TL;DR / Summary: Turn messy CSV onboarding into a timed detective game: new analysts import a deliberately broken file, find planted errors, and score points for each fix. They learn Power Query habits and sheet hygiene without a slide deck about “data quality.”
Onboarding that feels like a case file
New hires glaze over during lectures on delimiters and encoding. Give them a case: “This export from the fictional shop looks fine. It is not. Find the seven lies.”
The game frame creates permission to be suspicious. Suspicion is the skill you actually want in analysts.
Keep the story light. The learning is serious; the narrative only needs enough colour to make hunting fun.
Designing the sabotaged file
Plant errors across categories: a UTF-8 bombed city name, a date in DMY among MDY peers, a thousand separator that became a decimal, a duplicate order ID, a trailing space in a SKU, a blank revenue that is really zero text, and a column silently shifted after a comma inside an unquoted field.
Seven is enough for a fifty-minute session. Twenty turns the room into despair.
Ship a sealed answer key for facilitators only. Players should not see severity rankings until debrief.
Scoring that rewards method
- Points for each true positive found and correctly classified
- Bonus for documenting the fix steps in a short runbook cell
- Small penalty for “fixing” good rows into new errors
- Team bonus if every member can explain one finding aloud
Method bonuses stop the loudest person from claiming all the glory by glancing at filters.
Tools you allow in the first round
Round one: worksheet only—filters, trim experiments, text to columns, conditional formatting for duplicates. Round two unlocks Get Data / Power Query if your environment supports it.
The restriction teaches that GUI imports are not magic. Students see the same defects both ways.
Ban Python for this onboarding game even if the team uses it later. The point is spreadsheet literacy shared with non-coding stakeholders.
A clean chart of dirty inputs is how bad decisions get executive polish.
Facilitator moves during the hunt
Walk the room with a “hint budget.” Each team may buy two hints at the cost of points. Scarcity makes them think before asking.
When a team “fixes” dates by eye, ask them to prove the serial numbers. Eye fixes fail on the next month’s file.
Celebrate the team that finds the shifted column, not only the team that finishes first. That error teaches respectful parsing.
Debrief that sticks to operations
Map each planted bug to a real past incident if you have anonymised ones. Relevance converts the game into culture.
Draft a one-page intake checklist live with the cohort: encoding, delimiter, header row, date format, key uniqueness, null meaning, row counts versus source system.
Send them to practice more interactive drills at excelgames.cc so the detective mindset continues between formal trainings.
Remote and hybrid variants
Breakout rooms need a shared file copy each, not one shared workbook that fights over filters. Clone before the session.
Use a visible timer. Asynchronous Slack versions lose urgency and drag for days.
Record a two-minute loom of the weirdest bug for people who missed the live debrief.
Measuring whether onboarding worked
A week later, give a new sabotaged file with different cosmetics but similar bug classes. Score improvement, not perfection.
Watch whether new hires now ask source owners about null semantics before modelling. That behaviour change matters more than the game score.
If scores stay flat, shorten the story and lengthen the guided Power Query segment.
Safety with real company data
Never plant games inside production extracts. Synthesize. Real PII in a training gag is an incident report waiting to happen.
If leadership wants “realism,” scramble and subset until rows cannot identify customers.
State that rule at the start so enthusiastic detectives do not export live warehouses “for practice.”
Building a quarterly case rotation
Retire cases after two cohorts so answer keys do not circulate. Write new sabotage with the same skill targets.
Invite alumni of the onboarding game to design one planted error for the next cohort. Teaching locks the lesson.
Stakeholder cameos that raise the stakes
Invite a finance partner to play “client” for the last ten minutes. Teams must explain one defect in business language, not only in Excel menu names. The cameo teaches translation, which is half of analytics trust.
If finance cannot attend, record a sixty-second voice note reacting to a wrong revenue total caused by the text-zero bug. Hearing a human react lands harder than a rubric line.
Post-game kit every analyst keeps
End by exporting a personal checklist tab: encoding, delimiter, header integrity, date locale, primary key uniqueness, null semantics, and row-count reconciliation versus the source ticket.
Require that checklist on the first three real imports after onboarding, signed off by a buddy. Habits form under light supervision more than under a single game high.
Archive the sabotaged file and the facilitator key in a private drive with a retirement date. When the case expires, rewrite at least three planted bugs so Slack screenshots from older cohorts cannot spoil the hunt.
CSV detective hour will not replace deep analytics craft—but it stops new teammates from trusting the first import wizard smile.


