Skip to main content

​
Niche Signal Discovery

Every team knows the obvious ICP signals — company size, industry, funding stage. The signals that actually separate your wins from your losses are usually more specific: a particular tech stack combination, a hiring pattern, a revenue growth rate, a specific competitor on the account. This play finds those signals by mining your actual deal history.

​
The Play

1

Export wins and losses from CRM

Pull closed-won and closed-lost deals from HubSpot or Salesforce for the last 12 months. Include deal size, sales cycle, competitor mentioned, close reason, and any custom fields your team tracks.
2

Enrich with deep firmographic data

For each company, pull detailed firmographics: employee count, department breakdown, tech stack, hiring velocity by role, funding history, revenue estimates, geographic presence, recent news.
3

Run comparative analysis

Ask Claude to compare enriched wins vs losses across every dimension. Look for attributes where the distribution is meaningfully different between wins and losses.
4

Identify niche signals

The goal is non-obvious patterns. Examples from real analyses:
  • “We win 4x more often at companies with 3-8 open engineering roles”
  • “Deals close 2x faster when the company uses both Salesforce and Outreach”
  • “We lose 80% of deals where the primary competitor is [X]”
  • “Companies that raised Series B in the last 6 months close at 3x the rate of Series A”
5

Validate with holdout data

Test the discovered signals against held-out deals. If a signal predicts wins/losses on data the model hasn’t seen, it’s real. If it only works on training data, it’s noise.
6

Feed into account scoring

Add validated niche signals to your account scoring model. These signals become the weighted dimensions that separate generic ICP from your specific propensity model.

​
How to Run It

The /niche-signal-discovery skill automates this entire play. Give it your won and lost domain lists and it handles enrichment, analysis, and reporting. 
/niche-signal-discovery
Claude Code will ask for your won and lost domain lists (minimum 20 won + 10 lost), then run the full pipeline:
  1. Discovers your vertical - researches what you sell, who you sell to, and your competitive landscape
  2. Generates vertical-specific configs - keywords, tech stack tools, and job roles tailored to your buyer persona
  3. Enriches every company - pulls multi-page website content (~5 pages per domain) and job listings via Deepline CLI
  4. Runs differential analysis - computes Laplace-smoothed lift scores across every dimension to find what separates wins from losses
  5. Generates a full report with signal strength dashboards, scoring models, Apollo search recipes, and buyer persona quick-reference cards

​
What You Get

  • Signal strength dashboard with lift scores and visual bars (e.g., “Hiring 3+ SDRs simultaneously” at 4.2x lift)
  • Anti-fit signals that flag accounts to skip (e.g., “Consumer checkout language” at 0.2x lift)
  • Lead scoring model - 0-100 point system with three tiers (Core Fit, Buying Intent, Infrastructure Readiness)
  • Pre-built Apollo URLs for people and company searches based on discovered signals
  • Source evidence - exact quotes with page URLs and job listings backing every signal
  • Scoring cheatsheet with “How to Check” column for each signal

​
Cost

The skill uses Serper for page discovery and Firecrawl for content extraction - the cheapest enrichment path available.
StepToolCredits per companyTotal (60 companies)
Discover relevant pagesSerper Google Search0.021.20
Scrape ~5 pages per domainFirecrawl Scrape0.053.00
Job listingsCrustdata Job Listings0.4024.00
Total~0.47~28
That’s about $0.05 per company on managed credits, or $3 for 60 companies. BYOK users pay $0 on Deepline. The skill asks for credit approval before running any paid enrichment.
If you have an Apollo API key connected, job postings are BYOK (free on Deepline) - bringing the total down to ~0.07 credits per company.
If your Closed Won list is small (under 15), the skill can supplement with lookalike companies. It flags this in the report so you know the limitation.

​
What Niche Signals Look Like

Generic ICP attributes (company size, industry) get you to the right ballpark. Niche signals get you to the right accounts.
Generic signalNiche signal
”SaaS companies""SaaS companies using Salesforce + Outreach with 200-500 employees"
"Series A-C funding""Series B in the last 6 months with 20%+ headcount growth"
"Hiring GTM roles""3+ SDR/BDR roles open simultaneously"
"US-based""US-based with at least one office outside the Bay Area"
"Enterprise""$20M-$80M ARR with engineering team > 30% of headcount”
Run this quarterly. Your niche signals shift as your product evolves and your market changes. Signals that predicted wins 6 months ago may not predict wins today.
Account Scoring → | Closed-Lost Recovery → | I Have X, I Want Y →