This week we’ve rolled out a significant improvement to our reputation algorithm, affectionately known to the Mailgunners as Razor. In today’s blog, we’ll tell what you’ll start noticing about the changes, and some of the background around why we designed the system this way (and why it is good for our customers). We also just posted a look at the seedy underworld of spammers and phishers which we’ve learned a lot about during this process. Fun read if you’re interested. So…
No defensive & arbitrary sending quotas placed on new customers
We use a host of signals to detect spammers and shut them down as quickly as we can. We take advantage of this speed to let new customers send as much email as they need as soon as they signup. Why should good customers suffer because of a few bad apples?
Better Sender Scores on our shared IP address plans
Because our reputation algorithm is effective at detecting and disabling spammers, few spammers can make it onto our platform, diminishing the reputation of our shared IPs (a measure of an IPs reputation is its Sender Score). Functionally, this means that you’ll be able to sign up for a brand new Mailgun Standard account and send 10’s of thousands of emails per hour with no delay because our algorithm is smart enough to recognize the difference between a good and a bad email. We read a lot of books to make this happen. And we hope you like it.
There are evil people in the world. And they know how to use email. That means that every company who runs an email service of any kind needs to spend a significant amount of time fighting spam. It is just a cost of doing business.
There are a lot of ways to fight spam, and the method or methods chosen depend on how customers use (and can potentially abuse) the product. Gmail, for instance, limits how many emails can be sent per hour via their SMTP servers. This makes it really hard to spam from Gmail, but wouldn’t make sense for Mailgun, where our customers regularly want to send millions of messages a day, even the first day that they sign up.
Ideally what you want to have in your email product is a system that let’s good customers send as much email as they want while spammers are blocked completely. Our reputation algorithm does this.
One way to build a reputation system is to set some arbitrary and defensive cap on how many emails new customers can send, even important transactional emails like password resets or welcome emails. As you slowly monitor the quality of the email, you let them send more and more. Until that customer sends a spike in volume. Then you limit them again, just in case they are sending spam (even if that “spam” is a new double-opted in newsletter they are starting). This is called throttling and we think that it is the wrong approach to reputation management. Using sophisticated tools like content and link analysis, Mailgun can detect spam emails almost instantly and disable the offending account while letting our good customers send as much email as their business requires. Without this system, sending speeds will suffer leading to unhappy customers, and we can’t stand that.
Yes, unfortunately, Mailgun can only deliver email as fast as ISPs like Gmail and Yahoo will accept it. Theoretically, we could send millions of emails per second for our customers but there is no way that Gmail, for instance, would accept that amount of traffic from a single IP address, so we’re not going to send that fast. Mailgun automatically adjusts sending rates based on ISP feedback, so that your email is delivered as fast as it will be accepted. If our customers are getting slowed down by ISPs throttling their IP addresses, we can always add more IP addresses or employ some other creative solutions. Since content is the #1 most important factor that ISPs use to flag spam, follow our Email Best Practices guidelines for creating good email content. You’ll see your delivery speed into ISPs improve while having happier customers at the same time.
Till next week.
Last updated on August 28, 2019