Look-Alike Account Selection

Look-alike account selection finds companies that resemble your best existing customers across firmographic, technographic, and behavioral dimensions. Instead of defining the ICP from theory and hoping it works, look-alike modeling starts with customers who already buy, retain, and expand. Then it finds more companies that share the same profile.

The principle: your best customers are the best predictor of your next customers. A company that matches your top customer on size, stage, industry, tech stack, and GTM motion is more likely to buy than a company you've never sold to before. Look-alike selection makes this intuition systematic and scalable.

The Look-Alike Process

Step 1: Define your seed list (your best customers)

Pull 10-30 of your best customers based on one or more success criteria.

Seed list rules:

• Minimum 10 accounts. Below 10, the sample is too small to find reliable patterns. The look-alike model will overfit to coincidental similarities
• Weight toward retention metrics over acquisition metrics. Fast close is nice. But a customer who churns in 6 months isn't a "best" customer. Use NRR or retention as the primary ranking
• Exclude outliers. If one customer represents 30% of your revenue but is 10x the size of every other customer, they'll skew the profile. Include them in the data but don't let them dominate
• Use recent customers (last 18-24 months). Customers from 5 years ago were acquired when your product and market were different. Focus on customers that represent your current product-market fit

Step 2: Profile the seed list

For each seed account, capture the firmographic and technographic attributes.

Step 3: Find the patterns

Analyze the seed list to find the profile that defines "companies like our best customers."

Pattern analysis:

Across 20 seed accounts:
- Employee count: 70% are 50-300 employees (median: 140)
- Industry: 65% are B2B SaaS, 20% are fintech, 15% other
- Funding: 80% are Series A or Series B
- Geography: 90% are US-based
- CRM: 75% use HubSpot, 25% use Salesforce
- GTM: 85% are sales-led
- Growth: average 40% YoY headcount growth

The look-alike profile: "B2B SaaS or fintech, 50-300 employees, Series A-B, US, sales-led motion, using HubSpot or Salesforce, growing 20%+ YoY."

Step 4: Search for matches

Use the profile to search for net-new accounts.

Step 5: Score and prioritize matches

Run every matched account through your ICP scoring model (per icp-account-modeling skill). Rank by score.

Step 4 output: 500 matched accounts
  ↓
ICP scoring: score each on all dimensions
  ↓
Filter: ICP score ≥ 60 (Tier 1-2 only)
  ↓
Remove: existing customers, active deals, competitors
  ↓
Output: 200-300 net-new Tier 1-2 accounts
  ↓
Layer signals: which of these have a recent signal (funding, hiring, leadership change)?
  ↓
Signal-matched: 50-100 accounts with both look-alike fit AND a signal
  ↓
This is your target list

Step 6: Add signal layer

A look-alike without a signal is a well-matched account with no urgency. A look-alike with a signal is a high-probability prospect.

Look-Alike by Data Source

From CRM closed-won data

1. Export closed-won accounts with enrichment data
2. Profile on size, industry, stage, tech stack
3. Search for matches in Sales Nav or Apollo
4. Score matches against the profile
5. Remove existing CRM contacts

Best for: Most common approach. Works when you have 10+ closed-won deals.

From closed-lost data (re-engagement)

1. Export closed-lost deals from the last 12 months
2. Filter: loss reason was "timing" or "budget" (not "bad fit")
3. Check for new signals (funding, leadership change, tech shift)
4. Score against current ICP model
5. Re-engage with a new angle referencing the prior relationship

Best for: Highest-quality look-alikes because they already evaluated your product. Changed circumstances may make them ready now.

From competitor customer data

1. Find competitor customers from: G2 reviews, case studies,
   integration directories, LinkedIn
2. Profile: which competitor customers match your ICP?
3. Filter for switching signals: negative G2 reviews, contract
   renewal timing (if known), leadership changes
4. Outreach with competitive displacement angle

Best for: Accounts already in the category. They understand the problem and are using a solution. You just need to be a better one.

From product usage data (expansion look-alikes)

1. Profile your top customers by product usage patterns
2. Find similar-usage customers in your base
3. These are expansion look-alikes: similar usage → similar
   expansion potential
4. Target for upsell or cross-sell

Best for: Expansion pipeline. Finding existing customers who resemble your best expanders.

Validating the Look-Alike Model

How to know if your look-alikes are accurate

Validation rules

• Run the comparison on at least 50 look-alike deals before drawing conclusions. Fewer than 50 and the results are noise
• If look-alikes don't outperform random ICP lists, the profile is wrong. Either the seed list wasn't representative, or the matching criteria are too loose. Tighten the profile
• Re-validate quarterly. As your customer base evolves, the look-alike profile should evolve with it. A profile from 12 months ago may not predict today's best accounts

Tools for Look-Alike Selection

Tool selection rules

• Start with a spreadsheet. Export your seed account data. Profile manually. Search manually in Sales Nav. This builds intuition about what makes your best customers similar
• Graduate to Apollo or Sales Nav saved searches at scale. Once the profile is defined, codify it as a saved search. Run weekly for net-new matches
• Predictive tools (MadKudu, 6sense) at $10M+ ARR. Below that, the data volume doesn't justify the cost. Manual modeling is sufficient and builds more understanding

Measurement

Anti-Pattern Check

• Seed list is too small (3-5 accounts). The profile is based on coincidences, not patterns. One outlier dominates. Use at least 10 accounts, ideally 20-30
• Seed list includes churned customers. Churned customers are anti-models, not look-alike seeds. Only include customers who are active, retained, and growing. The seed defines "who we want more of"
• Look-alike profile based on aspiration, not data. "We want to sell to Fortune 500" when all 20 seed accounts are Series A startups. The profile must reflect actual customer data, not ambition
• No signal layer on look-alike matches. 300 accounts that match the profile but have no urgency signal. They're all "someday" prospects. Layer a signal (funding, hiring, leadership change) to identify "right now" prospects
• Never refreshing the seed list. Your customer base 12 months ago was different from today. New verticals adopted. Larger companies started buying. The seed list should be re-pulled and the profile re-calculated quarterly
• Look-alikes used for ABM without ICP scoring. A company that "looks similar" to your best customer may still fail on specific ICP dimensions (wrong geography, missing tech stack). Score every look-alike through the ICP model before adding to a target list
• Same look-alike list for 6 months. Lists decay. New companies emerge. Signals expire. Refresh the look-alike search monthly for new matches and quarterly for a full profile update