Cleaning a CRM by playing a game

Before a school can move its learners onto a new platform, somebody has to answer, for every person: which program did they finish, which did they buy, what kind of teacher are they now. The data for that lived in four places that disagreed, politely, about almost everyone.

The standard answer is a spreadsheet and a week of someone's life. The person who actually knew the answers runs community and alumni relations and has a phone, not a week. So we built her a game.

How it plays

One person at a time. The game shows what each source claims (program, purchases, teacher type) and asks which is right. Tap one. Or tap two, because sometimes two sources are both right: a 200-hour teacher who later did a 300-hour training is also a 500-hour teacher, and a merge is the honest answer. "Two of these are right" is a legal move.

Rounds are short. Levels have names. There are badges, and a small sound when a round closes. The avatars are randomized little placeholders, never real photos, because a game about real people's records should still feel light enough to play at 10pm.

What it had to get right

• Phone first, then iPad, then desktop. She plays on the couch.
• No keyboard hints on a phone. Obvious in hindsight.
• Partial merges. Not everything is a clean pick.
• Every decision is a rule. When she decides a pattern once, the game remembers it and stops asking.
• A list of people it couldn't resolve. Those become outreach, not guesses.

What happened

The prototype ran with sample data for a day and then got replaced by the real one, because the first thing she said was "I don't want to practice, I want to do it." Fair. The real build reads the exports straight from the shared folder, stores each decision, and produces the clean list the platform imports from.

Where it stands

Live, behind staff logins, being played. We'll know it worked when the import runs and nobody writes in to say they're listed as the wrong kind of teacher. The bigger lesson is one we keep relearning: when a chore has to be done by a person, build the thing for that person, not for the data.