# Checklist for updating AI lead scoring over time
> Learn how to refine your ICP for AI lead generation by analyzing false positives and adjusting your profile descriptions to improve lead quality.
- HTML: https://getkinlead.com/checklist-for-ai-lead-scoring-decay
- Markdown: https://getkinlead.com/checklist-for-ai-lead-scoring-decay.md

**Short answer:** Updating your ICP for AI lead generation requires a feedback loop where you analyze high-scoring leads that failed to convert and use those gaps to refine your profile descriptions or source URLs. You identify the specific traits of false positives and add them as negative constraints in your AI prompt.

## Identify your false positives

A high score means the lead matched the text you gave the AI. It does not mean the person can buy, has budget, or works at the kind of company you described. The first job is to find leads the AI was confident about that turned out to be wrong.

Start with leads that scored high and then stalled. That includes people who never replied, people who replied with a clear no, and people who were disqualified on a call. Open their profiles next to the description you used. Ask what words in the profile made the AI think this was a fit.

Keep a simple log. One row per false positive, with the profile URL, the score, the reason it was wrong, and the exact phrase from the profile that caused the mismatch. After ten or twenty rows, patterns show up. You will see the same title, company type, or headline phrase pulling in the wrong people.

Do not change the description after one bad lead. One bad lead can be a data problem or a person who changed jobs. Change it when the same kind of bad lead appears three or four times. That is a pattern in your targeting, not a one-off.

1. **Filter for high scores.** Sort your lead list to find people who scored 80 or higher but never booked a meeting or replied. If you do not have enough, include leads who replied once and went quiet.
2. **Sample the failures.** Pick 10 to 20 profiles from this group and open them in a new tab. Keep the original scoring description open beside them.
3. **Diagnose the mismatch.** Write down exactly why each person was a bad fit. Company size, funding stage, a title that sounds senior but is not, agency versus product company, or a service that only looks similar.
4. **Spot the pattern.** Look for a common keyword, industry, or headline phrase the AI is reading as a positive signal. That phrase is what you will exclude or qualify in the next step.

## Analyze the gap between score and reality

AI scoring reads text. It sees a headline, an about section, a company name, and a job title. It can match 'Founder' and 'Marketing' without knowing the founder runs a one-person freelance shop. It can match 'Head of Sales' without knowing the company has no budget for your product.

The gap is usually not random. It comes from a word that has two meanings in your market. 'Agency' can mean a 200-person paid media firm or a solo consultant. 'Founder' can mean a technical founder who builds the product or a founder who sells services. The AI needs you to separate those cases.

Read the profile the way the AI reads it. Look at the headline first, then the about section, then the current company. The company page often carries the detail that changes the fit: team size, what they sell, who they sell to. If the company page is vague, the AI has to guess.

Write the mismatch in plain words. 'This person is a freelance designer, not a SaaS sales leader' is more useful than 'bad fit'. The plain sentence becomes the exclusion you add to the description.

**What the AI saw**

- Founder of a marketing agency (matched 'Founder' and 'Marketing' keywords)
- Head of Sales at a FinTech startup (matched 'Head of Sales' and 'FinTech')

**The reality**

- Founder of a one-person freelance agency (too small for the product)
- Head of Sales at a seed-stage company with no budget (no buying power)

## Refine your ICP description

Once you know the pattern, change the description. Vague terms like 'Marketing Leader' invite false positives. Use restrictive language that forces the AI to be more critical. Name the title, the company type, the size, and the market.

Do not write one broad sentence and hope the AI fills in the gaps. Give it requirements and exclusions. The requirements tell it what to look for. The exclusions tell it what to ignore even when the rest of the profile looks close.

Keep the description short enough that you can read it out loud. If it takes a page to explain who you want, the AI will weight the wrong parts. Put the most important filter first: company type, then title, then size, then the work they do.

Test the new description against the false positives you already found. Run a small batch and check whether those profiles now score below your fit line. If they still score high, the exclusion is too soft or the positive signal is too strong.

1. **Add explicit exclusions.** Add phrases like 'Exclude freelancers, solo practitioners, and agencies with fewer than 10 employees' to your profile.
2. **Swap vague terms for specifics.** Change 'Marketing Leader' to 'VP of Marketing or CMO at B2B SaaS companies'.
3. **Define hard boundaries.** Specify the exact company size or funding stage required to be a fit. If you only want companies with a sales team, say that.
4. **Test a small batch.** Update the description and run a test on 50 leads. Check whether the scores for previous false positives drop below your fit line.

```
Target: VP of Sales at B2B SaaS companies with 50 to 200 employees. Focus on companies that sell to mid-market enterprises. Exclude: Seed-stage startups, freelance consultants, and agencies under 20 people.
```

## Audit your source website

If you use your website as the source for scoring, the AI treats your copy as the definition of your customer. That works when the page is specific. It fails when the page tries to cover every use case.

A homepage that says 'We help all businesses grow' gives the AI no reason to reject anyone. A page that says 'We help B2B SaaS companies with 50 to 200 employees fix sales follow-up' gives the AI something to match. The same product can produce very different lead quality depending on the page you point at.

You do not need to rewrite your public homepage. Build a page for the scoring source that describes your best customer in plain detail. Include the company size, the buyer title, the problem, and the cases you do not serve. Point the scoring source at that page.

Do not point the source at a blog post about a trend. A trend post describes a topic, not a buyer. Use a page that names the buyer and the company. If you only have a homepage, add a short section to it that states who the product is for and who it is not for.

1. **Review landing page copy.** Check if your homepage mentions broad audiences you do not actually want to target. Look for phrases like 'any business' or 'all industries'.
2. **Create a target-specific page.** Build a page that describes your ideal customer in detail. Include who it is for and who it is not for.
3. **Swap the source URL.** Use the URL of the specific ICP page instead of your homepage to guide the AI's scoring logic.

> If your website says 'We help all businesses grow', the AI will score almost any business leader highly because it thinks your market is universal.

## Set a review cadence

ICP refinement is not a one-time task. Your market changes, your product changes, and the profiles you attract change as your outreach reaches new groups. A description that worked six months ago can start pulling in the wrong titles.

Put the review on a calendar and give it an owner. The person who runs lead generation should keep the false-positive log and bring the patterns to the review. If you work alone, block 30 minutes a week and one longer session a month.

Use the weekly review to catch new patterns early. Use the monthly review to change the description. Use the quarterly review to check the source page and the sales playbook, because a change in what you sell can change who fits.

Write down what changed and why. When a lead quality problem comes back, you can check whether the description drifted or whether a new title entered your market. A short changelog saves you from guessing.

| | Frequency | Action | Goal |
| --- | --- | --- | --- |
| Weekly | Review the last 20 high-score leads that did not reply | Identify new false-positive patterns |
| Monthly | Update the ICP description | Remove common bad-fit traits |
| Quarterly | Audit source website and playbook | Align AI with current sales goals |

## Doing this with Kinlead

Kinlead scores each lead from 0 to 100 against the customer profile you describe or your website. Leads below the fit line are not kept. You can refine the target by updating the description or the website URL used for scoring, which changes how new leads are evaluated.

Leads are kept in [lead groups](/features/lead-groups-and-scoring), and any campaign can point at any group. You can export leads to CSV to run the false-positive analysis in this guide, or import a CSV of LinkedIn profile URLs when you want to recheck a known list.

## Related

- [Lead finders](https://getkinlead.com/features/lead-finders.md)
- [AI LinkedIn outreach](https://getkinlead.com/features/ai-linkedin-outreach.md)
- [Unified inbox](https://getkinlead.com/features/unified-inbox.md)

## Common questions

**Should I lower my score threshold or change my ICP?**

Change your ICP. Lowering the threshold lets more low-quality leads into your campaign. Tightening the description fixes the root cause.

**How do I handle leads that are a perfect fit but do not respond?**

That is a messaging problem, not a scoring problem. If the fit is correct, look at your sales playbook or the hook in your first message.

**Can I use multiple ICPs in one campaign?**

It is better to create separate lead groups for each ICP. That lets you use a specific sales playbook for each segment and keep the scoring descriptions clear.

**How often should I update the ICP description?**

Review it monthly and after any change to your product, pricing, or target market. If you see the same false positive three times in a week, update the description sooner.

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