∞Direct answer
Direct answer
Use AI to make a short, linked deal pack. Give it a set list and set files. Ask it to show sources, gaps, and conflicts. A reviewer must check key facts. The group must choose. NIST says AI that writes text can sound sure but be wrong, with false source links. [2]
Set a narrow job before you use AI
Start with one job. For example, ask AI to draft a deal brief from named files. Do not ask it to pick a deal or guess a return.
Write the questions first. They may cover fund fit, proof of need, team, cash, terms, conflicts, and open checks. This gives each deal the same start.
NIST gives four risk steps: Govern, Map, Measure, and Manage. Use them in a small way. Set an owner. Map the use. Test the output. Then act. [1]
- Purpose: make a first draft, not a decision.
- Input: name the files, sources, and allowed data.
- Output: require a claim, source, date, limit, and open question.
- Owner: name the person who can stop the workflow.
Build one evidence card for the meeting
Give the tool a fixed card. Put each key claim on its own line. Link it to the source file and check date.
Keep facts apart from views. A signed buyer order can back a sales claim. A founder view on later demand is a view. Both may matter. They are not the same proof.
Do not trust a link before you open it. NIST says teams should check sources and citations in AI output. [2]
- Claim: the exact point the committee must assess.
- Evidence: a source link, short quote, and check date.
- Limit: what the record does not show.
- Status: confirmed, unclear, disputed, or not checked.
Make human review a real control
Give a reviewer time and power to change the card. The reviewer should read the key source, fix errors, and flag facts we do not know. They should also be able to pause the deal.
Do not make a person a rubber stamp. People can trust AI too much. IOSCO says human checks need the right skills, clear ownership, and care not to trust tools too much. [4]
Raise items that can change the choice. Examples include a fund conflict, a cap table gap, a buyer claim, a data-rights issue, or a key term. Get legal, tax, or other expert help when the deal needs it.
- Reviewer: name and date.
- Change: what the reviewer corrected.
- Escalation: the issue, owner, and due date.
- Override: why the person changed an AI result.
Test changes and keep a short record
Test a new prompt, model, or rule on old deals first. Compare its draft with files people checked. Look for lost facts, weak sources, and changed answers.
Save the tool and prompt version. Save the file set, check date, and main edits. A later reader should see how the pack was made.
ESMA says some EU firms that use AI in investment work need controls, tests, checks, and records of use, data sources, and changes. [3] These are useful ideas. They are not a law for every fund or place.
- Test set: old deals with checked source records.
- Measure: missing facts, source errors, and reviewer edits.
- Change log: prompt, model, source set, and rule version.
- Stop rule: pause use when a material error appears.
FAQ
Frequently asked questions
Can AI draft a deal memo?
It can make a first draft from named files. It should show sources, gaps, and limits. A person must check key claims before the meeting. AI that writes text can make false facts and links. [2]
Should an AI score choose the deal?
No. A score can sort questions. It cannot fix a conflict, lost proof, or key risk. Keep the choice with the group.
What should we keep after the meeting?
Keep the choice, why it was made, source links, open checks, reviewer edits, and tool version. This helps the team test and improve the work. ESMA asks some firms in its scope to log AI use, data sources, and changes. [3]
Can we say that AI chose our deals?
Do not say this unless it is true and you can show proof. In 2024, the SEC charged two firms for false claims about AI use. Rules differ by place and work. [5]
REF
Sources
- 1
National Institute of Standards and Technology
AI Risk Management Framework ↗ - 2
National Institute of Standards and Technology
Artificial Intelligence Risk Management Framework: Generative Artificial Intelligence Profile ↗ - 3
European Securities and Markets Authority
Public Statement on Artificial Intelligence and Investment Services ↗ - 4
International Organization of Securities Commissions
Supervisory Toolkit for AI Use in Capital Markets ↗ - 5
U.S. Securities and Exchange Commission
SEC Charges Two Investment Advisers with Making False and Misleading Statements About Their Use of Artificial Intelligence ↗