ICP scoring is the practice of grading every account against your ideal customer profile and turning the result into a number or letter grade. The score measures fit: how closely a company’s stable attributes match the customers who already buy from you, stay, and expand.
Notice what the score ignores. It says nothing about whether anyone at that company opened your email or visited your pricing page yesterday. Fit exists before the first touch, and that is exactly what makes it useful for deciding where your team’s time goes.
I have built fit scoring models for three B2B software companies and audited about a dozen more since 2020. Honestly, most of them failed the same simple test: score last quarter’s won and lost deals, then check whether the model can tell them apart. So this guide covers what an ICP score measures, how it differs from lead scoring, what goes into the model, and how to build one that passes that test.
What Does an ICP Score Actually Measure?
An ICP score measures resemblance: how closely an account matches your best existing customers on attributes that change slowly, if at all. Typical inputs include industry, employee count, revenue band, geography, tech stack, and business model.
The phrase best existing customers deserves precision. I define it as accounts that closed at a healthy speed, renewed at least once, and expanded or referred others. A big logo that took 14 months to close and churned in year one does not belong in that set, no matter how impressive it looked in pipeline reviews.
You will also hear the same idea called fit scoring, account fit scoring, or simply a fit score. Some teams output a 0 to 100 number. Others use letter grades from A to D. The scale matters far less than the discipline behind it.
One more boundary is worth drawing. ICP scoring grades companies, not people. Whether your contact is a VP or an intern is a persona question, and it belongs one level down, in lead-level qualification rather than account fit.
📌 Example: In 2023 I scored the full customer base of a 40-person SaaS company. The sales team swore their best segment was enterprise retail. Yet the won-deal data said mid-market logistics: 2.1x the win rate and half the sales cycle. Nobody's opinion settled that argument. Their own closed deals did.
Is ICP Scoring the Same as Having an ICP?
No. An ICP is a written description of your best-fit customer, while ICP scoring is the measurement system that applies that description to every account in your market. One is a document. The other is an operation.
That difference sounds academic until you watch it in practice. Plenty of teams have a beautiful profile slide that says mid-market SaaS, North America, growing sales team. Then a list of 5,000 accounts arrives, and nobody can say which 200 match the slide. Scoring closes that gap by turning prose into arithmetic.
There is a second benefit I did not appreciate until I had built a few models. Scoring forces honesty. A slide can hold vague words like innovative companies forever. Scoring models cannot, because every attribute needs a field, a value, and a weight. If you cannot measure it, it quietly falls out of the profile, and what remains is the part you can actually operate on.
ICP Scoring vs Lead Scoring: What Is the Difference?
In one line: ICP scoring measures fit, lead scoring measures engagement, and blending the two into a single number is the most common scoring mistake I see. The two answer different questions, so they belong in separate fields.
Engagement tells you who is active right now. Someone downloaded a guide, attended a webinar, or replied to an email, and their score went up. However, activity without fit is noise. Students, competitors, and job seekers generate plenty of clicks and close exactly nothing.
Fit tells you who is worth pursuing even in total silence. A perfect-fit account that has never heard of you is a better outbound target than an eager visitor from the wrong market. Wikipedia’s entry on lead scoring covers the engagement side and its history in more depth.
Some platforms make the split explicit. Tools descended from Pardot keep an engagement score and a separate fit grade on every record, which is why you sometimes hear fit scoring called lead grading. Either way, the mechanics of the engagement half are well documented in Salesforce’s lead scoring guide.
| Dimension | ICP Scoring (Fit) | Lead Scoring (Engagement) |
|---|---|---|
| What it measures | How closely an account resembles your best customers | How actively a person interacts with you |
| Typical inputs | Industry, headcount, revenue, geography, tech stack | Email opens, page visits, downloads, replies |
| How fast it changes | Slowly, over months or quarters | Daily, sometimes hourly |
| Exists before first touch | Yes | No |
| Best for | Outbound prioritization, routing, disqualification | Timing follow-up, nurture triggers |
| Usual owner | RevOps and sales | Marketing |
The Fit and Engagement Matrix
The real payoff comes from crossing the two scores. Every account lands in one of four boxes, and each box has an obvious play. This 2×2 is the framework I draw on every whiteboard when a team asks where to start.
| Quadrant | Who They Are | The Play |
|---|---|---|
| High fit, high engagement | Your ideal accounts, actively evaluating | Fast lane: route to a rep within minutes, tightest SLA, no forms in the way |
| High fit, low engagement | Your ideal accounts, not yet aware of you | Priority outbound and targeted ads; never wait for a hand-raise |
| Low fit, high engagement | Wrong market, lots of clicks | Self-serve, community, or a polite no; protect rep time |
| Low fit, low engagement | Wrong market, no interest | Disqualify or recycle; remove from paid audiences to save budget |
Most teams already run the top-left box reasonably well. The money hides in the top-right one. High-fit accounts that never raised a hand rarely enter the funnel on their own, and only a fit score can surface them before a competitor does.
💡 Pro Tip: Report the four quadrants monthly, not just the pipeline total. If more than half of your open pipeline sits in the low-fit row, scoring is not your problem. Your demand generation is aimed at the wrong market, and no routing rule will fix that.
What Goes Into an ICP Scoring Model?
A working model has four parts: attributes, weights, thresholds, and tiers. Miss any one of them and the score stops driving decisions.
Attributes are the traits you measure, and they fall into three groups. Firmographic data covers industry, headcount, revenue, and geography. Technographic attributes cover the tools a company already runs, from its data warehouse to its billing platform. Signal attributes move slowly but reveal direction: hiring velocity, funding rounds, and new office openings.
That third group overlaps with buying signals, so keep the axes clean. Slow-moving signals like a growing sales team belong in the fit model. Fast-moving intent data, like topic research spikes, belongs on the engagement axis instead.
Weights encode how much each attribute matters. If industry separates your winners from your losers more than company size does, industry earns more points. For a public example, Aviso’s account scoring approach blends firmographic, technographic, and go-to-market signals into a single 0 to 100 score.
Here is a simplified weighted model I might ship for a mid-market SaaS seller. Yours will differ, and it should:
| Attribute | Example Ideal Value | Weight (Points) |
|---|---|---|
| Industry | SaaS, fintech, or logistics | 25 |
| Employee count | 100 to 1,000 | 20 |
| Region | North America or DACH | 15 |
| Tech stack | Modern cloud stack present | 15 |
| Hiring velocity | Sales or ops team growing | 15 |
| Revenue band | $10M to $100M | 10 |
Thresholds then turn the number into tiers your team can act on. Most teams use four:
| Tier | Score Range | What It Means | Treatment |
|---|---|---|---|
| A | 80 to 100 | Near-clone of your best customers | Priority routing, tightest SLA, top of outbound lists |
| B | 60 to 79 | Strong fit with one gap | Standard sales follow-up and sequences |
| C | 40 to 59 | Partial fit | Nurture and self-serve motion |
| D | Below 40 | Poor fit | Disqualify, recycle, or refer out |
Tiers beat raw numbers in daily use. No rep debates whether a 74 outranks a 71. An A next to the account name ends the discussion and starts the call.
How Do You Build an ICP Scoring Model From Won and Lost Deals?
You build one by mining closed-won, closed-lost, and churned accounts for the attributes that separate them, then weighting those attributes and backtesting the result. Expect the first working pass to take about a week, not a quarter.
Step 1: pull three lists. Export every closed-won, closed-lost, and churned account from the last 12 to 24 months out of your CRM. Churn matters as much as losses here. A deal you won and then lost to cancellation is the most expensive kind of bad fit.
Step 2: fill the attribute gaps. Most exports arrive with industry blank and headcount stale, and a model built on blanks scores blanks. This is where B2B data enrichment earns its keep. Teams typically append industry, employee count, revenue, and technographics with an enrichment provider such as CUFinder before modeling anything. That said, no enrichment tool rescues a model whose weights nobody ever validates against real outcomes.
Step 3: find the separating attributes. For each candidate attribute, compare win rate and retention across its values. Maybe logistics companies win at 34 percent while retail wins at 12. An attribute earns a spot only when the gap is that visible, and 6 to 10 attributes is plenty. PhantomBuster’s ICP score guide walks through a similar attribute-selection exercise if you want a second reference.
Step 4: set weights from the data, not the room. Give the widest win-rate gaps the heaviest weights. Resist the workshop urge to negotiate points by seniority. Your closed deals already voted.
Step 5: add disqualifiers and negative points. Some attributes should not just score low. They should subtract, or block outright. More on negative scoring below, because skipping it is one of the classic failures.
Step 6: backtest on last quarter’s deals. Score every deal that closed last quarter with the draft model. Wins should cluster in tiers A and B. If lost and churned deals also land in tier A, adjust the weights and run it again. I consider a model shippable when roughly 70 percent of recent wins score A or B and few losses do.
📌 Checkpoint: Backtest before rollout, every single time. A model that cannot separate last quarter's wins from last quarter's losses will not separate next quarter's either. Fifteen minutes in a spreadsheet here saves your SDR team a month of chasing the wrong tier.
Where Does the ICP Score Live and What Should It Trigger?
The score lives as a field on the account record, and it should trigger routing, response SLAs, outbound priority, and disqualification without a human touching it. A score nobody wired into workflows is a spreadsheet hobby, not an operating system.
Start with lead routing. When a form fill arrives from an A-tier account, it should reach a senior rep within minutes, not sit in a round-robin queue behind D-tier noise. HubSpot’s scoring documentation shows how score properties feed directly into lists and workflows, and every major platform has an equivalent.
Tiers also set service levels. A sensible ladder looks like this: A-tier inbound gets a human response within the hour, B-tier gets same-day, and C-tier flows into automated nurture. Nobody has to argue about coverage anymore, because the SLA is attached to the grade.
For outbound, the fit tier orders the territory. SDRs open their day on A-tier accounts, and campaign budgets follow the same ranking. Meanwhile, the score also powers lead qualification at the gate: D-tier hand-raisers get a polite self-serve path instead of a discovery call.
If you want a working example in public, PostHog’s handbook publishes theirs. Inbound leads get a fit score out of 70, accounts matching at least two of three fit criteria see an instant meeting scheduler, and lower scores route to manual review. Their stated aim from those scored leads is a 20 percent conversion rate.
One more trigger deserves a mention: advertising audiences. Sync the A and B tiers into your ad platforms as target lists, and suppress the D tier everywhere. In my experience this is the fastest payback of the whole project, because paid budget stops subsidizing accounts your own model says will never buy.
🔍 Field Note: In 2024 I audited a team whose model was mathematically fine but lived in a BI dashboard nobody opened. Reps kept cherry-picking accounts by logo. We moved the tier into the account list view and made it sortable. A-tier coverage went from 40 percent to 90 percent in six weeks, with zero changes to the model itself.
What Is Negative Scoring and Why Does Your Model Need It?
Negative scoring subtracts points for attributes that predict a bad outcome, and it protects your team from accounts that look busy but never pay off. Every model needs it, because addition alone cannot describe your losses.
Two mechanisms do the work. Negative points reduce the score, say minus 15 for a segment with a churn history. Hard disqualifiers zero it out entirely: competitors, sanctioned regions, or companies below your minimum viable size. Use points for risk and disqualifiers for never.
The usual suspects for negative points look like this:
- Churn-prone segments. Industries or sizes that cancel well above your base rate.
- Support-heavy profiles. Segments whose ticket volume erases the margin on the deal.
- Discount-dependent buyers. Segments that only ever closed at 40 percent off list.
- Shrinking companies. Layoffs and hiring freezes predict stalled procurement.
Here is why this matters in practice. In 2022 a client of mine kept closing marketing agencies. The deals were quick wins, so nobody questioned them. Then we ran churn by segment, and agencies cancelled at three times the base rate within a year. A minus 15 for agencies moved most of them to tier C, and the pipeline finally told the truth.
Without negative signals, your model only remembers the honeymoon. With them, it remembers the divorce too, and that memory is what keeps acquisition and retention pointed at the same customers.
How Often Should You Recalibrate an ICP Score?
Quarterly. Re-run the backtest on the most recent quarter’s closed deals every three months, and adjust weights whenever the model starts missing. The whole exercise takes an afternoon once the first version exists.
Why so often? Because the ground moves. Your product ships features that open a new segment. Pricing shifts you upmarket. A competitor floods your old niche. The model encodes yesterday’s winners, and yesterday keeps receding.
I watched this bite a company in 2025. They had moved upmarket over two quarters, but the model still gave full points to the 50 to 200 headcount band. As a result, SDRs kept filling pipeline with accounts the new pricing had already priced out. One recalibration fixed it in an afternoon, after two quarters of quiet damage.
Treat the review like closing the books. Same week each quarter, same owner, same three questions: did A-tier win more, did any tier surprise us, and what changed in the market that the attributes do not yet capture?
Certain events also justify an off-cycle review. A pricing change, a new product line, an acquisition, or entry into a new region can each invalidate a weight overnight. When one of those lands, do not wait for the calendar. Rerun the backtest that week and adjust before the pipeline fills with the wrong accounts.
What Are the Most Common ICP Scoring Mistakes?
The four I keep meeting are gut-feel weights, scoring the customers you wish you had, freezing the model forever, and ignoring negative signals. Each one has a cheap fix, and I have watched each one waste real quarters.
Gut-feel weights top the list. In 2021 I sat in a workshop where scoring weights were set by vote, sticky notes and all. The finished model ranked the company’s three biggest churned accounts as A-tier. Weights have to come from won and lost data, because the loudest voice in the room is not a dataset.
Scoring aspirational customers is subtler. Teams tune the model to the market they want, usually enterprise, instead of the customers who actually close. A founder I worked with in 2022 insisted on Fortune 500 attributes while his wins were 200-person companies. His version of the A-tier list produced two meetings in a month. The data-tuned version produced fourteen.
Freezing the model is the slow killer. One score forever means the model drifts further from reality every quarter, exactly like the upmarket story above. Put a recalibration date on the calendar before you ship version one.
And ignoring negative signals rounds out the set. A model that only adds points falls in love with every large account in a famous industry. Subtract for churn-prone segments and block the true never-buyers, or your best reps will keep winning deals you should not want.
🧠 Worth Remembering: An ICP score is a hypothesis about who buys, written in numbers. Hypotheses expire. The teams that win treat the model as a living document with an owner, a review date, and a changelog, not as a settings page someone configured once in 2023.
How Is AI Changing ICP Scoring?
AI is replacing hand-set weights with models that learn them straight from your closed deals. Instead of you deciding industry is worth 25 points, a predictive model runs the math across every field and finds the patterns you missed.
That approach is usually sold as predictive scoring, and TechTarget’s scoring definition explains how it extends the manual method. The appeal is honest: learned weights update themselves, and they can catch interactions between attributes that no workshop would spot.
Still, two cautions from the field. Small datasets lie. With under a few hundred closed deals, a learned model mostly memorizes noise, and a hand-weighted one will quietly beat it. Second, AI amplifies whatever data quality you feed it. Stale headcounts in, confident nonsense out. Clean and enrich the inputs first, then let the machine set the weights.
My advice for most teams is a staged path. Ship the hand-weighted model this quarter, because it forces you to understand your own winners. Once you have a few hundred clean closed deals and a working backtest habit, test a predictive layer against it. Keep whichever one wins the backtest, and stay suspicious of any score you cannot explain to a rep in one sentence.
Frequently Asked Questions
What is an ICP score?
An ICP score is a number or grade showing how closely an account matches your ideal customer profile. It is built from stable attributes like industry, employee count, revenue, and tech stack, weighted by how strongly each one predicts a won deal in your own history.
What is the ICP scoring model?
The ICP scoring model is the rulebook behind the score: which attributes you measure, the weight each one carries, the negative points and disqualifiers, and the thresholds that split scores into tiers. Put simply, the score is the output and the model is the machine that produces it.
How do you calculate an ICP score?
Score each attribute against your ideal value, multiply by its weight, subtract negative points, and sum the result. For example, a SaaS company with 300 employees in North America might earn 25 + 20 + 15 points on those three attributes alone. Most teams normalize the total to a 0 to 100 scale.
What is a good ICP score?
On a 0 to 100 scale, 80 and above usually marks A-tier, meaning the account closely mirrors your proven customers. The honest answer, though, is that a good score is whatever range your backtest validates: the band where last quarter’s wins actually clustered.
What is the difference between ICP scoring and lead scoring?
ICP scoring grades an account’s fit using stable attributes, while lead scoring grades a person’s engagement using recent behavior. Fit exists before any interaction; engagement only exists after. Strong teams keep both scores and cross them in a fit-by-engagement matrix to decide the next action.
What is the difference between lead grading and lead scoring?
Lead grading is an older name for the fit side, popularized by Pardot: a letter grade for how well a lead matches your target profile. In that vocabulary, lead scoring tracks engagement points while grading tracks fit. So lead grading and ICP scoring describe essentially the same idea at different levels.
What is ICP lead qualification?
ICP lead qualification means using the fit score as a gate in your qualification process. High-fit leads skip ahead to a rep, mid-fit leads get standard discovery, and low-fit leads are routed to self-serve or declined early. It saves discovery calls for accounts that can actually become customers.
How many attributes should an ICP scoring model have?
Six to ten. Fewer than six usually misses a real pattern in your won deals, while more than ten adds maintenance without adding accuracy. Every attribute must earn its place by showing a clear win-rate or retention gap in your own closed data.
So that is ICP scoring: a weighted, backtested measure of fit that decides where your team’s hours go before anyone clicks anything. Build it from your won and lost deals, wire it into routing and SLAs, subtract for the bad patterns, and recalibrate every quarter. The math is simple; the discipline is the product.
Bad
Okay
Good
Amazing
