Teaching People Analytics at Harvard
Visit us at https://www.ansaro.ai to learn more about how we’re using data science to improve hiring
Last week, I had the privilege of co-teaching two sessions of the course “People Analytics” at Harvard Business School. I was honored to join Professor Jeff Polzer, as well as our partner and client, Mike Metzger, the head of Recruiting and Admissions Strategy at Teach For America.
The case we taught focused on TFA’s data-driven approach to recruiting, and Ansaro’s work introducing new machine learning techniques to TFA recruiting, in particular Natural Language Processing (NLP). Our time in the classroom ranged from high-level strategy discussions to breaking out STATA and running models. (@Jeff Polzer, I still plan to convince you to adopt Python!)
After two days surrounded by young people excited to learn about People Analytics and expand its frontiers, I thought I’d share a few observations on this nascent, fascinating field:
Interest in People Analytics is skyrocketing. In the past year, the enrollment in Jeff’s relatively new “People Analytics” class more than doubled, to — by my count — almost 100 students. That means over 10% of Harvard Business School students have elected to study People Analytics. Even more striking was the diversity of students’ professional backgrounds: HR, engineering, marketing, finance, strategy, and many others. This diversity made the class discussion incredibly rich. For companies building their own People Analytics teams, there’s a lesson here that including professionals from different fields can yield immense benefits.
Technical competency matters. I was surprised and impressed by students’ eagerness to write statistical code and crunch data. (MBA students coding is a rare sight!) Students recognize that statistics, data science, and machine learning can be powerful assets to aspiring managers, but they can also be misleading and downright dangerous if used incorrectly. One way to avoid this is to “get into the weeds”: build a bunch of models yourself, so that years down the road, when you’re no longer the one coding, you still know the right questions to ask.
People Analytics can, and should, be used in culturally-aware ways. One of the most interesting parts of class focused on “exporting” recruiting models developed at Teach for America to similar international organizations (e.g., Teach for Colombia, Teach for Armenia). We asked students to vote whether Teach for America models should be exported (1) “as-is”, (2) with significant adjustments, or (3) not at all. “Export with adjustments” was a clear winner. Students argued that models should be adjusted to reflect differences in data collection processes and subtle differences in mission. At the same time, students recognized the data advantage that Teach for America has due to its larger size and longer history compared to its sibling organizations. Insights TFA has gleaned cannot be replicated by newer and smaller sister organizations, so exporting some of TFA’s people analytics makes sense only when done carefully and transparently.
There is no organization too small for people analytics. Even if you’re a 5-person company, you can start implementing the systems to collect reliable, detailed people data. More importantly, you can start building a culture that places value on collecting data, even if it isn’t going to be analyzed immediately. At this point, you can’t get statistically significant insights from that data. But if you wait to grow to 500 people and a statistically significant sample size before starting the people analytics journey, organizational culture is likely to have coalesced in a way that makes implementing new systems and practices harder.
Enormous thanks to Harvard and Jeff Polzer for inviting us. Jeff, I can’t wait to see the students from your class start changing the world (and maybe we can even convince a few to do so at Ansaro!)
Harvard Business Review will be publishing the case on Teach For America and Ansaro soon, and we’ll add the link here when it becomes available.