Chief data officer, MailChimp
Deluged with close to 2 billion e-mails a month in 2011 and struggling to stay ahead of spammers, e-mail-marketing firm MailChimp hired its first chief data officer. John Foreman, a former data scientist at Booz Allen Hamilton, did more than tackle the problem with a new automated vetting process; he mined MailChimp’s enormous datasets for the inspiration that led to a suite of new products. What were the keys to his success? Foreman, who is also the author of the forthcoming book on data science, Data Smart, claims his ability to communicate was even more important than his penchant for manipulating data and technology. Translating arcane math into a lingua franca is a prized skill, he says, that companies need to learn how to both appreciate and develop.
What’s it like joining a company as its first chief data officer?
The main hurdle to doing data science in an established company is really change-management. It’s not technical. The company was very design and user-experience driven and they were very, very good at creating a streamlined, well-edited, well-presented product. But with almost 4 million users and several billion e-mails sent a month, they wanted someone who could translate their ideas and their needs into actual data products, as well as someone who could dream up new ways to exploit all the data they were gathering. When I arrived I wanted a quick win to convince folks that I could offer them tools to make their lives easier. There was actually some bad will toward data science because some of the developers had tried to do a project themselves before I got here and failed. I was able to rectify that and convince them that data could be useful.
How important is it to get that quick win?
Each organization has to address this differently, but you’ve got two competing ideas here. One is that you don’t want to over-plan or overbuild infrastructure to the point where, once you deliver the product, it’s no longer useful for the business. You definitely want to move quickly, which is why people talk about doing data science in an agile way. But there’s also a pitfall to be aware of. If you do that over and over, you’ll eventually come to realize that there’s no overall plan. At some point you need to build an infrastructure for the next few projects. It’s not easy to predict what you might need before you actually need it without over-scoping, of course. It takes a bit of magic.
What is the most common mistake you see made by companies trying to use data?
Many organizations where I’ve worked are missing translators; people with technical backgrounds who can talk to multiple teams and understand what’s going on across departments. That has actually been my role a lot of the time. I can talk to folks who don’t know how to express what their analytic need is and then talk to the technical people and characterize it for them in a way they can understand.
Can you give an example of a scenario where translation would be necessary?
Think about dynamic pricing, where you build pricing models based on current or past bookings, what you know about competitors’ pricing, etc. It used to be just hotels or airlines that were interested in this, but now all sorts of companies do it. Building these models requires an intense collaboration between business folks who understand the marketing side and those who understand forecasting and how to build price-optimization models. That requires people who can talk between these groups and in all these different languages. When a company finds someone who can do that, they don’t want to let them go. They’re incredibly valuable.
How can a company find or develop such skills?
This isn’t a skill that is typically looked for by HR folks. More often they’re hiring people with Ph.D.s in statistics or computer science, people with specific pedigrees that allow them to be easily fit into a company silo. Only once someone is in a company and people see they can translate and speak multiple languages does it become apparent how important that skill is. It might be someone like me with a math background or it could be a person with an MBA who really liked their operations-research classes. But most people learn this through experience. Maybe they started as a technical person and then got promoted to the point where they manage all types of folks and eventually learn how to speak other languages. Or maybe they’re not technical, but they just have to learn it by necessity because there’s no one else who can do it. I encounter a lot of people that come this way by accident.
Any other warnings for companies that are venturing into big data territory?
Listen, data science is not magic. It’s not just cool or sexy or something you do because everyone else is doing it. Data science at its most basic level just takes raw transactional data and translates it through math and statistics so that it’s useful for the business. It allows you to make better decisions. I think a lot of companies aren’t thinking about this correctly. They’ve got all the best software, all the best hardware, all the best talent, but there’s a disconnect because there isn’t someone who can talk to everyone and understand what the business needs are and how to match that up with the technology.
How did you develop a culture of data science at MailChimp?
I’ve met a group of semi-analytics people throughout the company, just people who are interested in data and want to do something with it. I’ve found them in customer support, in marketing, in design, in finance—they’re everywhere. And so we collaborate together and we sit down and just talk. In fact, we do that every other week, get together and talk about how we can use data to actually improve the business. Sometimes it leads to writing some code and eventually turns out products. But the most obvious benefit is just having everybody training each other and sharing knowledge. That’s been super useful for us.