Methodology

Overview

UpsideList estimates the potential equity upside of thousands of private tech companies using a two-stage AI pipeline backed by deterministic heuristics and guardrails. Our goal is to help job seekers evaluate whether a company's equity is likely to grow relative to its current valuation.

Pipeline Architecture

Each company is processed through two search-grounded Gemini AI calls:

• Call 1 — Data Extraction: Gemini with search grounding extracts real-time financials, valuation, funding history, competitive landscape, market context (multi-segment TAM), and risk signals from publicly available sources. Returns structured data with cited URLs.
• Call 2 — Equity Analysis: A second search-grounded Gemini call receives the extracted data and produces an equity analysis: expected upside percentage, verdict, bull/base/bear scenarios with probabilities, risk factors, upside drivers, and a recommendation summary.

Both calls use Gemini's search grounding feature, meaning the AI retrieves and cites current web sources rather than relying solely on training data.

Deterministic Guardrails

AI outputs pass through a series of deterministic sanity checks before being stored:

• Revenue exceeding $100B is nulled (likely GMV confusion)
• Valuation below total funding is adjusted to funding × 1.5
• Expected upside is clamped to −100% to +500%
• Scenario probabilities are normalized to sum to 1.0
• If the AI's expected value math diverges >20% from the weighted scenario calculation, it is recalculated deterministically
• Verdict, confidence, and risk level must be valid enum values
• Early-stage companies (pre-Series A) receive a stage discount to account for higher failure rates

These checks ensure that AI errors — hallucinated numbers, math mistakes, or impossible outputs — are caught and corrected before reaching the site.

Ranking

Companies are ranked by expected upside percentage in descending order. There is no composite formula — the ranking is purely the AI's estimated 2-year upside after guardrails have been applied. This keeps the methodology transparent and auditable.

Equity Mechanics

Beyond headline upside, we assess factors that affect what employees actually take home:

• Preference stack risk: How much investor capital gets paid out before common shareholders in an exit
• Funding intensity: Total capital raised as a share of valuation — higher means more investor ownership
• Dilution risk: Likelihood of future fundraising that would shrink your ownership percentage
• Secondary liquidity: Whether you can sell shares before an IPO or acquisition

Data Freshness

The pipeline runs weekly. Each run processes all active companies with fresh search-grounded calls, so the data reflects what Gemini can find on the web at that time. Change tracking compares each run to the previous one and records deltas in upside, rank, revenue, and valuation.

Limitations

• This model has not been validated against long-term outcomes
• Private company data is inherently limited and may be outdated or incomplete
• AI-generated analysis may contain errors or hallucinations despite guardrails
• Valuation estimates are rough approximations, not precise calculations
• The model does not account for non-public information such as internal metrics, pending deals, or management quality
• Search grounding quality varies — some companies have more public information than others