ICP Account Modeling

ICP account modeling scores every company in your pipeline, CRM, or target list on how closely it matches your ideal customer profile. The output is a numerical fit score per account that drives prioritization, routing, ABM tiering, and resource allocation. Instead of "this looks like a good account" (subjective), the model produces "this account scores 82/100 on ICP fit" (objective, comparable, actionable).

The principle: the model should predict which accounts are most likely to buy, retain, and expand. Build it from closed-won data, validate against churn data, and update quarterly. A model built from intuition rather than data will drift from reality within 6 months.

The Account Scoring Model

Scoring dimensions

Scoring rubric

For each dimension, score 0-3:

Calculating the weighted score

ICP Score = (size_score × 0.20) + (industry_score × 0.20) +
            (funding_score × 0.15) + (geo_score × 0.10) +
            (tech_score × 0.15) + (growth_score × 0.10) +
            (gtm_score × 0.10)

Normalize to 0-100 scale:
  Raw max = 3.0 (every dimension = 3)
  ICP Score (0-100) = (raw_score / 3.0) × 100

Tier assignment

Building the Model from Data

Step 1: Define weights from closed-won analysis

Pull 50+ closed-won deals. For each dimension, calculate the win rate by score level.

Example: Company size analysis

The dimension with the largest win-rate spread between best and worst segments gets the highest weight. If company size has a 63pp spread (73% - 10%) and geography has a 20pp spread, size is weighted higher.

Step 2: Validate against churn data

Apply the scoring model to existing customers. High-scoring accounts should churn less.

If Tier 1 accounts don't churn significantly less than Tier 3, the model isn't predicting retention. Adjust weights.

Step 3: Automate scoring in CRM

On account creation or enrichment update:
  → Query enrichment fields (size, industry, funding, geo, tech, growth)
  → Apply scoring rubric per dimension
  → Calculate weighted score (0-100)
  → Assign ICP tier
  → Store: icp_score, icp_tier, icp_last_scored

Step 4: Quarterly calibration

• Re-run closed-won analysis with new data
• Check if Tier 1 still wins at 2x+ Tier 3 rate
• Check if Tier 1 still retains at 2x+ Tier 3 rate
• Adjust weights and score thresholds if needed

Negative Scoring (Anti-ICP)

Some attributes should disqualify an account regardless of positive fit.

Where the ICP Score Gets Used

Look-Alike Modeling

Use the ICP model to find new accounts that resemble your best customers.

Process

1. Score your top 20 customers on all ICP dimensions
2. Calculate the average score profile:
   Size: 2.7, Industry: 3.0, Funding: 2.3, Geo: 3.0, Tech: 2.1
3. Search for accounts with similar profiles:
   Sales Nav, Apollo, or Crunchbase filtered to match each dimension
4. Score the results with the ICP model
5. Tier 1-2 matches become your look-alike target list

Look-alike rules

• Look-alikes from closed-won data, not from wishful thinking. Model from your actual best customers, not from "the kind of company we wish would buy"
• Update the look-alike profile quarterly. As your customer base evolves, the "average best customer" profile changes. A profile from 12 months ago may not represent today's ICP
• Look-alikes are a starting point, not a guarantee. A company that looks like your best customer still needs a signal (timing) to become a real prospect. Look-alike + signal = qualified target. Look-alike alone = nurture

Measurement

Anti-Pattern Check

• Model built from intuition, not data. "I think mid-market SaaS is our ICP" is a hypothesis. "73% of closed-won deals are 51-200 employee SaaS companies" is a model. Use data
• No negative scoring. A government agency with 3 matching dimensions scores 45/100 and enters the pipeline. Without negative signals, bad-fit accounts pollute the pipeline. Add anti-ICP penalties
• Scoring but never using the score. The icp_score field exists in CRM but nobody looks at it. The score should drive routing, ABM tiering, lead scoring, and pipeline review. If it's not used, it's wasted
• Same model for 2 years. Markets shift. Products evolve. The ICP from 2 years ago may not predict today's best customers. Calibrate quarterly
• Over-weighting one dimension. "Industry is 60% of the score" because all your wins are in SaaS. But if 90% of your prospects are also SaaS, industry doesn't discriminate between fit and non-fit. Weight by predictive power (win-rate spread), not by frequency
• Scoring without enrichment data. 40% of accounts have blank industry and employee count fields. Their scores are wrong. Enrich first, score second