How MailChimp Does Deliverability

Issue 14

Joe Uhl is the VP of Operations at MailChimp. He leads the teams that support the entire company — operations, delivery, IT, facilities, and office management.

We sat down with Joe to chat about deliverability, infrastructure, and what sets MailChimp apart.

Joe, should agencies be concerned with an ESP’s infrastructure?

Yeah, I think it matters a lot, and there are a bunch of different pieces to it, too. Is the ESP reliable? How often do they experience downtime? Are they transparent about it when they do? Here at MailChimp, we try to be as transparent as possible. We have a status page that’s constantly updated with raw monitoring data, and we take uptime very seriously.

It helps that we’re as big as we are, because we’re able to identify a lot of good patterns. It’s provided the opportunity for us to develop the building blocks that we needed to be able to scale as we have. We have thousands of servers, our own fiber loops, we’re in several of the carrier hotels in Atlanta — we’ve built up a serious infrastructure that you couldn’t just go get if you were falling behind. 
What sets MailChimp’s infrastructure apart from the other ESPs?
Infrastructure-wise, we actually do a lot of things differently. Many competitors use big, third-party vendors, but we buy our own servers, use open-source software, and keep all of the expertise in-house. It helps us move faster. I’d rather find an expert and hire them to work here at MailChimp than flush that money away on a consultant somewhere.

For some companies, if they wanted to do a feature shift, for example, they would need to plan and coordinate with those third parties to make sure they had a big enough database. We don’t have that type of limitation. We’re unusual in the ESP market because we’re built more like a Facebook or a Google in terms of how we shard and stack. Our units of infrastructure are pretty small. There are no single points of failure in the way we build things because we’d rather spread everything out. If we want to iterate, or react, we can just order more servers and do it. We don’t need to go through anyone else. 
How is MailChimp’s approach to delivery unique compared to other ESPs?
 In many cases, when other ESPs talk about their delivery team, they’re really talking about account reps that work with customers and coach them on best practices. Our approach is different. It’s much more technical. It’s all about data. We’ve got good relationships with all the major mailbox operators. On the technical side, it’s a constant arms race — as the spammers’ tactics evolve, so do the rules and requirements across all of the different mailboxes. We’ve got to pay close attention to things like Reverse DNS, DKIM, and SPF Records. Our delivery team also has to track and maintain thousands of different provider rules too, things like how many IPs we can use, how many connections we can have per second, how many messages per connection, and how long a connection can be held open. It’s all constantly evolving, so you’ve got to have a really good team that stays on top of everything and is constantly monitoring for the slightest changes. I think we’re really good at that. 
Do you recommend that agencies invest in dedicated IPs for their clients?

There’s a lot of bad advice out there about sending email. Sometimes, people feel that they have to have a dedicated IP to get good delivery. The truth is that, in most cases, they’re better off just using our shared IPs because they’re older and they’re even more reliable. We care more about reliable delivery than we do about selling IPs. When you start getting to lists with hundreds of thousands of addresses, a dedicated IP can make sense, but for most folks, they can actually see better deliverability with our regular IPs.

How does MailChimp’s deliverability compare to smaller ESPs?

It’s all about scale. As an ESP grows, it becomes more difficult to keep up with that growth.

If you’re small, you can make it for quite some time, but it’s not always sustainable. We’ve grown into one of the biggest ESPs now, and we’ve gotten to the point where we’ve been able to build good relationships — and a good reputation — with many of the block list operators. They respect us and are willing work with us if we get in a jam because they know we take reputation seriously and are willing to enforce quickly.

The other big difference is that we’re far more strict about what we will and won’t allow in our system. We don’t allow affiliate marketing, we don’t allow adult content, and we’re very strict about the permissioning of the list. We’ve been really consistent in that stance over the years, and as a result, we’ve had to say no to some huge companies. We put more value in having a clean reputation than we do in generating revenue, and that’s been our stance since the beginning. But it’s paid off for us, and it’s helped us develop those good relationships I mentioned earlier.

Should an agency be discouraging their clients from importing a list into MailChimp?

It depends on the list. We trust people, so we won’t stop anyone from logging into MailChimp and importing a list. But we do a lot of sampling and prediction on each list, and we’ll shut it down if we think it’s going to perform poorly. And we’ll watch when folks send too, so if someone sends and it doesn’t perform well (maybe it generates high bounce or abuse rates, for example), we’ll shut that down, too. These policies do mean that we’re not always able to serve some of our potential customers, but in the long run, it helps all of our other users have more success with their own deliverability.

If you ran your own agency, what questions would you be asking potential ESPs?

I’d want to know how they monitor uptime, and how they handle incidents. How do they know when something is wrong, and, in turn, how do they show that to their customers? We just throw our uptime numbers on the status page. Most companies will have a grid of services and dates with a red/yellow/green and, when the incident is resolved, they’ll just flip it back to green so there’s not a public, historical record of the issue.

I’d ask about their infrastructure. Will I be able to track how things are going and what will they do when there’s an outage? How have you built your app? What kind of storage, networking, and servers do you use? What kind of people do you have?

This first appeared in MailChimp’s newsletter for agencies. Sign up for the newsletter here.