How I used a 6-step AI-assisted framework to pressure-test a $10,000 micro-business idea before any capital was committed.
Executive Summary
| Project | Venture Feasibility & Risk Assessment |
| Industry | Micro-business / early-stage venture (applicable across verticals) |
| Timeline | Single working session |
| Deliverable | Structured go/no-go recommendation with alternative paths |
| Workflow | Human-AI collaboration, 6 stages |
| Core method | Boundary-condition framing + adversarial risk interrogation |
Challenge
A friend was considering a full-time micro-business venture: roughly $10,000 in capital, no prior industry experience, targeting $1,600–$2,600 in monthly net profit. He had already started pricing out equipment and scouting locations.
His real anxiety wasn’t the product — it was whether the capital would ever come back, and whether he’d discover six months in that he’d worked hard for nothing. Most people facing this question default to discussing product details. That’s the wrong starting point. The real question comes first: does this business survive contact with reality?
Context
This is a common decision-quality problem, not a product problem: a resource-constrained, time-pressured decision made with incomplete information. The same pattern applies well beyond food carts — side-hustle evaluation, career pivots, even investment decisions share the same structure.
Diagnosis
Without hard numbers, asking AI to “write a business plan” produces generic, encouraging output — technically complete, practically useless. The fix isn’t a better prompt. It’s forcing precision at the input stage, before AI generates anything.
System Design
The workflow runs in six stages:
1. Define Boundary Conditions — Lock down the hard constraints before any analysis: capital available, category, relevant experience, target return, time commitment, and location status. Six data points, all confirmed with the client directly.
2. Convert Intuition into a Unit-Economics Model — Break the “feels viable” intuition into a numeric model: price per unit, margin range, resulting daily volume required to hit target profit. This step alone usually surfaces the first disqualifying constraint — a required output rate that no location realistically supports.
3. Force Adversarial Risk Review — This is the step that changes everything. The first prompt asked AI to “assess feasibility” — the output was encouraging and complete-looking, but named no real risks. Reframing the ask as “assess this from the perspective of the most demanding investor, and find the hidden reasons this fails” produced a completely different quality of output. This became a standing step in every subsequent workflow.
4. Propose Risk-Reduced Alternatives — Rather than a single go/no-go answer, generate a tiered set of paths: high-risk (commit fully now), medium-risk (test at small scale first), low-risk (validate demand before any capital commitment).
5. Stress-Test the Capital Buffer — Model how many months of runway remain after fixed costs, under a pessimistic early-traffic scenario.
6. Deliver a Conditional Recommendation — Not “yes” or “no,” but a recommendation with explicit preconditions attached.
Workflow Diagram
Define Boundary Conditions
↓
Build Unit-Economics Model
↓
Adversarial Risk Review
↓
Propose Risk-Reduced Alternatives
↓
Stress-Test Capital Runway
↓
Conditional Decision
Decision Points (Where the Value Actually Is)
- Deciding the prompt needed reframing, not just re-running. The first AI response wasn’t wrong — it was answering the wrong question. Recognizing that “is this good?” and “where does this fail?” produce fundamentally different output quality was the actual insight, not a prompting trick.
- Deciding what counts as “real” runway. The initial capital allocation looked sufficient on paper. Modeling working capital as a buffer against 1–2 months of underperformance — not just startup costs — reframed the entire risk picture.
- Deciding opportunity cost belonged in the model. A $1,300–$2,000/month “net profit” looks attractive in isolation. Netting it against foregone salary reduced the real incremental return to a number that changed the client’s risk tolerance for the decision entirely.
Prompt Principles
- Boundary First — No analysis begins until every hard constraint is confirmed as a specific number, not a range guess.
- Role Reversal — Always run a second pass in an adversarial role (“demanding investor”) before treating any assessment as complete.
- Ranges Over False Precision — Every projected figure is expressed as a range, never a single confident number — this better reflects real early-stage uncertainty and avoids overstating model reliability.
- Conditional Output — The deliverable is never a simple yes/no; it’s a recommendation with explicit preconditions.
Full prompt template: [GitHub link]
Results
- A capital-intensive, high-risk commitment (signing a fixed location lease) was replaced with a staged validation plan
- Three concrete risks were surfaced that the initial “encouraging” draft had missed entirely: capital buffer fragility, effort misallocation, and hidden opportunity cost
- A reusable 6-stage framework that now applies beyond venture evaluation
Reusable Framework
This workflow isn’t specific to food-service ventures. The same structure — Boundary Conditions → Model → Adversarial Review → Alternatives → Stress Test → Conditional Decision — has since been applied to:
- Side-hustle and freelance-offer evaluation
- Career-pivot decisions
- Early-stage investment screening
The deliverable changes. The decision structure doesn’t.
Appendix
Full prompt template and reproducible workflow files: [GitHub link]
This is Design Note #002 in an ongoing series documenting Human-AI collaboration workflows for content and business systems. Want to discuss how this framework applies to your own decision? Get in touch.
