Most lead scoring systems take weeks to configure. You define dozens of rules, assign point values to firmographic data, tweak weights, and still end up with a model that flags bad leads and misses good ones.
AI-powered lead scoring is different. Instead of rigid rules, you describe what a great customer looks like in plain language, and an LLM reasons about fit for each lead. The result: scoring that understands nuance, context, and timing — not just company size and industry.
This tutorial walks you through setting up AI-powered lead scoring in Scout from scratch. By the end, you'll have a working scoring model reviewing real leads.
That's it. No CRM integration required for this setup. No data uploads. Everything runs locally.
Open Scout and navigate to Settings → ICP Configuration.
Traditional tools make you fill out rigid forms — pick an industry from a dropdown, select a company size range, choose a geography. Scout takes a different approach: you describe your ideal customer in natural language.
Here's an example ICP definition for a B2B developer tools company:
Target: Engineering leaders (VP Eng, CTO, Head of Platform) at Series A-C
startups with 20-200 employees. They're building developer-facing products
and struggling with CI/CD pipeline reliability. Strongest fit when they've
recently scaled the engineering team (hired 5+ engineers in the last 6 months)
and are feeling the pain of manual deployment processes.
Anti-patterns: Enterprise companies with existing DevOps teams of 10+.
Agencies or consultancies. Companies that just raised a seed round
(too early for our price point).
Notice what's happening here. You're not picking from dropdowns — you're describing the reasoning behind your ideal customer. Why does company size matter? Because of team scaling pain. Why Series A-C? Because of budget and timing.
Claude uses this context to make judgment calls that rule-based scoring can't. A 15-person startup that just tripled their engineering team might score higher than a 200-person company with a stable team — because your ICP explained why team growth matters.
Tips for a strong ICP definition:
Navigate to Settings → Signal Sources.
Scout monitors the web for buying signals — events that suggest a company might need your product right now. You choose which sources to monitor and what to look for.
Default signal sources (enabled out of the box):
| Source | What it catches |
|---|---|
| Job postings, executive changes, company updates, employee growth | |
| X (Twitter) | Founder/exec posts about pain points, hiring announcements, tech stack discussions |
| Threads asking for tool recommendations, complaints about competitors, industry discussions | |
| Job boards | Roles that indicate team scaling or technology adoption |
| News/PR | Funding announcements, product launches, partnerships, pivots |
For each source, you can add custom signal keywords that are specific to your business. For our developer tools example:
Signal keywords: "CI/CD problems", "deployment pipeline", "DevOps hiring",
"scaling engineering team", "moved to microservices", "breaking builds",
"release process", "need better tooling"
These keywords help Scout prioritize signals that are relevant to your specific product. A generic "Series A funding" signal is useful, but a founder tweeting "our deployment pipeline is broken and we just hired 3 engineers" is gold.
Signal strength weighting:
Scout assigns a default weight to each signal type, but you can adjust these based on what converts best for your business:
| Signal type | Default weight | Adjust if... |
|---|---|---|
| Direct pain mention | High | Almost always a strong signal |
| Job posting (relevant role) | High | Lower if your ICP doesn't correlate with hiring |
| Funding announcement | Medium | Higher for tools that growing companies buy first |
| Competitor complaint | Medium | Higher if you have strong competitive positioning |
| Technology discussion | Low | Higher if your product solves a specific tech problem |
| General industry mention | Low | Rarely worth increasing |
Navigate to Settings → Scoring.
Scout uses Claude to score each lead on a 0-100 scale based on your ICP definition and the signals detected. You set thresholds to determine what shows up in your review queue:
For your first batch, keep the defaults. You can adjust after you see how the initial scoring feels.
Why AI scoring beats point-based scoring:
Traditional lead scoring: "Company has 50-200 employees (+10 points), is in software industry (+5 points), VP title (+8 points) = 23 points."
AI lead scoring: "This company fits the ICP well. They're a 120-person SaaS startup that just posted for 3 backend engineers and a DevOps lead — clear sign they're scaling infrastructure. The CTO tweeted last week about 'deployment nightmares.' They raised Series B six months ago, so budget is likely available. However, they're in fintech, which means longer procurement cycles. Score: 82."
The AI explains its reasoning. You can read why a lead scored high or low, not just see a number. This makes the review queue dramatically faster — you spend seconds validating AI reasoning instead of minutes re-researching each lead.
Navigate to Dashboard → Start Scan or click Run Scout in the toolbar.
Scout will:
Your first scan typically takes 3-5 minutes depending on how many signal sources are active and how broad your keywords are.
What to expect from the first batch:
Don't worry if the first batch isn't perfect. The scoring improves as you provide feedback (next step).
Navigate to Dashboard → Review Queue.
This is where the human-in-the-loop approach pays off. For each queued lead, you'll see:
For each lead, you have three options:
Approve — The message looks good, the lead fits. Scout sends the outreach (or stages it for your preferred send time).
Edit — The lead is good but the message needs tweaking. Edit the draft and approve.
Skip — Not a fit. Scout logs your skip and the reason, then uses this feedback to improve future scoring.
Calibration tips for your first review:
After your first scan and review, Scout enters its normal operating rhythm:
The 20-minute setup gets you a working system. The system gets smarter every day you use it.
Once configured, the daily lead scoring workflow takes about 15 minutes:
Compare this to the traditional approach: 2-4 hours of manual research, list building, and email writing — for the same number of quality touches.
Once you're comfortable with the basics, Scout supports multiple ICP definitions. This is useful when you sell to different segments:
Each ICP gets its own scoring model, signal keywords, and message templates. Leads are scored against all active ICPs and queued with the best-fit profile.
1. Making your ICP too broad. "B2B SaaS companies" will score everything. Be specific about pain points, timing, and anti-patterns.
2. Setting thresholds too high initially. Start with the default 70 and lower it if your queue is too empty. It's better to review and skip a few extra leads than to miss good ones.
3. Not providing skip feedback. When you skip a lead, the reason matters. "Not a fit" teaches Scout nothing. "Too early stage — pre-revenue" teaches it a lot.
4. Ignoring the watch list. Some of your best leads will start at 50-60 and climb to 80+ as Scout detects additional signals over the following weeks. Check the watch list weekly.
5. Over-engineering signal keywords. Start with 5-10 keywords that directly describe your customer's pain. You can add more after you see what signals are actually converting.
Everything described in this tutorial happens on your machine. Your ICP definition, your signal keywords, your lead database, your scoring feedback, your outreach messages — all stored locally.
Scout makes API calls to Claude for reasoning tasks, but your data is never stored on external servers. No aggregation with other customers' data. No training on your sales intelligence.
When you close Scout, your data is right where you left it: on your hard drive.
Ready to set up AI-powered lead scoring? Download Scout free and follow the steps above. You'll have a working scoring model in 20 minutes.