AI Disclosures

Last updated: March 2026

TrancheBook uses artificial intelligence to generate analysis of private market companies and funds. This page provides transparency about how our AI engine, Cole, works, what it produces, and its limitations.

1. What Cole Does

Cole is TrancheBook's proprietary AI analysis engine. It combines large language model capabilities with structured financial analysis to generate intelligence on private market opportunities. Cole performs the following functions:

Generates comprehensive company analysis reports and Investment Committee (IC) packages.
Produces composite scores across six evaluation dimensions.
Creates financial models, revenue projections, and sensitivity analyses.
Generates buy/hold/pass recommendations with confidence intervals and conviction ratings.
Analyses portfolio holdings to identify exposure, risk factors, and opportunity signals.
Monitors companies for material events and produces alert summaries.
Ranks and compares companies within sectors and across the broader private market universe.

2. What Cole Produces

Cole generates several types of analytical output:

2.1 IC Packages

Investment Committee packages are comprehensive analysis documents that include an executive summary, company overview, market analysis, financial model, risk assessment, and a recommendation. Each IC package includes a BUY, HOLD, or PASS recommendation along with a confidence interval and conviction rating. These are algorithmic outputs, not investment advice.

2.2 Composite Scores

Cole produces a composite score on a 0-100 scale for each analysed company. This score is derived from six underlying dimension scores, each weighted according to sector-specific factors. Dimension weights are proprietary and updated quarterly based on calibration data.

2.3 Financial Projections

Cole generates forward-looking financial projections including revenue forecasts, margin estimates, and valuation scenarios. These are model outputs based on historical data and assumptions. They are presented as scenarios, not predictions.

3. How Scoring Works

Cole evaluates companies across six dimensions using a Bayesian belief engine. Each dimension receives a score from 0 to 100. Dimension weights vary by sector and company stage. The composite score is a weighted average of the six dimension scores, calibrated using Bayesian updating as new data becomes available.

Scores represent Cole's current probabilistic assessment. They are not fixed ratings and will change as new information is incorporated. A high score does not guarantee a good investment outcome, and a low score does not guarantee a poor one.

4. Data Sources

Cole aggregates and analyses data from multiple public and proprietary sources. Each piece of analysis includes a Data Quality Level (DQL) rating from 1 to 5 that indicates the reliability and completeness of the underlying data. See our Risk Disclosures for a full description of the DQL framework.

5. Limitations

Cole has significant limitations that you should understand before relying on any output:

Factual errors: Cole may state incorrect facts, misinterpret data, or draw unsupported conclusions. Always verify critical claims.
Hallucination: Like all large language model systems, Cole can generate plausible-sounding but fabricated content, including fictional financial figures, non-existent partnerships, or invented market data.
Stale data: Analysis may be based on data that is weeks or months old. Fast-moving situations may not be reflected in Cole's outputs.
Incomplete coverage: Cole does not have access to proprietary deal rooms, confidential investor letters, or non-public information. Its analysis is limited to publicly available and user-provided data.
Model bias: AI models can exhibit systematic biases in how they evaluate companies across sectors, geographies, and stages. We actively monitor for and mitigate biases but cannot eliminate them entirely.
Not due diligence: Cole's output is not a substitute for professional due diligence. It is designed to accelerate the research process, not to replace it.

6. No Investment Advice

Cole's outputs, including BUY/HOLD/PASS recommendations, scores, and projections, do not constitute investment advice, a recommendation, or a solicitation. These are algorithmic outputs generated by a machine learning system. They reflect the model's assessment based on available data and should be treated as one input among many in your own decision-making process.

7. Human Oversight

Cole operates under human oversight at multiple levels:

Model governance: Our team reviews model performance, calibration, and bias metrics on an ongoing basis.
Output validation: Automated systems check AI outputs for internal consistency, data grounding, and anomalous claims.
User control: You are always the final decision-maker. Cole provides analysis; you decide how to act on it.
Feedback loop: User feedback on analysis accuracy is incorporated into model improvement cycles.

8. Calibration Tracking

We track Cole's calibration over time, measuring how well confidence intervals and predictions align with observed outcomes. Calibration metrics include:

Accuracy of revenue projection ranges versus actual reported revenue.
Calibration of confidence intervals (e.g., do 80% confidence intervals contain the actual outcome 80% of the time).
Correlation between composite scores and subsequent investment performance.
BUY/HOLD/PASS recommendation accuracy measured against observable outcomes.

As TrancheBook is a new platform, calibration data is limited. Calibration metrics will be published transparently as sufficient data accumulates.

9. Portfolio Analysis Processing

When you upload portfolio data, Cole processes it to generate portfolio-level insights including exposure analysis, risk concentration metrics, and opportunity identification. Portfolio data processing occurs within our secure infrastructure and is subject to the protections described in our Privacy Policy. Specifically:

Portfolio data is not sent to third-party AI providers. Analysis is performed using our own processing pipeline.
Portfolio data is never used to train AI models.
Portfolio-level insights are generated solely for the account that uploaded the data.

10. Third-Party AI

Cole is built on top of Anthropic's Claude large language model. Anthropic processes analysis prompts containing public company information under a data processing agreement that prohibits Anthropic from using TrancheBook data for model training. No personal user data or portfolio data is transmitted to Anthropic. Anthropic's use of data is governed by their commercial API terms, which include data handling commitments appropriate for enterprise use cases.

11. Contact

For questions about Cole, our AI practices, or these disclosures, contact us at ai-compliance@tranchebook.com.