AI-Aware Assessment Design (Process-Based Evaluation)

Teaching Problem

Traditional take-home exams and generic test-bank questions are increasingly vulnerable to AI completion. AI detection tools (Turnitin, GPTZero) have documented high false-positive rates and should not be used as primary evidence of academic dishonesty. The deeper problem: assessments designed around ‘compute the answer’ are no longer valid measures of student learning in an AI-available world.

How AI Addresses the Problem

Redesign assessments to emphasize process documentation, assumption justification, interpretation, and oral verification. The pedagogical shift: from ‘can the student compute this?’ to ‘can the student explain why this answer matters, what could change it, and whether to trust it?’ AI becomes a tool for generating individualized assessment variants at scale and for creating error-detection exercises that test higher-order thinking.

Deployment Details

Courses

All finance courses (UG and MBA)

Tools

ChatGPT (for generating variants), parameterized Excel/LMS templates, in-person oral components

Student Workflow

1. Receive individualized case parameters (different cash flow profiles, tax rates, betas, industry contexts) so each student works with unique numbers.
2. Complete quantitative analysis (with or without AI, per course policy clearly stated in the syllabus).
3. Write a justification memo (1–2 pages): explain each key assumption, interpret results in business terms, and identify the single assumption most likely to change the decision.
4. If AI was used at any stage, submit full prompts and outputs with annotations explaining what you accepted, what you modified, and why.
5. Complete a short oral check-in (3–5 minutes) with 2–3 targeted questions: ‘Walk me through your discount rate choice,’ ‘What happens to your recommendation if revenue growth is 2% instead of 5%?’

Sample Prompts

Generate 5 distinct capital budgeting case variants for an upper-division Corporate Finance exam. Requirements for EACH variant: (1) Different industry (manufacturing, healthcare, technology, retail, energy), (2) Project life between 3-7 years, (3) WACC between 8-12%, (4) Distinct cash flow patterns (one with heavy upfront investment and back-loaded returns, one with steady cash flows, one with a mid-project expansion option), (5) Include one variant where NPV is slightly negative (-$50K to -$200K) to test whether students recommend rejection. For each: provide complete cash flow tables, stated assumptions, and 3 interpretation questions requiring judgment, not just computation.

Write a 250-word equity research summary for a fictional mid-cap SaaS company called CloudMetrics Inc. Embed exactly 3 analytical errors that a well-prepared Corporate Finance student should catch: (1) a terminal growth rate of 6% in an industry with GDP-level long-term growth, (2) a P/E ratio applied to EBITDA instead of earnings, and (3) no mention of key risk factors (customer concentration, competitive entry). Make the prose confident and professional so the errors are not obvious. Do NOT label the errors.

Assessment Approach

• Computation accuracy20%
• Assumption justification quality25%
• Interpretation of results in business context25%
• AI critique and disclosure quality15%
• Oral defense performance15%

Publish the rubric in advance so students understand that explanation and interpretation carry more weight than getting the number right.

Grade-band guidance — Assumption Justification (25%): Full credit requires the student to explain why their specific discount rate reflects this company's risk profile, not just the industry average — e.g., citing the firm's leverage ratio, recent earnings volatility, or credit rating. Partial credit if the student states assumptions but offers no justification. Minimal credit if assumptions are asserted without any business rationale. The oral check-in for large sections can be conducted asynchronously: a 90-second video submission where the student walks through their two most consequential assumptions works well as an alternative to synchronous meetings.

Skills Developed

Evidence

Daigle, Li & Li (Monmouth, 2024) tested ChatGPT-3.5 on principles-of-finance test-bank questions. Abeysekera (2024) found ChatGPT scored at the 80th–90th percentile on introductory financial accounting assessments. OpenAI (2023) explicitly notes AI detectors are unreliable and recommends designing assessments accordingly. Bowen & Watson (2024, 2025) provide the progression framework: AI-prohibited to AI-transparent to AI-required.

Risks and Safeguards