Most target lists are built from a strategy document written before the company had customers. The list describes who the founders hoped would buy. The closed-won list describes who actually did, and the two are rarely the same.
Lookalike targeting is the practice of building your outbound list from the second one.
Start with the data you already have
Export your last 30 to 50 closed-won accounts. Not your biggest logos, not the ones in the case studies, all of them, including the unglamorous ones. Then answer these for each:
- Industry and, more usefully, business model, do they sell to businesses or consumers, subscription or project-based, high volume or high value?
- Headcount, and headcount at the time they bought rather than today
- Growth stage and whether they were growing fast at the time
- Tools they were already running
- Who signed, and who first replied
- What was happening at the account in the 90 days before the first meeting
This exercise routinely contradicts the stated ICP. Teams targeting enterprise discover their best customers are 40-person companies. Teams targeting one industry discover the pattern is actually business model, and it spans four industries.
Which attributes actually predict fit
Not all filters are equally useful, and the easiest ones to apply are often the weakest.
| Attribute | Predictive strength | Why |
|---|---|---|
| Technology stack | High | Proves they buy software for this problem area and can integrate it |
| Business model | High | Determines whether your value proposition applies at all |
| Growth stage / trajectory | High | Growth is what breaks the processes most B2B tools fix |
| Headcount band | Medium | Useful as a range, misleading as a precise cut-off |
| Industry | Medium | Often a proxy for business model. Use the underlying thing instead where you can |
| Revenue | Low–medium | Frequently estimated rather than known, and lags reality |
| Geography | Filter, not a predictor | Constrains what you can serve; does not indicate fit |
The pattern here is that behavioural attributes, what a company does and buys, predict better than descriptive ones. Industry and headcount are easy to filter on, which is why most lists are built from them, and why most lists are mediocre.
Test the attributes before trusting them
An attribute shared by all your closed-won accounts is only meaningful if it is not shared by everyone you lost.
Run the same analysis on 30 closed-lost accounts. Anything appearing at similar rates in both lists is a description of your pipeline, not a predictor of fit. What you want are the attributes where won and lost diverge sharply.
This step takes an hour and eliminates most of the assumptions people build lists on.
Building the list
- Translate the surviving attributes into filters. In Lead Finder that means combining technographic, headcount, growth and industry filters into one saved search across 500M+ records.
- Layer in timing. A lookalike showing a buying signal, hiring the relevant role, recent funding, fast growth, goes to the front of the queue.
- Score for fit. ICP Score ranks results against your profile, so a broad filter still produces a ranked list rather than an undifferentiated dump.
- Find the right person, not just the company. Lookalike work identifies accounts. You still need the individual who owns the problem, which is a separate filter on title and seniority.
- Cut to working capacity. Take the top tier your team can genuinely work this quarter and stop there.
Three ways this goes wrong
Copying the logo, not the pattern. One large customer becomes the template and the team spends a quarter chasing enterprises that behave nothing like the account that actually closed. Look for repeated patterns across many accounts, not the most impressive single one.
Filtering so tightly nothing is left. Six attributes stacked at once returns forty companies. Rank on the strong attributes rather than excluding on all of them, a scored list beats a hard filter.
Building it once. Your ICP moves as the product and the market change. Redo this every two quarters using the most recent closed-won cohort, not the one from launch.
What good looks like
A useful lookalike list is smaller than people expect and more specific than a strategy deck. Two or three hundred accounts, matched on attributes you have tested against closed-lost, ranked by fit score, with the ones showing timing signals at the top.
Worked properly, that list outperforms several thousand accounts matched on industry and headcount alone, and it costs less to run, because you are not paying in sends and deliverability for the accounts that were never going to buy.
Next: how to define an ICP from scratch, or scoring leads before they enter a sequence.
Frequently asked questions
What are company lookalikes?
Companies that resemble your existing best customers across attributes that predict buying, such as industry, size, growth stage, technology stack and business model. The idea is to stop guessing at your target market and derive it from accounts that already bought.
How do I find lookalike companies?
Start from your closed-won list, identify the attributes those accounts genuinely share, then use those attributes as filters in a company database. The hard part is picking the right attributes, not running the search.
Which attributes actually predict fit?
Usually technology stack, business model and growth stage predict better than industry and headcount, though most teams start with the latter two because they are easier to filter on. Test which ones separate your closed-won accounts from your closed-lost ones.
How many lookalike accounts should I target?
Enough to fill pipeline for a quarter at your realistic working capacity, not as many as the filter returns. A tight list of 300 well-matched accounts worked properly beats 5,000 loosely matched ones sent a generic sequence.
Do lookalikes work for a new company with no customers?
Partly. With no closed-won data, use your best-fit pipeline or the customers of a comparable competitor as the seed. Then revisit the model after your first 20 deals, because your real ICP is almost never the one you assumed at the start.