Guides

The Startup Guide to Building an AI-assisted ICP scoring model.

for under $150/month

Introduction

Ideal Customer Profile (ICP) scoring is not new. Every go-to-market team has some way to pull data about accounts to sort the good leads from the bad ones. But the problem is best-fit buyers don’t just look a certain way—they act a certain way. Companies of all sizes now need to assess intent signals, buying behavior, engagement data, and product fit to determine the true likelihood of conversion.

The problem? If you’re a founder or member of a scaling startup, you’re probably not ready to invest in the platforms needed to build a sophisticated scoring system. So you take every call you can get, reach out to every person with a particular title, and very often miss following up with the prospects actually worth your time.

Fortunately, AI and automation now make it possible to scale this process and make ICP scoring more precise than ever before. AND you don’t need a six-figure RevOps budget, expensive platform, or a PhD in data science to get started. You just need a better playbook—and maybe ChatGPT.

Why startups need to think about ICP scoring earlier

You’ve built a great product, that might even have a healthy mix of organic attachment, and founder-network sales. But soon investors will ask you to ramp up your GTM engine. If you’re able to hire, your first might look like this:

  • Founding marketer/community lead → To create actually helpful content
  • Founding seller/sales engineer → To build pipeline of the right audience
  • Founding CSM → To support the success of—and upsell—early customers

To execute on each of those ‘jobs to be done,’ you need to lock into who your product is best fit for, where those people are employed, and what they’re trying to solve. You need to not only score folks coming in the door, to conserve limited GTM resources, but build a target account list so you know who to proactively approach.

The faster you can codify an early version of your ICP for your team to rally around, the faster you can:

  • Create the right content, that actually converts
  • Focus sellers (or founders!) on the right audience
  • Find the right upsell triggers, to drive low-friction deals
  • Ensure you’re optimizing for the right use cases post-sale

A strong ICP scoring system helps with all.

Why traditional ICP data sources no longer cut it

For years, companies have relied on basic firmographic data from providers like ZoomInfo to identify their Ideal Customer Profile (ICP). While metrics like headcount, revenue, and location are useful starting points, they rarely capture what truly makes a customer "ideal" for your business. Knowing a company has 200 employees and is technically in “software” doesn’t tell you:

  • If they’re growing or flatlining
  • If they sell B2B or B2C
  • If they have a sales motion at all
  • If they’re currently in-market for what you sell

Sure, you can filter out obvious mismatches (e.g., a bakery chain probably doesn’t need your DevOps platform), but for any real precision, you need more signal. At Koala, for example, we need to identify:

  • B2B companies only (hard requirement)
  • Organizations with sufficient ACV to justify outbound sales
  • Companies with high website traffic or PLG models generating substantial lead volume

These specific traits simply don't exist in standard databases, making traditional ICP scoring models ineffective. Even with access to firmographic data, teams often assign arbitrary weights to different variables without validating their impact on actual conversion rates. This leads to a misalignment between ICP scoring and real-world sales success. So, you have two problems to solve:

  1. You don’t have the data you need. Most databases won’t tell you if a company is growing fast, using competitors, or has an outbound motion.
  2. You don’t have a scalable way to score that data. Even if you do pull those insights, how do you apply them across tens of thousands of companies without spending a fortune?

AI fixes both—if you know how to use it.

How AI makes ICP scores actionable

AI has revolutionized ICP-based outbound by automating the research tuning process that used to follow the moment an account was awarded a high score. By leveraging AI-driven automation, companies can:

  • Eliminate manual research time
  • Automate qualification based on hard-to-find custom data
  • Improve accuracy in identifying high-fit prospects
  • Surface use cases prospects care most about
  • Customize ICP scoring to fit their exact needs
  • Continuously refine ICP models based on real-time data

AI can analyze a company’s digital footprint, including website traffic, hiring trends, social media activity, and customer sentiment. This enables businesses to go beyond static firmographic data and create dynamic ICP models that evolve with market conditions.

For example, an AI model might analyze job postings to determine if a company is expanding its sales team, a strong signal that it might be investing in outbound sales. Or, it might assess web traffic patterns to identify companies with growing interest in solutions like yours. These insights help businesses prioritize leads with real potential, rather than relying solely on traditional firmographic factors.

How to build an AI-scoring system and target account list for under $150/month

A modern ICP model should do three things:

  1. Filter out the obvious “no”s.
  2. Enrich with traits that matter to your sales motion.
  3. Prioritize the companies most likely to convert, now.

The ability to build hyper-personalized ICP scoring is exciting, but it comes with a financial consideration. With most company databases containing 50M+ companies, even at just 10¢ per company for custom research, you're looking at a $5M price tag!

Without a thoughtful approach, data enrichment costs can quickly spiral out of control. The challenge is balancing accuracy with cost-effectiveness, ensuring that every dollar spent contributes to better sales outcomes. Let’s walk through exactly how to build that—and how much it’ll cost you.

Step 01

Define What Actually Makes a Good Customer

Start here: look at your last 20 closed-won deals. Not just company size or industry—look deeper. What do your best customers do before they buy?

  • Did they visit certain pages?
  • Hire certain roles?
  • Already use a specific tool?
  • Come from a certain referral channel?
  • Have a specific pain that shows up in their job listings or product updates?

These are the signals your ICP model should be scoring on—not whatever your data vendor gave you by default.

Example from Koala: We’re most useful for companies either trying to find their ideal audience, or for companies that need to scale an outbound motion for their whole sales team, using product, marketing, and publicly available signals. So we tend to look for whether a company:

  • Is B2B
  • Has an ACV over $10k
  • Runs outbound (job posts for SDRs, mentions of HubSpot or Outreach, etc.)
  • Has a marketing or product-led motion that generates signals
  • Is already showing interest (visiting pricing, reading docs, etc.)

We don’t get that from ZoomInfo. While we use our AI agents to extract it (accessible even in our Free tier), you can also use some well-crafted prompts in ChatGPT to help, which we cover next.

Step 02

Build a Cost-Efficient Scoring Funnel

Setting up AI-driven scoring is much more affordable than most teams expect, especially if you're smart about layering your filters and using GPT or scraping tools strategically. Here’s a rough cost breakdown based on a typical setup for a B2B sales org with a TAM of about ~100,000 companies:

Tier 1

Basic Filtering (Free)

Use existing data providers, CSVs, or ChatGPT to filter by:

  • Size (employee count, funding stage)
  • Industry
  • Region

This removes ~70% of the junk right away. If you’re using something like Crunchbase, you’re spending ~$0 here. Or use ChatGPT to classify based on scraped LinkedIn descriptions.

Prompt example: Here’s a company name: [paste it]. Where is the HQ for this company? How would you categorize the industry?”

Prompt Cost: ~$0

Tier 2

AI-Assisted Signal Extraction (GPT-3.5 or GPT-4 Turbo)

Use free GPT models or browser agents to pull lightweight signals, like:

  • What tools does the company mention on their careers page?
  • Is this B2B or B2C?
  • Have they raised at least $XM in funding?
  • Does their website mention integrations with your competitors?

You can build a cheap scraper with GPT or use a no-code tool (e.g. Browse AI, Bardeen AI, or even a Google Sheet + Apps Script).

  • Apply to: ~10,000–20,000 records monthly
  • Batch ~20–30 companies per prompt using structured, yes/no questions

Prompt example: “For each of the companies below, answer the following: 1. Is it B2B or B2C? 2. Are they hiring salespeople? 3. Have they raised at least $5M? 4. Do they mention outbound or cold outreach? 5. Is there a pricing page or PLG signal on the site?”

Prompt Cost: Free up to 250 accounts with Koala, else use GPT-3.5 for basic yes/no (at $0.0005/1K tokens) at ~$20–$40/month

Tier 3

Deep Scoring (Optional, Only on Hot Leads)

For leads that pass tiers 1 and 2, apply heavier AI research or paid enrichment. Only run this for your top 5-10% of leads or 1,000–2,000 records/month. Use Koala’s Free tier, or GPT-4 Turbo to summarize blogs, parse product pages, or qualify more nuanced traits, for this audience like:

  • Summarize blog content or job posts
  • Pull G2 intent data, Bombora surges, or product engagement
  • Pull recent social interactions, sponsorships, or event information

Prompt example: “Based on this company’s website, hiring page, social interactions, and recent blog posts, is this company likely investing in outbound sales? Summarize the evidence.”

Prompt Cost: Free up to 250 accounts with Koala, else ~$0.01–$0.03 per lead (depending on length) at ~$50/month if using GPT

Lean Monthly Budget Breakdown

Budget Breakdown

Total: ~$80–$150/month
Step 03

Apply Scores and Route to Action

Once you’ve got this data:

  • Assign points for each signal (don’t overthink weighting to start—just be consistent)
  • Bucket leads into A, B, C tiers
  • Auto-route A leads to sales or drop them into a high-intent sequence

Use a CRM workflow, a spreadsheet, or Koala Plays to make sure your reps don’t have to hunt for this. Your goal: no more lead soup, no more guessing.

Tips to Keep Costs Down:

To keep costs down, be thoughtful about when and where you apply enrichment, what models you use (GPT-3.5 vs GPT-4), and how much you automate vs batch manually. The result is a sharp, AI-assisted ICP scoring engine running monthly for under $150. To be clear, even a spreadsheet-powered v1 using ChatGPT in your browser could get you 80% of the way there before spending a dime!

  • Batch prompts (send 10–50 companies at once to GPT-4 Turbo)
  • Use GPT-3.5 where you don’t need complex reasoning (~10x cheaper)
  • Build a caching layer—don’t re-enrich the same company twice
  • Apply deep scoring only to your in-funnel or surging segments

That’s your whole market, deeply scored, for less than a Salesforce admin.

Bonus Free Tools to Stretch Further

Browse AIBrowse AI

Free plan allows up to 50 tasks/month to scrape job pages, pricing, or tech mentions.

Bardeen AIBardeen AI

Free Chrome automation for pulling site info into Airtable/Sheets.

GPT-3.5 via OpenAI APIGPT-3.5 via OpenAI API

Great for classification and short yes/no enrichment.

KoalaKoala

Great for signal and AI enrichment, even on the Free tier.

Best Practices for Smarter, AI-Driven ICP Scoring

Your ICP model doesn’t need to be perfect—it just needs to be better than whatever generic lead list you're working today. Even a lightweight scoring system that reflects your actual closed-won patterns will outperform random outreach. But if you want your model to get sharper over time and truly support scale, here’s how to keep it useful and cost-effective:

Start with your own conversion data

Use closed-won and closed-lost deals to define what “good” actually looks like. Don’t rely on gut feel or generic templates.

Validate scoring performance

Track how your A/B/C buckets convert through each stage of the funnel. If your “A” leads aren’t progressing, recheck your assumptions.

Get feedback from reps

Sales and SDR teams are already qualifying in their heads. Ask them what your model is missing—or what it’s over-weighting.

Use multiple signal types

AI works best when it has more than one lens. Combine firmographics with behavioral signals (like site activity or product usage), intent data, job postings, and tech stack info.”

Automate re-scoring cycles

Your TAM isn’t static, and your ICP shouldn’t be either. Set a cadence—monthly or quarterly—to refresh scores using AI agents or scraping workflows.

Keep prompt design clean

If you’re using GPT for enrichment, prompt structure matters. Simple, specific questions yield more accurate, repeatable results (e.g., "Does this company appear to have a sales team?").

Push scores into your sales stack

Don’t let scores sit in a spreadsheet. Feed them into your CRM, lead routing logic, or platforms like Koala to prioritize who gets worked (and how).

Apply scoring to engaged leads, too

Just because a lead comes inbound doesn’t mean it’s a fit—but it might be. Run the same scoring logic on free trials, whitepaper downloads, and signups. It’s often your warmest segment.

Use your model to disqualify, not just prioritize

A good ICP model keeps junk leads out of rep workflows. Set up thresholds or tags to filter out noise before anyone wastes a touchpoint.

Avoid common traps:

  • Don’t enrich leads before filtering—you’ll waste money fast
  • Don’t let weights get too theoretical—simple point-based models are easier to maintain

  • Don’t wait for RevOps to operationalize everything—start with a spreadsheet, prove it works, then automate

The Future of ICP Scoring with AI

As AI technology continues to advance, ICP scoring will become even more sophisticated. Predictive analytics, intent-based data, and machine learning models will further refine the ability to identify, prioritize, and engage with ideal customers. Companies that adopt AI-driven ICP scoring today will gain a significant competitive advantage, reducing wasted sales effort and increasing conversion rates. By adopting AI-driven ICP scoring, businesses can focus on the leads that truly matter, maximize sales efficiency, and stay ahead in an increasingly competitive market.

Need help pulling this together? Intent platforms like Koala that use AI agents for lead enrichment can help. Ensuring "warm" contacts that visit your website or start a product trial are part of your ICP consideration process helps surface high-potential leads that may have been overlooked in traditional ICP models, and enriching these leads with agentic qualifiers ensures focused attention on what really matters.

Shoot us a note for a deeper discussion, or set up your free account today!