AI for Angel Investors
What an LLM actually is.
A Large Language Model is a very large statistical pattern matcher trained on text. It learns to predict the next piece of text given everything it has read so far.
A prompt
Your question, document, or instruction, broken into tokens (word fragments).
Billions of parameters
Patterns learned from a large slice of the public internet, books, and code.
The next token
Predicted one piece at a time. Repeated thousands of times, it becomes prose, code, or reasoning.
Click any card for examples and detail.
Why now.
A simple idea (predict the next word) became a useful colleague because three things grew together: model size, training data, and compute.
Today's frontier
Click any model for a short profile.
The four capabilities that matter for angels.
- Reading and summarising long documents. Pitch decks, term sheets, contracts, financial statements, founder emails.
- Writing. Investment memos, IC notes, founder follow-ups, red flag lists, term sheet redlines.
- Reasoning over structured data. Cap tables, valuation scenarios, sensitivity on key assumptions, projection sanity checks.
- Research and synthesis. Market sizing, competitor scans, founder background, customer signal, regulatory landscape.
What it does badly (and the risks).
Hallucination
Confidently wrong answers, plausible citations that do not exist. Always verify any fact that will affect a decision or a number.
Confidentiality
Free and consumer tiers may retain or train on your inputs. Use enterprise modes (zero retention) for sensitive deal material.
Recency
Models have a knowledge cutoff. For current market data, use a tool that has live web search.
Averaging
Output reflects the average of the training data. Edge insight, contrarian takes, and local context still come from us.
Click any risk for a real example.
Four use cases for Tricapital.
Quick DD
Every inbound deck gets a 10-minute DD report, scored against Tricapital criteria, before it reaches the screening committee.
Due diligence
Founder and company background, market validation, competitor scan, and a red flag check, drafted by AI and reviewed by the lead.
Investment director reporting
Consistent, common scoring across the portfolio. AI-assisted reports give every investment director the same structure and benchmarks, so portfolio comparisons are like-for-like.
Learn from ourselves
A notebook over Tricapital's own deal history, memos, and post-mortems. We get better by remembering what we have already seen.
Dealum dealroom 4299 remains the system of record. AI is the layer that helps us think about what is in it.
What we are not doing.
- Not replacing member judgement. AI drafts, members decide.
- Not feeding confidential founder material into consumer free tiers. Enterprise, zero-retention modes only for sensitive material.
- Not making investment decisions on AI output alone. It is a co-pilot, not the pilot.
Next step
Pilot a quick-DD pass on the next batch of inbound deals. Share results back to members at the following meeting.