Why a Data Analyst Chose a Ten-Person Company

Probabilistically, It Was a Strange Choice

Last fall, I opened a spreadsheet with two offers in front of me. On one side was a company with 1,400 employees and a data organization of 40; on the other, an education company with ten employees total and zero data staff. I built twelve line items for things like salary, growth prospects, and commute time, multiplied in the weights, and the big company won. I closed that spreadsheet and picked the ten-person company. Not my proudest moment as an analyst.

My worries boiled down to two. First, there was no data infrastructure. What is an analyst supposed to do at a company with no warehouse, no pipelines, and no BI tool? Second, there were no peers. Could I keep getting better in an environment with nobody to review my queries and nobody to talk A/B tests with over lunch? Both worries turned out to be half right and half wrong. Which half was wrong is what this post is about.

Worry 1: No Infrastructure, No Analysis?

The reality I confirmed in my first week was worse than expected. Attendance data lived in the Rubric app, Epic Prep vocabulary quiz scores lived in seven spreadsheets, and counseling notes lived in the teachers’ paper notebooks. Numbers that would have taken a single line of SQL at my old job took two days to pull together.

But about a month in, I noticed something odd. Messy data and useless data are two different problems. At my old job, the share of my reports that actually made it into decisions was 20% by a generous estimate; the other 80% disappeared with a reply that said “we will take it into consideration.” I am cautious here because the sample is small, but of the twelve analyses I produced in the first half of this year, nine led to actual changes. A 75% adoption rate. The infrastructure is less than a tenth of what I had, and the adoption rate is more than triple.

There was trial and error, of course. Two months in, I built a metrics dashboard with 24 charts, exactly the way I would have at my old job. Two weeks later I checked the access logs: the view count excluding me was 3, and 2 of those were clicks I had made while demoing it. I can confirm that the ability to build dashboards nobody uses survives any change in company size.

Worry 2: No Peers, No Growth?

This is where my prediction missed by the widest margin. I have no data colleagues, but I have data users: all ten of us, myself included. One afternoon Rachel asked, “Why does this class’s homework submission rate drop specifically on Wednesdays?” I looked into it and found three students whose Wednesdays collided with another academy’s schedule, and starting the following week, that class’s homework deadline moved to a different day. Six days from question to classroom change. At my old job, a change of that size took a quarter on average.

This speed has side effects. If my analysis is wrong, next week’s classes actually change for the worse. With no colleague to check my work, I have become far more sensitive about sample sizes and verification steps than I used to be. My skills used to grow in the direction of building more sophisticated models; now they grow in the direction of noticing quickly when I am wrong.

What Turned Out Differently

Let me be honest. I expected analysis to be 80% of my job; in practice it is around 50%. The rest of my time goes to fixing input forms, teaching teachers spreadsheet shortcuts, and occasionally carrying chairs for Playfit trial classes. At a ten-person company, the usage frequency of the sentence “that is not my job” converges to zero.

The worry about missing peers also stayed half true. There is still nobody to point out the odd habits in my queries, so I attend an external analysts’ meetup once a month, and I am withholding judgment on whether that is enough. If you are the type who cannot sleep without code review, this environment is hard to recommend.

Still, the math can be run again. If I went back to last fall and opened the same spreadsheet, this time I would add one more line item: the probability that my analysis changes something next week. What I know now is that the moment you weight that item honestly, the result of the calculation flips.