OWNED DISCOVERY

The company list starts with our crawl.

You bought a list, watched it age, and still had to rebuild the target account universe yourself.

Workloom does not start with a rented company list. It crawls public sources, finds companies through search, aggregators, waterfalls and maps, then turns conflicting evidence into one canonical record. Your ICP comes from deals you won, not a form. That record can feed contact data, signals and the rest of the platform without a broker handoff.

Crawlers

35 worker types built in-house

Pipeline

three Kafka stages ordered handoffs

four search surfaces plus Maps

Browsers

two browser engines distinct modes

Reconciliation

source conflict checks scored records

ICP

won-deal ICPs learned from reality

WHY LISTS DECAY

A bought list stops at the broker

Bought lists decay because they stop at someone else's record. You inherit its coverage, refresh cycle and definition of a company. Workloom starts from the company itself. It crawls and enriches sources, then builds an ICP from your own won deals instead of asking a form to describe your market.

The result is a discovery layer tied to the same machinery that handles contact data and signals. No imported list has to survive a separate handoff before your outbound system can use it.

Fig 1 · workers, writers and normalizers form one canonical record

SEARCH4 engines

CRAWL35 workers

WRITE14 writers

RECONCILEconfidence

NORMALIZE13 normalizers

CANONICALone record

HOW THE CRAWL WORKS

Workers find it. Writers keep it. Normalizers make it one.

The pipeline runs over Apache Kafka in three separate stages. First, 35 worker types scrape. Then 14 writers persist what they found. Then 13 normalizers canonicalize those findings into one company record. Writers and normalizers are downstream stages, not extra names for the workers.

Discovery runs across Google, Brave, DuckDuckGo and Google Maps, alongside aggregator and waterfall discovery. Two Playwright engines do the browsing: a light headless engine blocks 20 analytics domains and moves the cursor along Bezier paths; a stealth headful engine uses persistent profiles, context pooling, and WebGL and canvas fingerprint spoofing. When our own engines get blocked, Serper and BrightData Web Unlocker pick up the work.

WHEN A SOURCE DISAGREES

Conflicting evidence stays visible

Sources disagree. Workloom does not quietly pick the first answer. Conflict detection flags disagreement, and confidence scoring weighs the evidence before sources are reconciled into the canonical company record.

That record carries the reasoning forward into the platform, rather than forcing contact data, signals and account discovery to make separate guesses. The same process also supports ICPs generated from your own won deals.

Fig 2 · light headless and stealth headful engines

Headless engineBlocks 20 analytics domains, Bezier cursor paths, stealth plugin

Headful enginePersistent profiles, context pooling, WebGL and canvas spoofing

Serper fallbackWhen a search engine blocks the crawl

BrightData unlockerWhen a target blocks everything else

Questions