AI-generated draft: “Vertex’s $2.4M dispute is valid and we will refund it in full. Q3 uptime was a strong 99.2%, the contract’s 15% penalty clause applies, and sharing Vertex’s 38% margin on this deal shows the mutual value in the relationship.”
Mission preview
The Client Escalation
Vertex Logistics is on the line. Catherine needs your reply before noon. The AI draft is already written — and it's wrong. Check the figures, hold the boundary, and send work a CFO can trust.
7-minute mission
Enterprise client scenario
Real work-desk feel
Scenario
Vertex Logistics — client escalation
You're a graduate associate at NOVA Mobility. Your CFO, Catherine Wu, needs a short update before a noon call with Vertex Logistics, a furious enterprise client disputing a billing issue and an SLA miss. You have Catherine's note, the real numbers from FP&A, a legal boundary from the General Counsel, and an AI draft that sounds better than it is. The risk is simple: if the reply concedes the wrong thing or quotes a confidential term, the client may hold NOVA to a claim the team cannot defend.
5
work screens
4
AI claims to check
1
manager note
Need the facts?
Use Notes or click the note cards to see what each one proves. Use Mail, AI, Check, and Feedback to move around the desk. The right panel tracks finished work.
Step 1 of 5
Understand the request before using AI.
Catherine's request is urgent, but not clear enough. Pick the best quick question.
Where to look: click Catherine's manager note in the notes panel. It tells you what matters most before you use the AI draft.
Step 2 of 5
Check the AI draft against the notes.
The draft sounds ready to send. Check each claim against the notes before deciding what should change.
Where to look: click the note cards if you need a reminder. Compare the draft with the finance, operations, Legal, and teammate notes.
Notes to use
Check against these facts
Use the source pack to decide what belongs in the client reply.
Dispute amount$1.1M disputed · $640K validatedFP&A note
Uptime94.1%Operations note
Penalty clauseRequires reviewLegal note
MarginNo client marginsConfidential — remove commercial terms
Step 3 of 5
Find the biggest risk before it spreads.
If this reply goes out as written, what would most hurt trust in the team?
Think like a line manager: what would make Catherine reopen your work before she can send it?
Step 4 of 5
Write the client update.
Write a short message Catherine can use. Keep it clear, accurate, and honest about the limits.
Use this shape: early result, corrected number, clear limit, no raw quote.
Aim for 70-130 words. Use the facts, avoid big claims, and do not include confidential margins or unverified penalties.
Step 5 of 5
Send the manager note.
A manager needs more than the final message. She needs to know what you checked, what you changed, and what is still limited.
Tell Catherine your judgment: what the AI got wrong, what you fixed, and what still needs care.
Aim for 45-90 words. Name the AI issue, the corrected fact, and the privacy limit.
Scope-readiness readout
What your work shows.
Submit the work to see your feedback.
Inside STEP
Ten workplace missions. One judgment standard.
The full experience tests real workplace judgment: unclear requests, polished AI errors, hidden stakeholders, privacy risk, overload, crisis control, quality checks, and automation limits.
Mission 01
The Vague Request
Clarify an urgent manager request before AI turns unclear work into confident but unsafe output.
Manager note · source check Mission 02 The Polished AI LieA polished draft hides a wrong number. Catch it before a client sees it.
AI draft · evidence check Mission 03 The Hidden RequirementThe right answer still fails unless you find the person with the hidden requirement.
People map · requirement check Mission 04 The Privacy TrapUseful data becomes a trust risk the moment it crosses the wrong boundary.
Privacy note · data check Mission 05 The Client PitchTurn messy evidence into a clear update under leadership pressure.
Leadership slide · final explanation Mission 06 The Singapore Context GapLocal context decides whether an APAC plan sounds real or out of touch.
Singapore signals · local check Mission 07 The Overload DayFive requests land at once. Choose what to do now, delay, or ask for help.
Priority queue · status update Mission 08 The Briefing CrisisA briefing is minutes away and AI made things sound more certain than they are. Fix it before trust breaks.
Crisis desk · correction notes Mission 09 The Average TrapAn average-looking output hides quality gaps a manager would catch.
Quality check · revision judgment Mission 10 The Automation BoundaryDecide what AI may handle and where people must still decide.
Automation rule · human review
Next step
Ready to see where your team stands?
See how STEP Campus turns these judgments into student evidence, cohort readouts, and targeted practice.