Research

We ran Riveter against OpenBenchmarks’ Lookalike benchmark and scored 79.2% — the highest Precision@100

Cody Watters· CTO, RiveterPublished · Updated

Riveter scored 79.2% Precision@100 on the OpenBenchmarks Lookalike benchmark: 100 companies similar to each of 48 seed companies, every one scored by a three-model judge panel. The top published arm, CompanyEnrich, scored 73.5%. At the top of the list Riveter scored 97.5% Precision@10 against CompanyEnrich’s 91.9%.

Bar chart of Precision at 100 by provider on the OpenBenchmarks Lookalike benchmark: Riveter 79.2%, CompanyEnrich 73.5%, Seltz 63.4%, Extruct 61.1%, Ocean.io 59.6%, Parallel 56.9%, Exa 48.6%, Discolike 34.6%
Bar chart of Precision at 10 by provider on the OpenBenchmarks Lookalike benchmark: Riveter 97.5%, CompanyEnrich 91.9%, PredictLeads 89.8%, Seltz 84.4%, Exa 80.8%, Extruct 75.0%, Ocean.io 74.3%, Parallel 69.2%, ZoomInfo 67.1%, Discolike 41.9%, CUFinder 38.9%
Precision@10, all eleven arms on the board plus our Riveter run.

Benchmark: openbenchmarks.com/lookalikes · Code: openbenchmarks-labs/lookalikes · Riveter arm: PR #14 · Our run: results CSV, 4,440 rows

What the benchmark measures

OpenBenchmarks publishes open-source benchmarks of data providers. Lookalikes gives every arm the same query for each of 48 seed companies across 12 categories. Toast’s reads: “Find companies similar to Toast (toasttab.com): restaurant point-of-sale platforms that also offer any of payments, online ordering, payroll, or inventory management.”

Three LLM judges from three labs score every company returned. It must be the same kind of company selling to the same kind of buyer as the seed, and share at least one of the seed’s listed capabilities. A majority makes it relevant.

Precision@100 is relevant companies among the first 100 returned, divided by 100 even when a provider returns fewer. Precision@10 is the same for the first 10. Three arms return at most 25 companies, so they appear on the Precision@10 board only.

How we ran it

Riveter has no lookalike index, no similar-companies endpoint, and no company graph. That is its strength here. Riveter’s dataset builder researches the live web at request time, so it builds a list of anything the web describes: companies, but just as readily products on a retailer’s site, restaurants in a city, authors, properties for sale, nonprofits. A list of restaurant POS vendors is one prompt among any other, and the list is as current as the web is.

For each seed we created a dataset holding the seed as its first row, asked the builder for 100 more rows with the seed’s benchmark query as the prompt, and extended the dataset when it stopped short, up to three builds. The rows, in the order they were generated, were the ranked list. No custom tools, no tuning, no filtering.

We ran it on October 20 and 21, 2026 and judged every candidate with the benchmark’s own judge panel, unchanged.

The result

Riveter CompanyEnrich (top published arm)
Precision@100 79.2% 73.5%
Precision@10 97.5% 91.9%
Relevant companies in the top 100, 48 seeds 3,803 3,526
Seeds judged 48 of 48 48 of 48

Riveter scored higher than every published arm with per-seed data on 24 of the 48 seeds, 90% or higher on 17, and 100% Precision@10 on 42.

Data breakdown

Precision@100 per seed for Riveter and the five published arms whose per-seed cells were in the benchmark’s snapshot at the time of our run; CompanyEnrich and Seltz publish board averages only. The best score in each row is bold, “—” marks a seed the arm was not judged on, and “Rows” is how many companies Riveter returned of the 100 asked for.

Riveter returned all 100 on 18 seeds and stopped short on 30. The benchmark divides by 100 regardless; among the companies Riveter did return, 85.7% were relevant. The lowest seeds were VTS at 25%, where the panel rejected residential and investor tools as a different kind of company, and Faire, Wise, and Mews at 42–45%, where the lists were short.

Every candidate for every seed, with the panel’s verdict, each judge’s vote, and the one-line reason: results CSV, 4,440 rows. Seeds, queries, and published cells © OpenBenchmarks, CC BY 4.0.

Show all 48 seeds
Seed Category Rows Riveter Extruct Ocean.io Parallel Exa Discolike
1Password cybersecurity 92 74 87 72 49 41 38
ABB industrial 100 96 10 48 3 51 27
Amplitude devtools 71 50 42 56 24 25 6
Braze b2b-saas 97 90 89 78 81 47 20
Brex fintech 100 75 58 84 80 72 79
Caterpillar industrial 99 88 33 34 10 41 6
CBRE real-estate 98 98 75 86 41 79 36
Cloudflare cybersecurity 95 80 57 33 27 35 21
CrowdStrike cybersecurity 100 76 78 74 91 41 47
Databricks devtools 89 80 40 63 24 33 11
Datadog devtools 63 61 81 68 93 69 33
Faire ecommerce 58 42 35 — 28 33 3
FedEx logistics 100 100 98 78 48 76 —
Fluence energy 100 80 88 81 92 45 38
Gorgias ecommerce 72 54 86 73 30 26 66
Grab logistics 99 63 71 37 35 47 33
Gusto b2b-saas 95 91 89 92 98 69 75
HubSpot b2b-saas 98 86 47 57 85 40 18
Intercom b2b-saas 93 84 90 80 90 45 70
Jobber home-services 100 87 90 77 95 76 14
Lightspeed hospitality 100 93 76 82 70 58 52
Marriott hospitality 100 91 70 75 2 41 6
Maven Clinic healthtech 75 63 65 73 8 37 35
Mews hospitality 45 45 91 91 97 38 76
Miro b2b-saas 99 81 62 20 53 33 20
Motive logistics 100 98 0 0 91 45 0
NextEra Energy energy 100 100 76 41 27 56 23
Notion b2b-saas 97 82 46 41 92 37 8
Nubank fintech 99 92 81 43 78 48 10
OpenAI devtools 100 80 35 7 42 26 3
Plaid fintech 81 80 0 0 81 40 0
Procore b2b-saas 94 79 79 90 83 62 67
project44 logistics 100 71 36 73 50 56 35
PTC industrial 100 81 26 12 55 39 15
Revolut fintech 100 77 95 84 4 48 59
Rippling b2b-saas 99 95 71 73 96 68 91
Roto-Rooter home-services 100 99 90 90 95 86 86
Sea (Shopee) ecommerce 99 92 20 5 5 34 15
ServiceTitan home-services 100 91 81 83 82 68 27
Shopify ecommerce 95 95 60 68 45 32 23
Siemens industrial 100 98 6 52 — 46 25
Snyk cybersecurity 99 71 80 86 52 43 46
Teladoc healthtech 98 83 92 74 79 55 48
Toast hospitality 100 95 92 89 45 47 81
Vanta cybersecurity 94 78 79 73 73 42 64
Veeva healthtech 99 71 34 31 54 33 27
VTS real-estate 94 25 47 72 13 33 42
Wise fintech 54 42 0 0 79 89 0

Build your own list

The same builder makes any list you can describe, companies or otherwise. Describe the rows you want, get them, then add the columns you need. Start free or book a demo.

Related reading: How to build a company list from criteria · Riveter vs Exa · Riveter vs Parallel


Run on October 20–21, 2026 against the 48 seeds of the OpenBenchmarks Lookalike benchmark (dataset lookalike-2026-q3; runner and judge code as in the repository’s commit ac0901e), judged with the benchmark’s published three-judge panel, unchanged. Board figures from openbenchmarks.com, last updated October 16, 2026. Runner code for the Riveter arm: pull request #14.

See Riveter in action

Build your first dataset in minutes.