Introducing Parallel FindAll

## Turn the web into your own structured dataset

**FindAll** finds any set of entities (companies, people, events, locations, houses, etc.) based on a set of match criteria. For example, with **FindAll, **you can run a natural language query like “Find all dental practices located in Ohio that have 4+ star Google reviews.”

This is a powerful way to discover the complete long tail of interesting entities from the web and filter them down with match criteria that are personalized to your unique use case. The result is an extensible tool that can produce high-quality datasets on demand, as opposed to buying static, stale, and generic datasets.

## How FindAll works

FindAll executes a three-stage pipeline optimized for both coverage and efficiency:

**1. Generate candidates from web data: FindAll** searches across our proprietary web index to identify potential entities matching your query. Unlike traditional search, which returns a fixed result set, **FindAll** generates candidates dynamically based on your specific criteria.

**2. Evaluate against match conditions: **Each candidate is evaluated against your match conditions using multi-hop reasoning across web sources. Only candidates which satisfy all conditions reach matched status and are included in the results. This staged approach means you only pay to process entities that actually matter.

**3. Extract Structured Enrichments: **For matched entities, **FindAll** automatically orchestrates our **Task API**[**Task API**](https://docs.parallel.ai/task-api/task-quickstart) to extract any additional fields you've specified— from basic attributes like revenue and employee count to complex data points like the strategic initiatives a company is prioritizing.

## State-of-the-art performance

To test the performance of **FindAll**, we created our own benchmark of 40 complex multi-criteria queries covering public companies, startups, SMBs, specialized entities, and people (e.g., executives, researchers, and professionals). Recall measures the proportion of all correct matches within the entire competitive set of successfully identified entities.

Some sample questions:

• - "Find all former McKinsey & Company consultants who are currently employed in C-level or VP positions at healthcare technology startups with Series A or later funding" — combines employment history, current role level, industry focus, and funding stage.
• - "Find all wedding venues in Florida with capacity between 150-300 guests that offer both indoor and outdoor ceremony options, provide in-house catering, and have availability in 2025" — combines location, capacity range, facility features, service offerings, and temporal availability.
• - "Find all climate technology startups that have active pilot programs with Fortune 500 companies, raised pre-Series A funding, and focus on carbon capture or renewable energy storage" — combines industry focus, corporate partnerships, funding stage, and specific technology areas.

****

**FindAll Pro** achieves state-of-the-art results with 61% recall, ~3X higher than OpenAI Deep Research, Anthropic Deep Research, and Exa. Higher recall means that Parallel **FindAll **finds more correct matches for a given query.** FindAll** **Base** also achieves 30% recall while being the lowest cost on the market, making it the most cost-effective yet performant option.

## About Parallel Web Systems

Parallel develops critical web search infrastructure for AI. Our suite of web search and agent APIs is built on a rapidly growing proprietary index of the global internet. These solutions transform human tasks that previously took days and weeks into agentic tasks that now take seconds and minutes.

Fortune 100 and 500 companies use Parallel’s web intelligence APIs in insurance, finance, and retail, as well as AI-first businesses like Clay, Starbridge, and Sourcegraph.

## Related Posts64