∞Direct answer
Direct answer
Ask for a short match record before an introduction. Show your brief, the firm facts used, the source and date for each key fact, gaps, and the rank reason. Name the person who can change it. NIST separates what happened, how a choice was made, and why it matters. [1] A score does not prove a firm fits your brief.
Ask what the match means
A match tells you to look closer. It is not a buy view. It may show only facts the tool had.
Start with your written investment thesis. This says what firms you want to back. Name stage, place, field, cheque range, and hard limits.
Ask which parts of the brief made the match. A useful answer names facts. It does not just give a score or label.
NIST says a record can show what happened. It says a clear reason can show how a choice was made and why it matters. [1]
- The written investor brief used at the time.
- The fields that supported the match.
- The fields that counted against the match.
- The fields that were not known.
Check the facts behind the rank
Read the source for each key fact. Check who put it out and when. A firm page may tell of a product. It may not prove sales, owners, or a fund plan.
Keep the claim apart from the source. Mark a client count as a claim if you lack proof under your check rule.
A good match record can say a fact is old, missing, or not checked. That is more useful than a neat number with no trail.
NIST says an open record can cover data, planned use, roll-out, and who made choices. [1]
- Fact used for the rank.
- Source link and source date.
- Date you checked the fact.
- Status: checked, claimed, old, or not known.
Look for the missing and the unfair
A blank field is not proof of poor fit. It may show a new firm, a closed process, or poor data.
Ask if the tool treats a blank as zero, a loss, or no result. Ask if small field changes shift the rank a lot.
Check a few firms that look like your target but sit low on the list. This can show if the brief or data rule leaves out good cases.
NIST names fair use, privacy, work that holds up, open records, and clear reasons as key traits. The mix depends on use. [1]
- Which fields are often blank?
- What does the system do with a blank?
- Which data sources are not covered?
- Can a reviewer correct a wrong field?
Keep a human choice at the gate
Set a simple rule for an introduction. A person should read the match record and choose if the firm gets contact.
Save the name, date, and reason for that choice. Save the reason for a pass or hold too. This saves a record for a later check.
The ICO guide says teams should choose clear reasons for the use and effect. It also covers data work, staff prep, and saved notes. [2]
A human check does not make data right. It gives you a point to question the data, rank, and next step.
- Review the fit facts before contact.
- Check conflicts and existing relationships.
- Choose introduce, hold, or pass.
- Write one plain reason for the choice.
Test the process and describe it truthfully
Check a small sample on a set plan. Match the record to the source. See if a person can grasp the reason with no tool demo.
Track simple faults. Count old links, blank fields, wrong facts, rank shifts, and human changes. These counts do not show gains.
NIST says its AI risk guide is optional. It can help firms add trust checks to AI design, use, and review. [3] Use steps that fit your brief and risk level.
Be exact when you tell others about an AI process. The SEC charged two firms over false or misleading AI claims in 2024. [4] Do not say a tool proves fit, tells future gains, or makes choices it does not make.
- Sample date and sample size.
- Facts that matched the source.
- Facts that needed a correction.
- Rank changes and human overrides.
- Known limits and the next fix.
FAQ
Frequently asked questions
What should an AI match record show?
Show the brief, facts used, source links, dates, gaps, and the rank reason. Add the reviewer and choice. This lets a person check the match before contact.
Is a score an explanation?
No. A score can sort a list. It does not show which facts made the result or if they are new. Ask for facts, weights if used, gaps, and a clear reason. NIST keeps the record apart from how and why a result was made. [1]
Should I introduce a company when key data is missing?
It depends on your brief and check rule. Do not hide the gap. A blank field can mean many things. A reviewer can ask for more facts, hold the match, or make a short intro that names the gap.
Can I market an AI matching process as proven?
Use care. A process record can show how your team worked. It does not prove fit or future gains. The SEC has acted on false or misleading AI claims by firms. Rules vary by facts and place. Seek expert rule advice for your case. [4]
REF
Sources
- 1
National Institute of Standards and Technology
AI Risks and Trustworthiness ↗ - 2
Information Commissioner's Office
Explaining decisions made with AI ↗ - 3
National Institute of Standards and Technology
AI Risk Management Framework ↗ - 4
U.S. Securities and Exchange Commission
SEC Charges Two Investment Advisers with Making False and Misleading Statements About Their Use of Artificial Intelligence ↗