Line goes up
If you work at a publicly traded company, chances are your job comes down to 2 things: increase revenue and reduce cost.
Sometimes the line between action and outcome can get a little blurry. Execs make decisions to maximize revenue and minimize cost 10 years down the line. External comms might publish content to build brand, which will drive more revenue a year down the line. But on a principal basis, it all comes back to the same thing. Revenue up, cost down.
The question of “how to be a great analyst” might vary from team to team and person to person. But there are a reasonably well defined set of outcomes the team can drive:
- Make a decision
- Create a feedback loop (aka make multiple decisions)
- Identify opportunities
- Diagnose threats
- Enable others to do more of the above
Each has a variety of tools, best practices, and quirks. Each probably has a dozen books for how to excel at a particular niche. What I’d like to encourage is to start with the outcome, then work backwards to the task.
Make a decision
The OG “pls pull this number for me” ask. Should we increase the fee cap or not? Should we launch the pilot globally? How many operators should we hire in Q3? At the lowest level this might be 6 lines of SQL. At the highest level this might be a 6-page causal inference analysis of the brand marketing impact on conversion. The sweet spot is somewhere in between, and most likely consists of a clear recommendation, a few charts or tables, and at times some supporting analysis. 90% of experiments and causal analysis land here.
The key is:
- Most data asks are intended to drive a decision (to hopefully increase revenue or reduce cost)
- If it’s not clear what decision or next step your analysis will drive, take a pause and figure out what that is.
- Your final analysis should probably come with a recommendation.
Create a feedback loop
Aka make the same set of decisions multiple times. Air conditioners keep the room calibrated to 70 degrees. Analyst-driven feedback loops keep your team calibrated to their north star.
The tools of the trade here are dashboards and narratives.
A dashboard isn’t a dashboard, it’s the first step to your feedback loop. A place you or a stakeholder can visit to make a repeated set of decisions. A great dashboard will surface the details needed to make repeatable, self-serve decisions at the top.
Narratives go a step further by:
- Calling out recent trends (the temperature is down to 55 degrees)
- Diagnosing why (neighbor left the door open on his way out)
- Making a recommendation (ban neighbor from visiting)
The biggest thing I’d like to push is treating reporting as a product. What is the problem the user is trying to solve, and how can you enable them to solve it as easily and reliably as possible.
Diagnose a threat
If feedback loops are the air conditioning, threat diagnostic is the collision avoidance module on your self-driving car. Are things trending in the wrong direction, and what can we do to course correct? In the best case, we anticipate trends and adjust course pre-emptively. In the worst case we introduce large scale problems and don’t catch them until 6 months later, or ever at all.
Identify opportunities
The words “strategic” and “exploratory” start to get thrown around a lot here. You might have a hypothesis (“customers in X space are underserved”) or even a vague idea (“we’ve never investigated this stage of the funnel - is there any low-hanging opportunity to improve”).
A fun(?) characteristic of exploratory analysis is that, much like experiments, they’re not guaranteed to yield a positive result. Often enough, the resulting insight is “we didn’t find anything worth prioritizing in the next quarter.” On the flipside, these analyses can also have some of the largest impact - causing shifts at the scale of organizational strategy.
Enable others to do more of the above
- Building and improving ETL tables
- Building and improving internal tools and processes
- Code reviews and coaching
- Recruiting new members (writing and conducting interviews)
- Keeping members around (organizing events, supporting teammates)
Footnotes
1. Empires have risen and fallen on the art of lifecycle marketing analytics. Great men dedicated lives to DAU. ↩
2. You can make the case that “shipping ML models to prod” is just feedback loops at a bigger scale. ↩