Build an ICP Account Scoring Model: A Worked Walkthrough
An ICP scoring model gives every target account a score based on how closely it matches the customers you win and keep, so reps spend time on the best-fit accounts first. Build it from your closed-won and closed-lost deals, weight a handful of criteria, add hard disqualifiers, and review it twice a year.
You don't need a data science team. A worksheet with six to eight criteria and clear point values is enough to change where a small team aims its calls and emails. The walkthrough below builds one from scratch, and the example is illustrative, so replace every criterion and weight with what your own deal history shows.
Vendors Covered in this Article
Disclosure: We may earn a commission if you buy through some links on this page. It doesn't change what we recommend.
Where do the scoring criteria come from?
Start with the deals you've already closed. Pull your last 20 to 30 wins and a similar number of losses and stalls, then compare them on fields you can actually source:
- Firmographics: industry, employee band, revenue band, geography.
- Technology in use: the CRM, billing system or platform the account runs on, if it matters to your product.
- Buying triggers: a new executive, a funding event, a hiring push in the function you serve.
- Engagement: visits to pricing pages, replies, event attendance.
Keep only the traits that separate your wins from your losses. If both groups share a trait, it isn't a criterion. If you can't fill a field for most accounts, drop it or accept that missing data will score low.
How to turn criteria into a 100-point worksheet
Give each group a share of 100 points based on how well it predicts a win in your history.
Say your wins cluster tightly by size and industry, and less so by technology. In this example you might give firmographics 40 points, buying triggers 25, technology 20 and engagement 15. Then break each group into levels, such as 40 points for the size band you win most in, 20 for an adjacent band and zero outside it.
Follow these steps:
- List criteria per group and mark which ones you can fill from your CRM or data tool today.
- Assign points that add to 100 across the groups.
- Write the exact rule for each level so two reps would score the same account the same way.
- Score 15 to 20 past deals with the worksheet and check that wins land near the top.
- Adjust weights until the wins and losses separate cleanly, then stop tuning.
Which disqualifiers should override the score?
Some traits should end the conversation no matter how well an account scores elsewhere. Add a short list of disqualifiers that set the tier to "skip" or subtract enough points to drop the account out:
- A market or country you can't legally or practically serve.
- A business model your product doesn't fit, such as consumer-only companies if you sell to B2B teams.
- Companies below a size where the price can't work.
- Existing customers or accounts in an open deal, so nobody double-works them.
Keep the list short. Every disqualifier is a rule someone has to maintain, and too many will hide accounts that could have been good.
How to convert scores into outreach tiers
A score is only useful if it changes what people do. Map bands to actions:
- Tier 1: the top band gets personal research, multi-channel outreach and an account plan.
- Tier 2: the middle band gets a lighter sequence with some personalization.
- Tier 3: the bottom band gets marketing nurture only, or nothing.
Pick the cut lines from your own score distribution instead of a fixed number. If most accounts land in one band, your model isn't separating anything and you need sharper criteria. For account-based motions, tiering and scoring for ABM goes deeper, and lead scoring with AI is the next step once you have enough history.
How do you keep the model honest over time?
Markets shift and the model will drift. Every six months, compare the accounts you scored high with what happened: how many became opportunities, and how many closed. The average B2B new-logo win rate is 19 percent1, so use that as a rough yardstick, but judge the model mainly against your own baseline. If your highest tier doesn't convert better than your lowest tier, change the weights.
Watch for three failure modes. First, rewarding traits that only correlate with deal size, not fit. Second, scoring on data that's stale, since old technology and headcount fields mislead. Third, letting reps override scores without a note, which teaches the model nothing. Data tools like Apollo or ZoomInfo can fill and refresh the fields, but the rubric itself is your call.
What Good Looks Like
Every target account has a score built from criteria that separate your past wins from your losses, and each score band maps to a defined outreach action.
Building The Capability (5-Stage Skill Ladder)
How to Get Started
Disclosure: We may earn a commission if you buy through some links on this page. It doesn't change what we recommend.
Frequently Asked Questions
What is the difference between ICP scoring and lead scoring?
ICP scoring rates how well an account fits your ideal customer, using firmographic and technology data. Lead scoring rates an individual person's behavior and engagement, such as page visits and replies. Use ICP scores to pick who to target and lead scores to decide who to follow up with now.
How many criteria should an ICP scoring model have?
Six to eight is a workable range for a small team. Fewer than that misses real differences between accounts, and more makes the model hard to explain and maintain. Keep only traits that separate your wins from your losses.
How do you score accounts with missing data?
Decide the rule up front: score missing fields as zero for that criterion, or estimate from a proxy like company size. Then flag accounts with many gaps for a quick manual check before you tier them. Don't guess silently, since that hides data problems.
Should reps ever contact low-scoring accounts?
Rarely, and only with a reason. If a low-scoring account is inbound or referred, treat it as a special case and note why. Otherwise keep reps on the top tiers, and use the notes on exceptions to test whether your model is missing something.
Sources
Where we quote a benchmark, we show its source. Other figures in this guide are estimates or general guidance, so check them against your own numbers.
- Average B2B new-logo win rate. Ebsta x Pavilion 2025 GTM Benchmarks Report, 2025.
Related Guides
A Simple ABM Tiering Model Reps Will Actually Trust
Why most ABM scoring models collapse under their own complexity, and how to build a three-input tier system that stays legible and gets used.
Where AI Lead Scoring Actually Earns Its Keep, and Where It Doesn't
AI lead scoring can spot patterns humans miss, but it can also quietly encode last year's bad targeting. Here's how to use it without trusting it blindly.
Building a B2B Sales Forecast Worksheet Your Team Will Use
Set up a B2B sales forecast worksheet with deal columns, forecast categories, coverage math and a weekly routine, plus how to measure forecast accuracy.
Building a Lead Score Reps Actually Trust
Why most lead scoring models get ignored, how to pick inputs that actually predict a close, and how to test a score before rolling it out.
Predictive Lead Scoring: Machine Learning Model or Manual Points
When a manual point-based lead scoring system is enough, and when it's genuinely worth building a machine learning model instead.
Building a Lead Score From Real Engagement, Not a Guess
How to build a dynamic lead score from actual reply and click behavior instead of static demographic points, so reps stop chasing leads that already went cold.