# How to validate AI lead scores manually
> Audit AI lead lists with a manual spot check to confirm the buyer persona matches the people scored and targeted in your LinkedIn campaigns today.
- HTML: https://getkinlead.com/checklist-for-ai-lead-scoring-validation
- Markdown: https://getkinlead.com/checklist-for-ai-lead-scoring-validation.md

**Short answer:** To validate AI lead scores, pull a random sample of leads and check each profile against a short list of binary fit criteria. Mark every lead fit or no fit before you look at the score. If high scores keep failing the manual check, tighten the persona description or raise the fit line so the same mistake stops reaching your campaigns.

## Define your non-negotiables

AI scoring is useful when the instructions are specific. It gets shaky when the profile description is a pile of words that sound right but do not describe a buyer. A lead can have Sales in their title and still be a junior coordinator who cannot sign a contract. Before you trust a score, decide what makes a lead viable at all.

Write a short checklist where every line has a yes or no answer. Do not use a scale for this part. A lead either clears the bar or does not. If a lead fails one item, mark them no fit even if the AI gave them a high score. That rule keeps the audit honest.

Keep the list small enough to use in under a minute per profile. Current title, company size, industry and geography cover most B2B cases. Add one or two items that matter for your product, such as whether the company sells to the same buyer you serve or whether they have a team that would use your tool.

| | Dimension | Fit (yes) | No fit (no) |
| --- | --- | --- | --- |
| Current title | The person can approve or influence the purchase | Junior, intern, student or unrelated function |
| Company size | Inside the range you can serve well | Too small to pay or too large to move |
| Industry | A sector where your product solves a known problem | An adjacent sector with no use case |
| Geography | A region where you can sell and support | A country outside your service area |
| Business model | The company sells to the same buyer you serve | The company is an agency, consultant or reseller |

## Run a small blind spot check

You do not need to read every profile. A small random sample shows whether the scoring is drifting. Pick 20 leads from different parts of the list, not the first five that look good. The point is to catch a pattern, not to prove the list is perfect.

Hide the score while you review. If you see a 92 before you open the profile, you will look for reasons to agree with it. Open the profile, read the current role, check the company, then decide fit or no fit. Write the result down before you look at the score again.

Do the same check after every change to the persona description. If you edit the description and skip the review, you are guessing. A short audit takes less time than a week of messages to people who were never going to buy.

1. **Pick the sample.** Open your lead group and choose leads from the top, middle and bottom of the scored list. Do not filter by score first.
2. **Hide the score.** Use a spreadsheet or a blank note. Copy the profile URL and leave the score column empty until you finish the review.
3. **Read the current profile.** Check the headline, current role, company page and recent activity. A title from three years ago is not a current title.
4. **Apply the checklist.** Mark fit or no fit. If the lead fails one non-negotiable, the result is no fit.
5. **Compare the results.** Count how many high scores failed. If several fail for the same reason, fix the persona description before you run another campaign.

## Look for the reason a high score failed

A false positive is a lead the AI scored highly that fails your checklist. One false positive is noise. A group of them with the same cause is a targeting bug. Read the profiles that failed and write down the reason in plain words.

Most failures come from a few places. The AI may match a keyword without reading seniority. It may pick up an old role from the experience section. It may treat a service provider as a buyer because both use the same industry words. It may also match a large company when your buyer only exists in a small team inside it.

Do not fix this by adding more adjectives to the persona. Fix it by adding constraints the AI can check. Marketing leaders is weak. Heads of Marketing at B2B software companies with 50 to 200 employees, excluding agencies and freelancers, gives the scorer something to reject.

**Weak signal**

- Title contains Marketing
- Company is in software
- Profile mentions leads or demand
- Location is in the right country

**Buyer signal**

- Current title is Head of Marketing or VP of Marketing
- Company sells software to other businesses
- The person owns demand generation or pipeline
- Company size and location match your service area

## Tighten the persona description

The persona description is the instruction the scorer follows. If it is vague, the scorer fills the gaps with guesses. Rewrite it as a set of statements about the buyer, not a list of buzzwords. Say what the person does, where they work and what makes them a poor match.

Add exclusions. If you sell to companies that run their own sales team, say that consultants and lead generation agencies are not a fit. If you sell to a specific department, name the department and the seniority. The scorer can only reject what you tell it to reject.

Test one change at a time. If you change the title, company size and industry all at once, you will not know which change fixed the problem. Run the same spot check after each edit and compare the failures.

- **draft:** Target: Sales leaders at B2B companies.
- **you:** That is too broad. Use: VPs of Sales at B2B software companies with 50 to 200 employees. Exclude consultants, agencies and freelancers.
- **draft:** Target: HR managers.
- **you:** That is too vague. Use: HR managers who own retention and culture programs at healthcare companies with 200 to 800 employees. Exclude staffing firms.
- **draft:** Target: Founders.
- **you:** Add the stage. Use: Founders of bootstrapped B2B service businesses with 10 to 50 employees. Exclude venture-backed startups and solo consultants.

> If a lead would need an explanation for why they count as a buyer, they are not a fit. The checklist should be boring enough that two people on your team reach the same answer.

## Check borderline leads with a second pass

Some leads will sit on the line. The title is right but the company is a little small. The industry is close but not exact. Do not let borderline leads slide into the campaign just because the score is high. Put them in a separate pile and review them after the first pass.

A second pass is also where you catch mistakes in your own checklist. If you keep marking the same kind of company as a fit even though they never buy, the checklist is wrong. Update the checklist before you update the AI instructions.

Keep a simple audit log. You do not need a CRM for this. A CSV with the profile URL, the score, your manual result and the reason is enough. After two or three audits, the reasons will repeat and the fix will be obvious.

```
profile_url,score,manual_fit,reason,persona_rule
https://www.linkedin.com/in/example-one,91,no,agency owner,exclude agencies
https://www.linkedin.com/in/example-two,84,yes,,
```

## Doing this with Kinlead

Kinlead finds people on LinkedIn who match the customer you describe, or your website, and scores each lead from 0 to 100 against that profile. Kinlead does not keep leads below the fit line. It keeps leads in lead groups, and any campaign can point at any group.

If your audit shows high scores going to people who fail your checklist, rewrite the customer description or raise the fit line in your [lead group settings](/features/lead-groups-and-scoring). That keeps the same mistake from reaching your [campaigns](/features/campaigns-and-send-windows). You can also export a lead group to CSV for the audit log.

- Scores each lead from 0 to 100 against the profile you describe.
- Drops leads below the fit line.
- Keeps leads in groups that campaigns can use.

## Related

- [Lead groups and scoring](https://getkinlead.com/features/lead-groups-and-scoring.md)
- [AI LinkedIn outreach](https://getkinlead.com/features/ai-linkedin-outreach.md)
- [Find leads](https://getkinlead.com/tools/find-leads.md)

## Common questions

**How often should I manually validate AI lead scores?**

Run a spot check when you start a new campaign, change the persona description or notice a drop in reply quality. If you are scaling outreach, do a small audit every week.

**What if the AI matches the title but the company is wrong?**

Your company-level constraints are too loose. Add industry, company size or business model rules to the persona description, then rerun the same spot check.

**Can I track a manual audit in a CSV?**

Yes. Export the lead group, add columns for manual fit and the reason, and review the reasons after each audit. The pattern matters more than the individual score.

**Should I delete a lead that fails the manual check?**

Remove them from the campaign and from the lead group if the same profile keeps appearing. If the failure came from a vague rule, fix the rule first so the next search does not bring them back.

[Create a Kinlead account](https://getkinlead.com/signup) or [book a call](https://getkinlead.com/demo) to talk through pricing and setup.
