Lookalike Benchmark

Why an LLM judge instead of a hand-labeled set

A fully hand-labeled lookalike set would require labeling K × seeds × vendors candidates (100 × 48 × 7 = 33600judgements) every time we re-run a snapshot. That doesn't scale, and it isn't how the buyer actually evaluates a vendor in the wild - the buyer reads the list and decides "close enough to my ICP, yes or no".

The judge approximates that decision with a consistent rubric: given the seed's name, domain, and description, is this returned candidate plausibly the same kind of company a B2B seller would target as a lookalike? The judge's rationale is persisted alongside the binary label so any cell can be audited by a human in seconds. When the model swaps, the cohort re-runs with the same prompt; deltas are visible.

How Exa and Parallel are prompt-tuned

Exa and Parallel accept natural-language company queries. For each seed, we run the same three fixed prompt variants, judge each returned list, and retain the variant with the highest Precision@100 for that provider-seed cell. Ties break on more judged candidates, then lower latency. The full attempts and winning config are preserved in the raw audit trail.

1. Full framing · Find companies that are lookalikes of {Company}: companies with a similar core product and buyer. {Company description} Return companies a buyer would realistically evaluate alongside {Company}.
2. Product-and-buyer framing · Find companies that are lookalikes of {Company}: companies with a similar core product and buyer. {Company description}
3. Buyer-evaluation framing · Find companies that are lookalikes of {Company}: {Company description}Return companies a buyer would realistically evaluate alongside{Company}.

This is a per-seed best-of-prompt result, not a single fixed production prompt average. Domain-first providers are not given this prompt sweep.

How each vendor was queried

• Ocean.io · /companies/lookalikes with seed domain. Default similarity model, K = 100.
• Exa · /search with category: company and query text like Find companies that are lookalikes of HubSpot, using the seed description, K = 100.
• Parallel · Responses API research prompt with the seed name, description, and an explicit similar-product-and-buyer instruction, K = 100.
• PredictLeads · /api/v3/companies/{domain}/similar_companies; ranks via shared tech, news, and jobs co-signals; evaluated through Precision@25.
• Extruct · /v1/companies/{domain}/similar, K = 100.
• CUFinder · /v2/fcl with the seed domain; the endpoint returns at most 10, so it is evaluated through Precision@10.
• Discolike · /v1/discover with the seed domain and max_records=100.

Verify any number end-to-end

The full benchmark — runner code, judge prompt, benchmark snapshot, and per-cell raw audit trail (the literal HTTP request/response we sent each vendor + the literal LLM judge prompt/response per candidate) — is mirrored in a public repo: openbenchmarks-labs/lookalikes. Auth headers are scrubbed via an allow-list; everything else is verbatim.

To audit a single cell, open data/lookalike-runs/<dataset>/<seed>/<vendor>.raw.json in that repo and replay any of the vendor_calls[] with your own credentials, or re-score with your own LLM by replaying judge_calls[].messages against any OpenAI-v1 compatible model — useful for measuring judge bias or drift across model versions.

Inclusion queue and how to request a provider

Live: Ocean.io, Exa, Parallel, PredictLeads, Extruct, CUFinder, and Discolike.

Requested but not directly comparable: ZoomInfo (company lookalikes are sales-gated, no self-serve API), Clay (lookalike runs inside Clay tables), Apollo (no public lookalike endpoint), Lusha (`/v3/companies/lookalike` requires 5-100 seeds per request, incompatible with the per-seed cell unit of this benchmark), Coresignal, CompanyEnrich, Explorium, and Surfe (enrichment or prospecting surfaces whose lookalike quality is not an independently measured Precision@K number here).

To request a provider, email founders@openbenchmarks.com with a link to the public API docs and pricing page.