> ## Documentation Index > Fetch the complete documentation index at: https://www.landbase.com/docs/llms.txt > Use this file to discover all available pages before exploring further. # Expand Lookalikes > Provide seed companies and get a scored lookalike audience at scale — up to 20,000 results from a single query. # Expand Lookalikes You know your best customers. You want more accounts like them. The question is what "like them" actually means — and whether your tools can capture it. Most lookalike features match on firmographic attributes: same industry, same size range, same geography. That catches the obvious overlap but misses the real patterns. Your best customers might share a go-to-market motion, a technology architecture, a business model, or a market positioning that doesn't map to any dropdown filter. Expand Lookalikes finds companies that are similar across what they do and how they describe themselves — not just what category they're filed under. ## How it works You provide a set of seed companies — typically your best customers or closed-won accounts. The system finds what those companies have in common at a deeper level than firmographics, then surfaces thousands of similar companies you haven't found yet, ranked by how closely they match. * Give it 10 seed companies, get back up to 20,000 scored lookalikes * Each result is ranked by a confidence score so you know which matches are strongest * Results combine with your other filters (size, geography, funding) so you're not just getting lookalikes — you're getting *qualified* lookalikes **What's happening under the hood** Seed companies are matched against a pre-computed similarity graph where every company has similarity scores to thousands of others, computed from description embeddings, keywords, and firmographic signals. If you provide a large number of seeds, clustering on the embedding vectors selects representative companies that span the diversity of your seed set. Scores are aggregated using Root Sum Square — a company moderately similar to 8 of your 10 seeds scores higher than one very similar to just 1 seed. This rewards broad ICP fit over single-account similarity. ## Similarity goes deeper than filters Matching on industry + size + geography is a proxy for similarity. It catches companies in the same category but misses the ones that *do the same thing* in a different category. A developer tools company and a data infrastructure company might both be your ICP — same buyer, same budget, same pain point — but they're in different industry classifications. Expand Lookalikes matches on what companies say about themselves (descriptions, keywords) and how they're structured, not just which checkbox they fall under. ## Scoring rewards breadth, not just closeness A company that's moderately similar to 8 of your 10 seed accounts is almost certainly a better fit than one that's very similar to just 1 seed but unlike the rest. The scoring formula captures this — it rewards companies that match the *pattern* of your ICP, not just one data point. **Why Root Sum Square?** RSS scoring computes `sqrt(score_1² + score_2² + ... + score_n²)` across all seeds. Squaring emphasizes stronger individual similarities while the summation rewards consistency across seeds. The square root normalizes the final score. A company similar to many seeds accumulates a high aggregate even if no individual score is extremely high. ## How it connects to Scout Lookalikes [Scout Lookalikes](https://www.landbase.com/docs/overview/capabilities/scout-lookalikes.md) and Expand Lookalikes are the same concept at different scales: * **Scout** — fast, small (25 results), pre-computed. For exploration and validation. * **Expand** — slower, massive (20K results), real-time computation. For production audience building. A typical workflow starts with Scout to validate the pattern, then moves to Expand to build the full audience. ## Result limits Up to 20,000 scored results per query. This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.