ICP Scoring Without a Data Vendor: A Step-by-Step Guide
The Problem with Off-the-Shelf Data
Data enrichment vendors are a default part of the modern GTM stack, but they are expensive and opaque. Industry analysis shows that annual contracts for platforms like ZoomInfo or 6sense frequently range from $25,000 to over $100,000 for a mid-market team. For this investment, we get access to vast datasets governed by black-box scoring models that make it impossible to know why a specific lead received a certain score.
The underlying data is also a significant issue. Recent research shows that B2B data decay is accelerating. A 2024 analysis by RevenueBase found a 3.6% monthly decay rate for business email addresses, pushing the potential annual decay above 35%. Other studies suggest the rate is between 30% and 40% per year. According to Gartner, this poor data quality costs organizations an average of $12.9 million annually. Building our own system gives us full control and transparency.
Step 1: Define Our ICP from First-Party Data
The most accurate data we have is our own customer list. We start by analyzing our 10 to 20 best customers—those with the highest lifetime value, fastest sales cycles, and best retention. We look for common, objective attributes across this group.
We focus on three categories of signals:
- Firmographics: These are the basic identifiers. We note the industry, employee count ranges (e.g., 100-1000), and geographic locations.
- Technographics: What technology do our best customers use? This is a powerful qualifier. We note their project management tools, communication platforms, or any other software that is complementary to our product.
- Buying Signals: We look for recent events that signal need. This could be a new funding round on Crunchbase, a key executive hire, or a surge in hiring for a specific department visible on their career page.
From this analysis, we create a simple rubric with our top five to seven attributes. We avoid subjective measures like “innovative culture” and stick to verifiable data points.
Step 2: Identify Public Data Sources
We can find the data for our rubric using publicly available sources. This manual research is the foundation of the model. The goal is to verify that the signals we chose are accessible without a paid subscription.
Our primary sources are:
- Company Websites: The most reliable source for a company's core business, product offerings, and team structure.
- LinkedIn Sales Navigator: Essential for verifying employee count, finding specific roles, and tracking company growth trends.
- Job Boards: LinkedIn Jobs, Wellfound, and company career pages show hiring signals. A company hiring for a “Head of Engineering” has a clear pain point we can address.
- Free Tech Lookups: Tools like BuiltWith, Wappalyzer, and WhatRuns offer free browser extensions for single-site lookups to get a basic picture of a company's tech stack.
It is important to acknowledge the limitations of these sources for scaled operations. Free tools have strict rate limits, and platforms like LinkedIn explicitly prohibit automated scraping in their terms of service. These methods are for building and validating a model, not for running a full-scale GTM motion.
Step 3: Build a Simple Weighted Scoring Model
Once our attributes are defined, we can build the scoring model in a spreadsheet. We create a column for each attribute and assign a weight based on its importance. To make this concrete, let's use an example. Imagine our company, SyncUp, sells an AI project management tool to mid-market tech companies.
Based on an analysis of our best customers, we create this 100-point scoring rubric:
| Category (Weight) | Attribute | Scoring Criteria | Points |
|---|---|---|---|
| Firmographics (40%) | Industry | SaaS / Technology | 15 |
| Employee Count | 251-1000 employees | 15 | |
| Geography | North America or Europe | 10 | |
| Technographics (30%) | Tech Stack | Uses Slack & Jira | 20 |
| Remote Work Tools | Uses Miro, Loom, etc. | 10 | |
| Buying Signals (30%) | Hiring Roles | Hiring for Project Manager or Head of Engineering | 15 |
| Recent Funding | Raised Series A, B, or C in last 18 months | 10 | |
| Negative Signal | Recently signed with Asana, Monday.com | -20 |
Now, we score a prospect. Let's look at a fictional company, Innovatech Solutions:
- Profile: A 350-employee SaaS company in Austin, TX. They use Slack, Jira, and Miro. They are hiring a “Senior Project Manager” and raised a Series B nine months ago.
Here is their score based on our rubric:
- Firmographics: 15 (SaaS) + 15 (350 employees) + 10 (North America) = 40 points
- Technographics: 20 (Slack & Jira) + 10 (Miro) = 30 points
- Buying Signals: 15 (Hiring PM) + 10 (Series B) = 25 points
Innovatech Solutions scores a 95/100. This turns a vague “good fit” into a quantifiable reason to prioritize them for outreach. We then run our best customers and a few known bad-fit customers through the model. If the scores don't clearly separate good from bad, we adjust the weights until they do.
Step 4: Putting the System to Work
With a calibrated model, the final step is to validate it on a fresh list of prospects. We manually research and score a batch of 50 target accounts like Innovatech Solutions. This process establishes a baseline for the time it takes and the quality of the qualified leads the system produces. We find the results are more accurate than any vendor's, because the model is tailored perfectly to our business.
This manual work is high-value but not scalable. Once we have a proven, in-house scoring model, we can use an AI agent like Drevon to automate the data collection and scoring. We can instruct it to find companies matching our rubric, gather the specific data points from our chosen sources, and deliver a scored list. This automates the repetitive work without sacrificing the accuracy and control of a custom-built system.
Anyone can start building this model this weekend. A simple spreadsheet and a few hours of focused research are all that is needed to take control of lead quality.