Score Inbound Leads Without Letting AI Make the Decision

Not every inbound lead deserves the same response time, but a black-box score should not decide who gets attention. A transparent evidence summary can help a person prioritize without hiding the judgment involved.

This guide helps you create a simple evidence-based lead score with visible reasons, unknowns, and human review rules.

What you need

• Your minimum qualification rules
• Positive, negative, and disqualifying signals
• A small set of historical leads with known outcomes

The workflow

1. Choose observable criteria

Use facts such as service area, problem fit, timing, required capability, and stated budget range. Do not score protected traits, personality guesses, writing style, or proxy variables.

2. Separate fit from priority

A lead can be a good fit but not urgent. Create separate fit and timing scores so your team can nurture instead of incorrectly rejecting.

3. Make unknown an allowed answer

Missing data should trigger a question, not a negative assumption. This one rule prevents many unfair and inaccurate scores.

4. Test against old outcomes

Run the rubric on a small historical sample. Look for qualified leads it would miss and poor-fit leads it would promote.

Copy this working prompt

Replace every bracketed field with verified information from your business. If you do not know something, write unknown instead of guessing.

Help me design and apply a transparent inbound lead-priority rubric.

Offer and minimum fit:
[DESCRIBE THE OFFER AND NON-NEGOTIABLE REQUIREMENTS]

Observable positive signals:
[LIST SIGNALS]

Observable negative signals:
[LIST SIGNALS]

True disqualifiers:
[LIST ONLY FACTUAL REASONS THE BUSINESS CANNOT SERVE THE LEAD]

Lead data:
[PASTE REDACTED INTAKE INFORMATION]

Rules:
- Use only provided facts
- Unknown is not negative
- Do not infer protected traits, personality, wealth, intelligence, intent, or seriousness from names, language, location, or writing style
- AI may organize evidence but may not accept or reject the lead

Return:
1. Fit score from 0 to 5 with evidence
2. Timing score from 0 to 5 with evidence
3. Known positive signals
4. Known risks
5. Unknowns that need a question
6. Any factual disqualifier
7. Recommended human next step: respond now, ask questions, nurture, or review manually
8. Confidence level

Finish with a plain-language explanation a team member can audit.

Run a second-pass review

The first output should not be the final answer. Use this challenge prompt to find weak evidence, hidden assumptions, or avoidable risk.

Audit this rubric for bias, proxy variables, hidden assumptions, and criteria that reward polished writing instead of actual fit. Recommend safer observable replacements.

What good looks like

• A teammate can reproduce the score
• Unknowns become questions
• AI never makes the final acceptance or rejection decision

Review before you use it

• Could any criterion unfairly disadvantage a protected group?
• Does the score use information relevant to delivering the service?
• Are low-confidence cases routed to a person?

Privacy and judgment guardrails

• Remove passwords, payment details, private health information, and confidential customer data before pasting anything into an AI tool.
• Treat the output as a working draft. A person remains responsible for the decision, promise, price, and final send.
• Do not let AI invent customer quotes, financial figures, legal terms, capabilities, deadlines, or proof.

Your next action

Test the rubric on ten old leads and compare it with what actually happened before using it on new inquiries.

If your AI tool still lacks the context to do this well, first Build a clear business context brief.