ICP scoring is the practice of turning your ideal customer profile into a grade attached to every prospect on your list, so you can sort best-fit accounts to the top and focus outbound on the companies most likely to buy. For outbound specifically, you score on fit (the structural question of whether you should be selling to this account at all) using firmographics, tech stack, and the facts you can scrape about each company. Done well, it is the highest-leverage filter in your pipeline: teams that score against their ICP report two to three times higher conversion, because reps spend their effort on accounts that can actually close. This guide shows how to score an outbound list against your ICP inside a spreadsheet.
What is ICP scoring?
ICP scoring converts your ideal customer profile from a description into a number or grade on each row of your list. Instead of a slide that says "we sell to mid-market logistics firms," you get an explicit label on every prospect: this one is a strong fit, this one is possible, this one is not a fit at all. The score is the operational version of the profile. If you have not defined the profile yet, start with the ideal customer profile entry, because the scoring is only as good as the ICP it encodes.
The point of making it a score rather than a description is that a score sorts. A list of ten thousand prospects is useless until you can rank it, and ICP scoring is what produces the ranking.
Fit versus intent: score them separately
The single most important design decision in ICP scoring is to keep fit and intent as two separate numbers. They answer different questions and they change at different speeds.
- Fit answers: should we be selling to this account at all? It is built from firmographics (industry, company size, revenue, geography), tech stack (the tools they already run), and persona (is your contact a decision-maker). Fit is relatively stable, because a company's industry does not change month to month.
- Intent answers: is this account showing buying behaviour right now? It is built from behavioural signals and changes constantly.
For cold outbound, fit is what you score. A prospect can be a perfect ICP match with zero current engagement, and that is not a reason to skip them: it is the definition of an outbound target. You are reaching out precisely because they have not raised their hand yet. So when you score an outbound list, you are scoring fit, and intent is a separate concern you can layer on later if you have the data.
What goes into a fit score?
A fit score is built from the attributes your ICP cares about. The common ones across 2026 scoring models are:
- Firmographics: industry or vertical, employee headcount, revenue band, geography.
- Tech stack: the tools and platforms the company already runs, which can signal both fit and a specific need.
- Growth signals: funding stage, hiring pace, recent expansion, which suggest budget and momentum.
- Persona match: whether the contact you have is a decision-maker or an influencer.
- Disqualifiers: the traits that rule a company out entirely, so a single hard mismatch can drop the score to zero regardless of everything else.
A useful pattern many teams use is a letter grade (A to F) for fit, kept separate from any numeric intent score. The grade is what you sort and filter on for outbound.
Start with rules, graduate to prediction
Around three-quarters of B2B companies are expected to use some form of AI-driven scoring by the end of 2026, but the right starting point for most teams is still a rule-based model, not a predictive one. Predictive scoring needs conversion history to train on, and a new outbound motion does not have that history yet. Begin with explicit rules that encode your ICP, then graduate to prediction once you have enough closed deals to learn from. The rules are also easier to inspect and correct, which matters when you are scoring a list you are about to spend money on.
How to score an outbound list in a spreadsheet
A spreadsheet is the natural place to score, because the firmographic data is already sitting in columns and the score lands as one more column you can sort and filter. In ReplyLabs the scoring step is an AI prompt that reads the relevant columns and writes a verdict back. ReplyLabs prepares the list inside Google Sheets, it does not send the mail, so the scored, filtered list is what you hand to your own email sender. This is the targeting step in the wider cold email from Google Sheets workflow.
The sequence:
- Make sure the rows hold the inputs your ICP cares about. Industry, employee count, and ideally a scraped "About" column with real text. If the firmographic columns are thin, enrich first via lead enrichment in Google Sheets, because the score is only as good as the data it reads.
- Write a scoring prompt that encodes your ICP. For example: "Based on
{{Industry}},{{Employee count}}and{{About}}, reply with exactly one of: Strong fit, Possible fit, Not a fit. A company is Not a fit if it has fewer than 50 employees or is outside B2B software." - Sample twenty rows and read every verdict. Confirm the model is applying your rules the way you meant. Adjust the prompt where it disagrees with your judgement.
- Run the full list. The verdicts land in a new column.
- Filter to the fits. Keep Strong and Possible, drop Not a fit, and carry only the survivors into enrichment and personalisation.
Because the AI runs server-side rather than as a spreadsheet formula, there is no six-minute Apps Script ceiling: one scoring prompt can grade fifty rows or fifty thousand in a single pass.
Why score before you personalise and send
Scoring is the filter that comes before you spend real money and real sender reputation on a list. Deals sourced from ICP-fit accounts have been reported to close at around 68% against 22% for non-fit accounts, so the rows you drop at the scoring step are mostly rows that were never going to convert anyway. Filtering them out first means you never spend an AI-personalised opener, or a precious send from a warmed inbox, on a company outside your ICP.
There is a deliverability benefit too. Per-inbox volume limits in 2026 are low, around 15 to 25 cold emails per inbox per day, so your sending capacity is scarce. Scoring keeps the final list small enough to fit that capacity and ensures every send goes to a company worth contacting.
What does ICP scoring cost in a spreadsheet?
The scoring step is an AI step, priced at the provider's raw cost times 1.25 plus a small base fee per succeeded row, with the exact figure shown in a cost preview before the run. You are charged only for rows that return a verdict. New accounts get $20 of free credit, enough to score and prepare a real list end to end. On higher tiers you can bring your own AI key and pay only the raw provider rate. See getting started to set up a first run.
Common questions
What is the difference between ICP scoring and lead scoring?
ICP scoring grades fit: how well a company matches your ideal customer profile, based on firmographics and tech stack. Traditional lead scoring usually mixes in intent: behavioural signals like email opens and site visits. For cold outbound you score fit, because your prospects have not engaged yet by definition. Keep fit and intent as separate numbers.
Should I score on fit or intent for outbound?
Fit. An outbound prospect is, by definition, not showing intent yet, which is why you are reaching out cold. A perfect ICP match with zero engagement is the ideal outbound target, not a reason to skip them. Score fit, filter to the fits, then personalise.
Can I score a list without a defined ICP?
Not usefully. The score is just your ICP turned into a grade, so without a clear profile there is nothing to grade against. Define the ideal customer profile first, including the disqualifiers, then encode it in a scoring prompt.
Does ReplyLabs send to the scored accounts?
No. ReplyLabs prepares the list in Google Sheets: enrich, score, and personalise. You export the filtered, scored columns and send through your own email service provider or sequencer. Sending and reply handling live in your sending platform.
How accurate is AI scoring against an ICP?
Accurate enough to be useful as a first-pass filter, especially when you sample twenty rows and tune the prompt before the full run. Rule-based prompts that encode explicit ICP criteria are predictable and easy to correct. Treat the score as a sort-and-filter tool, not a final verdict, and the accuracy is more than sufficient for outbound targeting.