Vertical versus Horizontal Impact
Two ways that you can make impact as a data scientist. Three main functions — exploratory analysis and metrics definition (output answers “what is worth looking at and how”), confirmatory based on experiments or causal inference on observational data (output is “we should do X and Y will result”). Predictive work is a bridge between these two things. If you predict doing X leads to Y, then we should try doing X.
Vertical — build crazy good models, find deep insights, push a decision forward. This is all about moving the needle.
Horizontal — unify people’s definitions for efficient communication, deploy a platform for doing analyses, or running experiments. This is about scaling other people’s efficiency at doing data analysis.
It’s an Applied Science
The job of DS is not to discover new techniques. It’s to apply existing techniques thoughtfully to their domain. Thus, DS is closer to epidemiology and neuroscience than it is to Computer Science or Statistics. The job of a CS or Statistician is to create new ways of doing things. The epidemiologist spends their time consuming the statistical and computational literature, to be applied thoughtfully.
Getting these two things mixed up is probably a core reason behind companies who are not getting value from their DS. They hire these smart PhDs in statistics only to find that they want to spend all their time writing proofs.
It’s important to understand the foundational work, but a DS should be using as much of it as possible before trying to extend it.
Short versus Long Term Goals
You need a delicate balance between known-wins and longer term investments. This is the classic exploration/exploitation dilemma. You have, at any moment, a large range of projects you could be working on. Some are known to be useful and just needs the time. Others are bets that have <50% of paying off but likely to pay off big. This is the difference between hole digging and well digging. Say you’re in a village and you need holes and wells. At the end of the day, you’re going to have a hole. But the mentality is completely different.
Digging a hole is working on things that you know are going to work. This is the optimization work, where you’re >90% sure that you’ll get a bounded upside. It’s just a matter of putting in the time. You should always be working here to provide value and knock out marginal increases to the bottom line. Wells is when you dig somewhere with the hope that you’ll hit water. You dig, but lots of the time you’ll just end up with a hole in the wrong place. For this work, you need to look at your hit ratio. Is it going up?
The right balance is probably 70/30 holes to wells.