VC Portfolio Risk: Why the Highest-NPV Startup Isn't Always the Best Bet

Goal

A venture capital fund is evaluating five early-stage startups, each starting with the same $5M initial balance. The companies span food-tech, education, enterprise AI, energy storage, and biotech — deliberately diverse to stress-test the evaluation method. The question isn't which company has the best pitch deck. It's which investment delivers the best risk-adjusted return over a 5-year horizon. We built two sets of models: traditional deterministic projections (the kind you'd build in Excel — all fixed inputs, no variance) and probabilistic Foresight simulations with Monte Carlo analysis. The gap between what the spreadsheet says and what the simulation reveals is the entire point of this case study.

Challenge

Every startup's financial model looks good in a spreadsheet. Fixed revenue growth, predictable expenses, smooth hockey-stick projections — that's what founders present and what many VCs evaluate against. The problem is real businesses don't follow a single path. Revenue ramps are uncertain. Regulatory approvals have unknown timing. Customer acquisition costs fluctuate. And most critically — a company that goes bankrupt in Year 2 never gets to enjoy its Year 5 projections, no matter how impressive they looked on slide 14. The five companies under evaluation: 1. **Food-Tech Central Kitchen** — A proven food operation with $400K/mo satellite sales and a corporate catering stream. Expanding via a centralized production model. Revenue is real, growth is incremental. 2. **Premium Daycare & Early Education Network** — $380K/mo gross tuition fees with after-school programs. Stable enrollment in a high-demand market, but significant facility rent ($90K/mo), staff payroll ($150K/mo), and regulatory requirements. 3. **B2B Enterprise AI Agent Platform** — The highest-ceiling company in the portfolio. $400K/mo ARR subscriptions plus implementation fees. Massive TAM, but heavy burn: $350K/mo engineering payroll, $100K/mo cloud compute, and enterprise sales commissions. 4. **Solid-State Energy Storage** — Deep-tech play with $100K/mo pilot grid deployments, $50K/mo government R&D grants, and a large IP licensing stream ($800K/mo). R&D operations at $200K/mo with hardware prototyping costs. 5. **AI-Driven Drug Design** — The moonshot. Pharma R&D grants ($250K/mo), patent licensing ($50K/mo), and a massive one-time Phase II buyout payment ($25M) — but only if the clinical trial succeeds. $250K/mo compute and wet lab burn, plus trial operations. In a deterministic Excel model, all five companies use fixed monthly amounts with no variance. Companies 3 and 5 dominate on NPV and IRR. The question is whether that ranking holds when you account for what can go wrong — and when it goes wrong.

Model Summary

We built 10 financial models — two for each company. **Approach 1: Deterministic (Excel-style)** — 7 events each Each company gets a single-path projection using fixed monthly amounts: fixed revenue, fixed expenses, no uncertainty, no risk events. For example, Food-Tech shows $400K/mo satellite sales and $120K/mo ingredients — every month, guaranteed. This is what most pitch decks contain and what many VCs build their DCF models around. **Approach 2: Foresight (Monte Carlo)** — 8-9 events each The same five companies, same base economics, but now with realistic uncertainty: • **Distributions replace fixed amounts**: Food-Tech's satellite sales become Normal($400K, σ=$50K). Corporate catering becomes Triangular($20K/$80K/$150K). B2B AI's ARR uses Normal($400K, σ=$120K) — that $120K standard deviation reflects real enterprise sales volatility. Energy Storage's pilot deployments use Triangular($50K/$100K/$200K). • **Risk signals model disruptions**: Each Foresight model adds 1-2 signal events that can trigger cascading consequences via logic gates. Food-Tech has a Supply Shock signal. B2B AI has a Base Model Shift signal that triggers a $300K Platform Rewrite. Drug Design has a Phase II Trial signal that gates the $25M buyout payment — if the trial fails, that revenue never arrives. • **Expenses carry variance too**: B2B AI's cloud compute is Normal($100K, σ=$60K) rather than a flat $100K/mo. Drug Design's clinical trials are Triangular($70K/$110K/$180K). Ingredients for Food-Tech are Triangular($100K/$120K/$250K). **What stays the same**: $5M initial balance, 5-year simulation period, 10% discount rate, and the fundamental business model of each company. **What changes**: Every assumption a founder says 'we expect' becomes a distribution. That's a more honest representation of what early-stage financials actually look like.

Results

The comparison tells a clear story. ## Deterministic Results (5-Year, Fixed Inputs) All five startups were modeled with fixed monthly amounts — no variance, no risk events. The numbers below are what a standard Excel DCF would produce: | Company | NPV | IRR | P50 Balance | ROI | |---|---|---|---|---| | B2B AI Platform | $56.3M | 135.5% | $92.5M | 1,751% | | AI Drug Design | $9.0M | 48.3% | $24.0M | 380% | | Energy Storage | $8.6M | 39.8% | $24.8M | 396% | | Food-Tech | $2.0M | 29.2% | $13.8M | 177% | | Daycare | $1.9M | 28.6% | $13.7M | 174% | All five show 100% Prob Success, 100% Min Survival, and a "Strong Accept" DCF verdict. The spreadsheet says every investment works. B2B AI dominates with $56.3M NPV and 1,751% ROI. ## Monte Carlo Results (Probabilistic Inputs) The same five companies, same base economics, but with distributions replacing fixed amounts and risk signals modeling disruptions: | Company | NPV | IRR | Prob Success | Min Survival | P50 Balance | Verdict | |---|---|---|---|---|---|---| | B2B AI Platform | $41.5M | 113.2% | 100% | **70%** | $74.2M | Conditional | | Energy Storage | $19.0M | **88.0%** | 100% | **100%** | $34.8M | Strong Accept | | Daycare | $1.8M | 28.3% | 100% | **100%** | $13.5M | Accept | | Food-Tech | -$87K | 20.1% | 100% | **100%** | $12.8M | Accept | | AI Drug Design | -$2.4M | -8.7% | **34%** | 52% | -$2.2M | **Reject** | Notice what changed: B2B AI's NPV dropped from $56.3M to $41.5M and its Min Survival fell to 70%. Drug Design collapsed from $9.0M NPV to -$2.4M with only 34% success. Meanwhile, Food-Tech went from $2.0M to slightly negative, but kept 100% survival — the downside is contained. | Company | NPV | IRR | Prob Success | Min Survival | P50 Balance | Verdict | |---|---|---|---|---|---|---| | B2B AI Platform | $41.5M | 113.2% | 100% | **70%** | $74.2M | Conditional | | Energy Storage | $19.0M | **88.0%** | 100% | **100%** | $34.8M | Strong Accept | | Daycare | $1.8M | 28.3% | 100% | **100%** | $13.5M | Accept | | Food-Tech | -$87K | 20.1% | 100% | **100%** | $12.8M | Accept | | AI Drug Design | -$2.4M | -8.7% | **34%** | 52% | -$2.2M | **Reject** | Notice what changed: B2B AI's NPV dropped from $56.3M to $41.5M and its Min Survival fell to 70%. Drug Design collapsed from $9.0M NPV to -$2.4M with only 34% success. Meanwhile, Food-Tech went from $2.0M to slightly negative, but kept 100% survival — the downside is contained. ### Food-Tech Central Kitchen 100% probability of success, 100% min survival. NPV is slightly negative (-$87K) under uncertainty due to variance in ingredients and catering costs. The downside is contained — zero scenarios where the company goes bust. A stable, cash-flow-positive operation. ### Premium Daycare & Early Education 100% success, 100% min survival. NPV of $1.8M with a tight P5–P95 band ($13.2M–$13.8M). The Regulatory Audit signal can trigger a $75K compliance fine, but it's manageable. Stable enrollment in a high-demand market keeps the floor high. ### B2B Enterprise AI Platform Highest median NPV ($41.5M) and highest ending balance ($74.2M). But min survival drops to **70%** — in 30% of scenarios, the company hits insolvency mid-simulation before recovering. The $120K standard deviation on ARR combined with $350K/mo fixed engineering payroll creates a cash flow squeeze. The Base Model Shift signal can trigger a $300K platform rewrite, adding pressure during down months. ### Solid-State Energy Storage Strong NPV ($19.0M), **highest IRR (88.0%)**, and critically — **100% min survival**. The IP licensing revenue ($800K/mo) provides a stable floor, and the R&D Breakthrough signal creates upside without threatening solvency. This is the balanced play: strong returns with no path-dependent insolvency risk. ### AI-Driven Drug Design Only **34% probability of success**. Median ending balance is negative (-$2.2M). The Phase II Trial signal gates the $25M buyout — when the trial fails, the company burns through its $5M balance on $250K/mo compute costs. Runway exhaustion projected by September 2029. DCF verdict: **Reject**. ## Risk-Adjusted Composite Ranking When we apply a weighted composite score — NPV (25%), Min Survival (25%), IRR (20%), Success Probability (15%), Ending Balance (15%) — with a squared penalty on the survival metric: | Rank | Company | Composite Score | Key Factor | |---|---|---|---| | 1 | Solid-State Energy Storage | 0.82 | 100% survival + highest IRR | | 2 | B2B Enterprise AI Platform | 0.80 | Penalized for 70% min survival | | 3 | Premium Daycare | 0.55 | Stable but modest returns | | 4 | Food-Tech Central Kitchen | 0.48 | Safe but negative NPV | | 5 | AI-Driven Drug Design | 0.00 | 34% success, negative NPV | The B2B AI Platform doesn't disappear from the portfolio — it's still the highest raw-return investment. But it's not the safest primary allocation. The composite scoring reveals that Energy Storage delivers strong returns without the path-dependent insolvency risk.

Decision

## The Spreadsheet vs. The Simulation The spreadsheet said B2B AI Platform ($56.3M NPV, 135.5% IRR). The simulation said Solid-State Energy Storage ($19.0M NPV, 88.0% IRR, 100% min survival). This isn't because the AI platform is a bad company — it has the highest raw return potential in the portfolio. It's because deterministic models hide a specific type of risk: **what happens between now and the exit**. With Normal($400K, σ=$120K) monthly ARR and $350K/mo fixed engineering payroll, a few bad sales months can create a cash crisis that the fixed-amount Excel model never shows. ## Why Min Survival Changes Everything The Min Survival metric measures the **lowest probability of solvency at any point** during the simulation — not just at the end. B2B AI shows 100% success at Year 5 but only 70% min survival — meaning in 30% of scenarios, the company hits insolvency before recovering. That window is invisible in a standard DCF. ## Capital Allocation Strategy For a VC making real capital allocation decisions: - **Primary allocation → Solid-State Energy Storage** — Best risk-adjusted return. $19.0M NPV, 88.0% IRR, and 100% min survival. The IP licensing revenue provides downside protection while pilot deployments and grants add upside. - **Secondary allocations → Food-Tech and Daycare** — Stable businesses that anchor the portfolio floor. Both show 100% survival across all scenarios. - **Opportunistic allocation → B2B AI Platform** — Highest return potential ($41.5M NPV) and still a strong composite score (0.80). But 70% min survival means this is a calculated risk. Size the check accordingly. - **Option-value play → AI-Driven Drug Design** — Small check, binary outcome. 34% success rate, runway exhaustion by September 2029. The $25M Phase II buyout makes it high-reward if it hits. None of this is visible in a standard Excel DCF. The numbers are the same — it's the uncertainty around those numbers that changes the decision.