∞Direct answer
Direct answer
Due diligence means checks before you invest. Master Collective's AI due-diligence feature can sort deal files. It can find facts and raise questions before an introduction. It is not full investor due diligence. It is not regulated advice. It does not decide to invest. People must check facts. A human must review it.
1. Set the task before file upload
Set the company, round, due date, approved files, and key questions. Give AI small tasks. It can list files, find terms, match sums, or draft questions. Do not ask it for a final invest or decline call.
Name a person for finance, law and IP, sales checks, data safety, and the deal lead. This is a work guide. It is not expert advice.
2. Make a proof map
For each claim, save the file, page or slide, source words, version, date, and trust note. Link to the source file. A reviewer must be able to open it again.
NIST lists data source logs, human checks, IP, and test logs as key AI controls. [1] Keep the proof trail. It is a real control, not just a source list.
3. Parse first. Check next.
Use AI to find names, dates, duties, sums, clients, and gaps. Check the pitch deck, cap table, contracts, and cash files. A cap table is a list of who owns shares. Mark each point as found, inferred, or missing. An inference is not proof.
Check key points in the main record. Ask the team or an outside party when needed. Fix the proof map. A smooth AI story is not proof. A missing file is not a fact.
4. Match money terms first
Make a metric sheet. Add the meaning, time, money unit, book rule, source, and writer. Split bookings, contract sales, booked sales, cash in, gross margin, and burn. Like labels may mean different things.
Match the pitch, accounts, bank or ledger support when given, cap table, and plan. Do not blend gaps. Find the cause. Log the proof, reviewer, and state: fixed, open, or not found.
5. Send law and IP points to people
AI can list company files, founder and contractor agreements, and IP assignments. It can list open-source notes, licences, client terms, privacy notes, and key disputes. It can flag missing signatures, name gaps, end dates, and terms that need review.
A lawyer must judge who owns what, legal force, licence risk, privacy duties, and local law. NIST also flags IP and private or key data. [1] Do not upload locked files without an approved data path.
6. Test claims and guard the work
Ask users what they bought and who used it. Ask what changed, what is still open, and if they would buy again. Check consent, the link to the company, and any conflict before you use a client or partner reference.
Keep untrusted file text apart from system rules. Prompt injection can make a model skip its task or leak data. NIST calls for tests of such attacks. [1] Use least access. Limit data. Check vendor terms for storage, training, place, and deletion. The ICO UK guide is under review. Get local advice when needed. [2]
7. End with a risk log and vendor check
Keep a risk log with the issue, source link, impact, owner, next step, state, and last check. Save prompts, model type, source set, output, edits, and notes. British Business Bank says due diligence uses desk work, talks, refs, and law checks. AI should help with each check. [3]
Ask an AI vendor: Can it cite sources? Can access split by deal and role? Are storage and deletion terms clear? Are sub-tools and training terms clear? Can the team export an audit log? Are attack controls clear? Are harm steps clear? Can a person correct or hold output? If not clear, log a vendor due-diligence question.
FAQ
Frequently asked questions
Can AI perform investor due diligence alone?
No. It can sort and read files. People must check key facts, get legal or expert help when needed, and make the investment decision.
What proof should an AI finding have?
Save the file, page or slide, source words, version, date, reviewer, and a tag: found, checked, guessed, or open.
What if a file is missing?
Log it. Ask for the file or a reason. Note the limit it makes. Do not treat the gap as good news.
Is an AI tool safe for deal data by default?
No. Check access, data flow, storage, deletion, sub-tools, training, safety, logs, and privacy first.
REF
Sources
- 1
National Institute of Standards and Technology
Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile ↗ - 2
Information Commissioner's Office
Guidance on AI and data protection ↗ - 3
British Business Bank
Due diligence ↗