How investors can compare startup pipelines without hiding bias

Start with the full list

Do not compare only deals that reached a partner. Start with each lead that could enter. Give each lead a date and route. Add mandate fields. Mark a missing fact as not known.

Use one time window and one meaning for a lead. If not, one list can look better. It may just be cleaned first. A route is how the lead arrived. It is not proof of fit.

An NBER study found many VC deals came from work links. Other deals came from fund introductions and active search. One in ten came from company managers. [1] Label the route. A known route should not get a hidden edge.

• Entry date and source route.
• Stage, sector, place, and cheque range.
• Known, unknown, or not a fit for each field.
• A link to the evidence and its check date.

Set a rule before the first review

Write a short rule card before people compare leads. Name stage, sector, place, cheque size, and conflict rules. State what proof can support each field. Keep a not-known fact as not known.

When you can, hide the referrer name in the first fit check. Do not hide facts needed for a real conflict. Do not hide facts needed for a legal duty. The aim is not blindness at all costs. The aim is a clear order for using facts.

NIST lists more than one type of bias. It says bias in people can change how they use facts. It can change how they fill gaps. [3] A written rule will not end bias. It makes changes easier to see.

• Rule: the exact condition being tested.
• Proof: the record that supports the condition.
• Status: known, unknown, or mismatch.
• Reviewer: who moved the lead and when.

Compare movement, not only wins

Compare each source at the same steps. Count all leads. Count mandate fits, first checks, deep checks, and passes. Show the count beside the rate. A small group can change fast.

Write why a lead stopped. Use clear reasons. They may be no mandate fit, no proof, conflict, timing, no reply, or team load. Do not let one label hide many reasons.

ILPA asks firms to say how they find, check, and record deals. It also asks for rules used to judge a deal. [2] This is a useful prompt for a startup lead list. It is not a rule for every investor.

Do not judge a source only by a few later outcomes. Those results show leads your team chose. They do not show each lead your team could see.

• Source route and total lead count.
• Count and rate at each review step.
• Reason for a pass, hold, or advance.
• Sample size and review period.

Check people and tools

If a tool sorts or scores leads, save its input and version. Save its output and human changes. Ask for linked facts and gaps, not a verdict. A person should make the next choice.

NIST says teams should test bias. It says to save results and track risks over time. [3] Use this to check tool results and human changes.

NIST warns that AI can give false facts or false sources. It says teams should check sources in tests and later. [4] Open the record before you use it.

Name one owner for the review. Change one rule at a time. Test it on saved leads before use across the list. Check some passed and held leads. Do not check only leads that moved on.

• A named owner and review date.
• A version note for each rule or tool change.
• A sample of passes and holds for recheck.
• A way to pause the process when evidence fails.