Intent Score Model

This drill builds the scoring model that converts raw intent signals into a prioritized outreach queue. It defines what signals matter, how much each one weighs, and what score thresholds trigger action.

Input

• ICP definition (from icp-definition drill)
• At least one active signal source: website visitor identification, G2, Bombora, or manual signal logging
• Historical deal data in Attio (10+ closed-won deals recommended for calibration)

Steps

1. Map your signal universe

Catalog every signal you can capture. Organize into three tiers by predictive strength:

Tier 1 — Direct intent (highest weight):

• Visited pricing page
• Requested a demo or started a free trial
• Viewed competitor comparison pages
• G2 "alternatives" or "compare" signal
• Multiple return visits within 7 days

Tier 2 — Research intent (medium weight):

• Visited case studies or docs
• Downloaded a resource
• Bombora surge score above 70 on relevant topics
• G2 category browsing
• Read 3+ blog posts in one session

Tier 3 — Contextual signals (lower weight, multiplier effect):

• New executive hire in buyer persona role
• Funding round in last 90 days
• Hiring 3+ roles in your product domain
• Adopted a complementary technology
• Competitor technology detected on their site

2. Assign signal weights in Clay

Using the clay-intent-scoring fundamental, set up a Clay table with columns for each signal and build the weighted scoring formula. Start with these default weights:

| Signal | Points | Max | |--------|--------|-----| | Pricing page visit | 15 | 15 | | Multiple site visits (per visit) | 5 | 25 | | G2 alternatives/compare signal | 20 | 20 | | G2 category browsing | 10 | 10 | | Bombora surge (scaled) | 0.2x score | 15 | | Recent funding | 10 | 10 | | New exec hire | 5 | 5 | | Job postings (per posting) | 2 | 10 | | Competitor tech detected | 5 | 5 |

Apply time decay: signals older than 7 days lose 15%, older than 14 days lose 40%, older than 30 days lose 70%.

3. Define tier thresholds

Using the formula output, define action tiers:

• Hot (70+): Contact within 24 hours. Fully personalized outreach referencing specific signals.
• Warm (40-69): Contact within 72 hours. Semi-personalized outreach using category-level messaging.
• Cool (15-39): Add to nurture sequence. Monitor for score increases.
• Cold (<15): No action. Re-evaluate when new signals arrive.

4. Calibrate against historical data

Pull closed-won deals from Attio using attio-lists. Enrich each deal's account through Clay with the signal data available at the time of engagement. Score them retroactively:

• If fewer than 70% of won deals score Hot or Warm, your weights undervalue the signals that drove those wins. Increase weights on signals those accounts had.
• If more than 50% of lost deals also score Hot, your model is too generous. Raise thresholds or reduce weights on common-but-not-predictive signals.

5. Push scores to Attio

Create a custom attribute in Attio using attio-custom-attributes:

• intent_score (number) on the Company record
• intent_tier (select: Hot/Warm/Cool/Cold) on the Company record
• intent_signals (text) logging which signals fired

Push from Clay to Attio after every score calculation. Create Attio lists for each tier using attio-lists so outreach agents can pull the right accounts.

6. Track scoring accuracy

Log every score-to-outcome pair in PostHog using posthog-custom-events:

Event: intent_score_assigned
Properties: { company_domain, intent_score, intent_tier, signals_fired }

Event: intent_outcome_recorded
Properties: { company_domain, outcome: replied|meeting|deal|closed_won|closed_lost }

After 4 weeks, compare: do Hot accounts convert at 3x+ the rate of Cold accounts? If not, the model needs recalibration.

Output

• Clay table with scored accounts and tier labels
• Attio lists segmented by intent tier
• PostHog tracking for score-to-outcome analysis
• Documented scoring model with weights, thresholds, and decay rules

Triggers

• Re-run calibration monthly or after every 20 new closed deals
• Re-run whenever a new signal source is added