How to Score B2B Accounts: ICP Scoring with Real-Time Buying Signals

Most ICP definitions describe who could buy from you. A good scoring model identifies who is ready to buy right now. The difference is timing -- and timing determines whether your outreach lands in an active evaluation or gets archived.

Why Firmographic Filters Alone Are Not Enough

Firmographics -- company size, industry, revenue, geography -- are the starting point. They define the universe of possible buyers. But within that universe, readiness varies enormously.

Two companies with identical firmographic profiles can have completely different levels of urgency. One is actively evaluating solutions. The other won't have budget for two years. Firmographics can not tell you which is which.

• Headcount range: necessary filter, not a buying signal
• Industry: defines the universe, not the timing
• Revenue: useful for deal size estimation, not urgency
• Geography: an exclusion filter, not a scoring factor

Treat firmographics as the gate. Signals are the score.

What Are Real-Time Buying Signals?

Buying signals are changes in a company's behavior that indicate an active need, new budget, or a decision being made. The best signals are time-sensitive -- they have a window of relevance before the moment passes.

How Do You Build the Scoring Model?

A scoring model assigns numerical weights to each signal. The total score determines priority tier. Build it in Clay so scores update automatically as signals change.

Fit score (max 50 points)

• Headcount in ideal range: 20 points
• Industry exact match: 15 points
• Revenue in target range: 10 points
• Geography match: 5 points

Signal score (max 50 points)

• Active relevant job posting: 20 points
• Funding event in last 90 days: 15 points
• Relevant tech install detected: 15 points
• Executive hire in last 60 days: 10 points
• Intent data signal active: 10 points
• Headcount growth over 20% YoY: 5 points

How Do You Find These Signals at Scale?

Manual signal research does not scale. For a list of 1,000 accounts, you need automation.

• Job postings: Proxycurl, People Data Labs, or direct LinkedIn scraping via Clay
• Funding: Crunchbase and Harmonic integrated directly into Clay tables
• Tech stack: BuiltWith or Clearbit Reveal via Clay enrichment columns
• Executive hires: LinkedIn job change alerts or Clay's people enrichment layer
• Intent data: Bombora or G2 Buyer Intent piped into Clay via webhook
• Headcount growth: LinkedIn follower count trend or People Data Labs headcount history

Clay lets you pull all of these into a single table, compute the score in a formula column, and automatically tier every account.

How Do You Tier Your Total Addressable Market?

1. 1Start with firmographic filters to define your TAM -- every company that could theoretically buy.
2. 2Run signal enrichment across the full TAM. This is where Clay earns its cost.
3. 3Score every account using your weighted formula.
4. 4Tier: P0 gets immediate personalized outreach. P1 enters standard sequences. P2 goes on a monitoring cadence.
5. 5Export P0 and P1 to your sequencer. P2 stays in Clay for automatic re-scoring as signals change.

How Do You Know If Your Signals Are Predictive?

After 60-90 days of outreach, look back at your closed deals and positive replies. Which signals were present in those accounts?

• Pull all accounts that converted or replied positively
• Check which signals were present at time of outreach
• Compare signal presence rate vs. your full contacted list
• Signals overrepresented in wins get higher weights
• Signals not present in wins get reduced weights or removed
• Repeat this analysis every quarter

Your best customers will tell you exactly what to look for in the next ones. The model improves automatically if you feed it the right data.

What Are the Most Common Mistakes?

• Using too many signals -- a model with 15 inputs is harder to trust than one with 6
• Weighting all signals equally -- not all signals predict the same urgency
• Never revisiting weights -- signals that predicted buying last year may not predict it today
• Scoring accounts but not updating scores -- a signal from six months ago is not the same as one from today
• Skipping the fit layer -- a high-signal company outside your ICP is still a bad fit