Ask the average RevOps leader if their company uses account scoring, and the answer is almost always yes. Ask them if the scores actually correlate with closed-won revenue, and the answer gets much murkier. Account scoring is ubiquitous in B2B sales — and largely broken in practice.
Why Most Scoring Models Fail
The most common failure mode is building a scoring model based on what the team thinks good accounts look like, rather than what the data actually shows about past wins. This produces models that reward the attributes of your ideal customer profile document — industry, company size, tech stack — without incorporating the behavioral signals that actually predict buying readiness.
The second failure mode is static scoring. A score that was accurate three months ago is not accurate today if the account's intent behavior has changed. Models that update weekly or monthly are already fighting yesterday's battle. The accounts your AI should be surfacing today are the ones showing signals this week — not the ones that looked good last quarter.
The third failure mode is the one no one talks about: scores without explanations. If a rep cannot understand why an account has a high score, they will not trust it — and if they don't trust it, they will not change their behavior. A black-box score that produces the right answer but generates zero behavioral change is worth nothing in practice.
The Four Dimensions of a Predictive Score
Fit Score: How well does this account match your historical ICP? This should be based on the characteristics of your actual closed-won accounts, not your ICP document. These two things are often significantly different, especially for companies that have been selling for more than 18 months.
Intent Score: What behavioral signals is this account generating right now? This is the most dynamic dimension and the most valuable for timing decisions. It should incorporate multiple independent signal sources and weight recent activity more heavily than older signals.
Timing Score: Is this account in an active buying window? This combines intent signals with contextual triggers — recent funding, executive changes, tech stack changes, competitive displacement signals — to assess whether the company is in a period where decisions are likely to be made.
Engagement Score: What is the depth of your existing relationship with this account? First-party data from your own marketing channels, email engagement, and event participation provides a complementary dimension to third-party intent signals.
Validating That Your Model Actually Works
The only valid test of a scoring model is whether high-scoring accounts convert to pipeline and revenue at higher rates than low-scoring ones. Run this analysis at least quarterly:
- Sort all accounts worked in the last 90 days into score quartiles (Q1 = highest score, Q4 = lowest)
- For each quartile, calculate: meeting rate, qualified opportunity rate, closed-won rate, and average ACV
- If your model is working, you should see a consistent monotonic decline across all four metrics from Q1 to Q4
- If Q3 outperforms Q2, or Q4 is within 20% of Q1, your model is not generating meaningful signal and needs retraining
This validation exercise should be on every RevOps team's monthly calendar. Models drift as your ICP evolves, as your product positioning changes, and as the market dynamics shift. A model that was highly predictive 12 months ago may be significantly degraded today — and no one on your team will know unless you are actively measuring it.