
Company-size segments show nothing across three thousand accounts, and re-cutting the identical rows by what each user was trying to finish that day produces a clean split
Same Rows, Different Question — Volume I, Understanding Humans Before Products · Jobs & Progress · Investigation · Foundations. Company-size segments show nothing across three thousand accounts, and re-cutting the identical rows by what each user was trying to finish that day produces a clean split
01 · Opening
Company-size segments show nothing across three thousand accounts, and re-cutting the identical rows by what each user was trying to finish that day produces a clean split
02 · The setup
Sam has spent a week on the segmentation and comes back with nothing, which he is honest enough to say.
Sam: Small, medium, large. Every segment behaves the same. There is no story in it.
03 · The setup
Week-4 retention by company size — Under 20 staff: 42%; 20 to 200: 44%; Over 200: 41%; Accounts: 3,010
04 · The evidence
Maya: That is a real finding. Company size does not predict anything here.
Sam: It is the only cut the tool gives me.
05 · The evidence
Maya reads twenty of the first-week support conversations and stops sorting people by who they are.
Maya thinks: Every one of these opens with something they were trying to finish that day.
06 · The evidence
Maya: Can we tag each account by the first thing they completed rather than by how big they are?
Priya: Same three thousand rows. Different column.
07 · The evidence
The identical accounts, sorted by the job they arrived with.
Week-4 retention by first job completed — Reported to someone else: 71%; Replaced a spreadsheet: 63%; Explored, no clear task: 18%; Accounts: 3,010
08 · The evidence
Sam: Seventy-one against eighteen. That was sitting in the data the whole time.
Maya: It was. It just needed a different question asked of it.
09 · The evidence
Demographics describe who someone is. They are easy to collect, they are what every analytics tool offers by default, and they are frequently unrelated to what a person is trying to get done on the day they meet your product.
A job is a situation with a deadline attached — reporting to a manager on Friday, replacing the spreadsheet that broke last week. Two companies of wildly different sizes can arrive with the identical job, which is why the size cut flattened a real difference into noise.
Nothing was missing from the data. The rows were the same three thousand rows both times. What changed was the question, and that is usually cheaper than collecting anything new.
10 · The evidence
Dev: So we build for the reporting job.
Maya: Or we work out how to give the eighteen percent a job on their first day.
11 · You make the call
The story does not tell you first.
A. The second analysis used better data
B. Size averaged together users who arrived with very different jobs
C. Company size is never a useful segmentation
12 · What happened
The first-run screen starts by asking what the user came to finish, and the eighteen percent stops being a category.
Maya thinks: A week of segmentation and the answer was one column we already had.
13 · Complete
A flat segmentation often means the wrong question rather than an absent effect, and re-cutting costs nothing. Next: the competitor that never appears in a competitive analysis because it has no website.