3 Ways MailChimp Uses Data to Scale
MailChimp launched in 2001 as a paid email marketing service, featuring a cute monkey mascot named Freddie. In 2009, MailChimp went freemium, with a customer base of 85,000 users. The new offering, “Forever Free,” permitted businesses to send emails for free, so long as the distribution list was fewer than 500 subscribers (a cap that was gradually raised to 2,000).
Just one year after the introduction of Forever Free, that customer base had grown 5x, to more than 450,000 users.
Today, MailChimp boasts 4 million customers, and they send more than 6 billion emails a month. The company is self-funded and profitable, and its 200-person team is entirely based in Atlanta. If you’re seeking a poster child for smart scaling, MailChimp might be it.
In the midst of aggressive growth in 2011, MailChimp hired data scientist John Foreman. He was tasked with digging into millions of email lists and addresses to find trends that might help MailChimp more deeply understand the email ecosystem, create a better user experience and scale efficiently.
We spoke with Foreman — who’s releasing a book in November entitled Data Smart, an extension of his blog, AnalyticsMadeSkeezy.com — about MailChimp’s three core tenets of scaling with data.
1. Use Data to Liberate Humans
Foreman says, “One of the cool things about having the free plan is that we get a lot of data from the small users,” which is rare in the email marketing space. So when the rapidly growing customer base — and, in turn, a growing number of spammers — started to strain the resources of MailChimp’s customer service and compliance teams, Foreman put the data to work.
Foreman dug into the data and found that most MailChimp spammers are more accurately defined as abusers — they’re legitimate businesses that employ distribution lists — as opposed to typical spammers sending emails about “v1@gra” or Nigerian princes. Often, these businesses are experiencing slipping revenues, so they turn to abusive email marketing practices in a desperate effort to hit certain metrics.
After studying email behaviors, Foreman developed Omnivore, an artificial intelligence model that predicts both “bad” and “good” users. “We want to predict if you’re bad before you send, because if we wait and let you send once, the damage is already done,” says Foreman. “We can’t let a user send spam from one of our IP addresses, because that’s going to negatively affect the entire ecosystem.”
Digging into metadata helps MailChimp determine whether a user-uploaded list is legitimate. Rather than waiting for the compliance team to put the kibosh on a user, Omnivore does it automatically, saving employees valuable time and resources. (Fun fact: Lists scraped from Facebook are the most “toxic,” meaning they’re abused often. So if a business’ email list has 70% correlation to the public database of addresses on Facebook, Omnivore assumes you’re in the spammer category.)
Just as it’s important to stop bad users before they abuse the system, it’s key to let the good users sign up and start emailing without hassle. In MailChimp’s early days, businesses would sign up, go into a review queue and await assessment by the compliance team — a process that could take up to a day. In a world where customers are spoiled by real-time, that doesn’t fly. So Foreman’s algorithm also predicts who is likely to be “good,” and Omnivore automatically approves these users and allows them to start sending emails immediately.
There’s also a “grey” category, in which case MailChimp may let you in, but throttle how much email you can send until it gets a better read on your behavior and intentions.
“Omnivore is all model-based and automated now, so that’s really helped us grow,” says Foreman. He adds that the compliance queue has dramatically whittled.
Another algorithm Foreman built helps alleviate burdens on the customer service team. With a global audience, MailChimp needs hands on deck to help customers 24/7 — but knowing just how many hands posed a challenge, and round-the-clock shifts were difficult to schedule around other meetings and breaks. As a solution, Foreman built an optimization model that takes into account the number of employees, shifts, lunch breaks and capacity, and then spits out an optimized schedule. The system ensures that MailChimp has enough hands (but not too many) on deck to help customers in real time.
2. Don’t Do Data Because Data Is En Vogue
Data is only useful if you know what to do with it. Foreman has put it to use to reduce the workload on MailChimp’s employees and make processes more efficient — but he doesn’t apply data where it need not be applied.
“Everyone talks about how data science is going to solve the world. It’s really not. I feel like my job is just to serve the rest of the organization, and if they’re already doing something well without having a metric involved, my job is not to come in and create some stupid metric for everyone to look at,” says Foreman.
Some metrics, he says, can actually degrade service. So the metrics MailChimp focuses on are forecast growth, the need for servers and other back-end infrastructure metrics that help the company plan for the future. “There’s not one metric that we look at to say, This is what success means. We just want people to feel good about the product.”
For example, MailChimp’s billboard advertising campaign features just Freddie the chimp’s face — no URL, no call to action. With this effort, MailChimp appeals to its existing user base, practically winking at its own customers. Tactics like this help MailChimp engage with users and get a read on brand health, while injecting some fun into an industry that tends to focus on open rates and average revenue per user.
“People will often ask us, What’s the ROI for that? What metrics are you tracking? And the truth is, we’re not,” says Foreman. “We just stay in constant conversation with our customers. We’re constantly taking their pulse to get their sense of what’s right, what’s wrong — and I really love that, which is weird for a numbers guy.”
He says the problem with having one success metric is that people will cheat to drive those numbers. He cites the fast food industry’s drive-thru trick where they tell you to pull to the side because they’ll bring your food. They’re just trying to reset the clock — your customer experience is not reflected by the timer.
3. Use Data to Help Your Customers Do Better
Not everyone is savvy enough to understand the art of email marketing; as Foreman points out, there’s still a cohort of people who use the term “email blasts.” One common misstep is the belief that “more is better.” Foreman studied the data and found that there is a sweet spot for send frequency. “There’s actually a point for your business where sending more gets you aggregately less engagement,” says Foreman.
Foreman cites Birchbox as a content and engagement paragon, but says that even if you have a solid content strategy, you might see engagement plateau or decline. He warns businesses against going into panic mode and sending frequent (and often replicated) emails — a practice that will likely cause engagement to drop even lower. “[That tactic] is actually kind of great for me because I can use that as part of my predictive model,” says Foreman.
To combat that fatigue and keep customers’ audiences engaged, MailChimp provides a blog with helpful tips and a Knowledge Base that covers best practices for send frequency, copy suggestions and content marketing, as well as video guides about leveraging email.
But just because the Knowledge Base exists, doesn’t mean people are going to find it. Once again, Foreman turns to the data — he’s working to create systems that better serve the educational content based on assessments of a business’ competence. First, the system tracks your clicks as you navigate through MailChimp. If you move through the site with ease, you’re likely a pro; a more tentative and exploratory click pattern indicates a novice. MailChimp also checks out your company website — does it have parallax scrolling, social integration and ecommerce, or is it a simple WordPress site? — and then serves up content that fits your perceived prowess.
Another benefit of Omnivore is that MailChimp had the infrastructure in place to build fun and useful features for customers, like the “discover similar segments” tool. “This was a really fun project because MailChimp is the only email service provider that could have done it — we’re the only one with this many users,” says Foreman. Most email recipients subscribe to more than one MailChimp email, so the company is able to detect interest patterns. This Omnivore byproduct, Wavelength, unveils clusters within an email list and helps a business understand its audience’s other passions — which can lead to strategic partnerships and inform the content strategy. More targeted content means better engagement, which often leads to increased brand affinity and more revenue. It’s a win-win.
Why MailChimp Loves Data
MailChimp has 200 employees, but more than 4 million customers — the application of data has enabled the company to stay lean and maintain a tight-knit culture while serving a very large audience.
Foreman brings up Mad Men and how Draper and company bend over backwards to please big clients like Lucky Strike — a practice he says MailChimp wants to avoid. Many other email marketers build up massive sales teams and focus heavily on core metrics. MailChimp hasn’t. Instead, the company brings a solid product and a little quirkiness to the decidedly unsexy industry of email marketing — and it uses data to become even bigger and better.
Source: mashable.com
