How to Humanize a Business Case Analysis After ChatGPT
ChatGPT case analyses invent market share and ignore exhibits. Lock every number from the assigned packet, name one recommendation with a tradeoff, then humanize synthesis once.
A business case analysis is a decision under constraint using only the packet. ChatGPT does the opposite by default: adds a competitor from training data, rounds revenue to a plausible percentage, and closes with “pursue a balanced, customer-centric strategy.”
Professors have taught the same case for years. They know Exhibit 4. Polished fiction still fails.
Human Writes is the voice pass after every number is packet-true. It will not invent a recommendation or fix a missing tradeoff. Related: case study assignment (same packet discipline, assignment framing), policy memo, MBA admissions essays (different artifact), ethical AI checklist.
What rubrics punish
| Element | Weak AI draft | Strong packet analysis |
|---|---|---|
| Evidence | “Margins look healthy” | Exhibit 2: gross margin 34% in FY22 |
| Recommendation | “Explore all options” | Choose Option B; defer Option A |
| Tradeoff | Ignored | Capacity at Plant 2 hits 94% if you choose B |
| Framework | SWOT wallpaper | One lens from the syllabus |
| Tone | Brochure | Names the protagonist and the decision date |
Detectors may flag smooth Discussion. Instructors flag invented numbers first.
Packet-only rule (non-negotiable)
Before any model help, build a packet fact sheet from the PDF your instructor assigned:
| Field | Your entry |
|---|---|
| Company + case year / decision date | From the narrative |
| Three defendable numbers | Revenue, share, cost, headcount - only from exhibits |
| Two constraints | Budget, regulation, capacity, brand |
| One competitor move in the text | Not from memory of the industry |
| Recommendation (one sentence) | What you would do Monday |
| What you will not invent | Outside market stats, fake interviews, Bloomberg color |
If a figure is not on an exhibit or in the case narrative, delete it. ChatGPT fills gaps with plausible lies.
Do not paste confidential employer data into a consumer chatbot. Use only the version distributed for class.
How “analysis” differs from a case write-up dump
Some courses ask for a long narrative. Others want a memo. Either way, analysis means you weighed options against criteria and can defend the choice in cold call.
| Weak pattern | Strong pattern |
|---|---|
| Summarize the story for three pages | One short context paragraph, then decision |
| List every exhibit | Cite the three exhibits that move the decision |
| “Both options have merits” | Pick one; state what you give up |
| Paste a generic Porter five-forces essay | Use the framework your professor graded last week |
If your prompt says “analyze,” do not deliver a book report. Deliver a choice.
Analysis structure that graders recognize
Adjust to your rubric; most business case analyses look like:
- Decision - what the protagonist must choose
- Criteria - from class (not a random framework dump)
- Options vs criteria - short matrix
- Recommendation - one path + implementation steps
- Risks - packet-backed, not generic “execution risk”
| Section | Humanize? | Why |
|---|---|---|
| Exhibit numbers, quotes | No | Accuracy |
| Criteria labels from class | No | Course vocabulary |
| Analysis narrative | After matrix locked | AI flatters every option |
| Recommendation prose | Yes, one pass | Often the stiffest block |
Paste analysis and recommendation paragraphs into Human Writes once. Put back every number exactly as printed.
Before and after
AI recommendation sludge:
The company should leverage synergies and digital transformation to unlock sustainable growth across key segments while carefully managing stakeholder expectations.
After packet facts + one pass:
Enter Segment B now and defer Segment A (Exhibit 3 share: B at 18%, A at 41% but declining). Plant 2 utilization is already 94% through Q4 (Exhibit 5); adding A this year breaks the capacity constraint the CFO flags on page 7.
Second version has an ask, numbers, and a cost from the file you were assigned.
Undergrad vs MBA / exec format
| Undergrad case analysis | MBA / exec write-up | |
|---|---|---|
| Length | Often 3 to 5 pages | Sometimes 1-page exec summary + appendix |
| Voice | Clear student analysis | “What would you do Monday?” |
| Evidence | Named exhibits | Same - no outside quotes |
| AI failure | Generic industry essay | Fake consultant brochure |
Same workflow either way: packet sheet, recommendation, one humanize pass, re-check exhibits. Assignment twin: business case study assignment.
Cold-call readiness check
Before you submit, ask whether you could answer these without opening ChatGPT:
- Which exhibit number backs your biggest claim?
- What is the strongest objection to your recommendation inside the packet?
- What happens to cash, capacity, or share if you are wrong for two quarters?
If you cannot answer, you still have a voice problem layered on an analysis problem. Fix the second first.
Workflow
- Read the case once without a model. Highlight every number.
- Fill the packet fact sheet by hand.
- Build a small options-by-criteria table from the syllabus.
- Draft analysis and recommendation from the sheet only.
- One Human Writes pass on narrative sections.
- Re-check every figure against the exhibit page.
- Disclose AI use if your syllabus requires it (ethics checklist).
Bottom line
A business case analysis is a test of packet discipline, not of how consultant-smooth your prose sounds. Lock numbers, constraints, and one recommendation first. Then polish voice. Human Writes helps the second step - not the first.