A 100-point model
| Part | Points | Example rules |
|---|---|---|
| Company size | 25 | In your sweet spot: 25; adjacent: 10 |
| Industry | 20 | Core industry: 20; secondary: 10 |
| Region | 10 | Sales territory: 10 |
| Pages read | 20 | Pricing or comparison: 20; product: 10; blog only: 3 |
| Source | 15 | ChatGPT ad on a decision question: 15; AI answer: 12; other: 5 |
| Repeat visits | 10 | Second visit in 14 days: 10 |
Thresholds
- 75 and above: alert the account owner today.
- 60–74: weekly list for sales.
- Below 60: nurture and audiences only.
Keep it honest
- Review the model with sales every quarter.
- Check which scores turned into meetings, and adjust weights.
- Exclude customers, partners, competitors and internet providers.
Testing the model
After three months, look at companies that got meetings and those that didn't:
| Score band | Companies | Meetings | Meeting rate |
|---|---|---|---|
| 75+ | |||
| 60–74 | |||
| Below 60 |
If the 60–74 band books meetings as often as 75+, lower the alert threshold.
Questions
Should source get points?
Yes, but modestly. A perfect-fit company is worth a call wherever it came from.
How do we score companies with missing data?
Give neutral points and let behaviour decide. Don't punish companies for gaps in your data provider.
Should we use AI to score?
Start with a simple rule-based model. You need to be able to explain every score to sales.
Why two dimensions beat one score
A single score blends who the company is with what it did. That hides useful differences: a perfect-fit company reading a blog post and a poor-fit company reading pricing can end up with the same number. Keeping fit and behaviour visible separately, even if you also show a total, lets sales choose the right action: research and reach out to the first, ignore the second.
Where the data comes from
- Fit: company size, industry and region from your identification or data provider.
- Behaviour: pages, visits and source from analytics and identification.
- Context: the buyer question from your UTMs.
A warning about precision
A score of 78 isn't meaningfully different from 76. Use bands, not exact numbers, when routing.
Start here: Company identification from AI. More articles in Company signals.
