Our first deliverable on most engagements is not a model or a dashboard. It is a document defining eleven words, signed by people who have been quietly disagreeing about them for years.
Retention. Persistence. Completion. Enrolled. Active. Withdrawn. Deferred. Full-time. Attempted. Passed. Cohort. Every institution we have worked with had at least three of these defined differently in different offices, and almost none of them knew it before we asked.
How you find out
Ask three offices for last year’s retention rate. If you get three numbers, you do not have a reporting problem, you have a definitions problem, and buying analytics on top of it will produce a fourth number that nobody trusts.
This is not a criticism of anyone. The definitions diverged for good local reasons — the registrar counts on a census date because that is what the regulator asks for, the dean counts at term end because that is when they can act, and the IR office counts on an annual basis because that is what the ranking submission needs. All three are correct for their purpose.
Three offices, three numbers, all defensible. That is normal. What is not normal is building a prediction system on top of it without noticing.
What the document has to contain
- The definition, in a sentence a non-specialist can check.
- The census point or window it applies at.
- The exclusions, named explicitly — approved breaks, transfers out, deceased students.
- Who signed it, and when.
- A version number, so a year-on-year comparison cannot silently compare two different questions.
Why it has to be signed
Because the argument will resurface. Twelve months in, someone will say the number looks wrong, and the only thing that resolves that conversation quickly is a document with their colleague’s name on it. We have watched projects lose a term to this and it is entirely avoidable.
You do not need a vendor to do this. If you are considering analytics in the next year, start the definitions workshop now — it is the highest-value unpaid work available to you, and it will make every subsequent procurement conversation shorter.
Our data dictionary template for Indian higher education is published as EPT-2025-04, including notes on which definitions are most contested and why.
