switch. Let me know if I can help :-) rb [at] bolt.com
"Mouse movements, keystrokes, capitalization, clipboard usage, and more make sense, because we understand all the elements of the DOM" ...
Such data is super valuable for fraud detection.
That is - if ability to model mouse movements as well as similar behavior features over time is indeed part of the offering and included into modeling [1]. Pretty much no one to speak of is productizing it today.
[1]
https://www.splunk.com/blog/2017/04/18/deep-learning-with-sp...