In 2020, a police department interested in artificial intelligence encountered a problem no algorithm could fix: money. The Los Angeles Police Department had trialed Voyager Labs’ software. Then, according to the released procurement records, projected pandemic budget shortfalls prevented it from buying VoyagerAnalytics licenses. The machine promised to shorten investigations. The purchasing process had its own clock.
- Voyager turns unstructured data into leads about people, topics and networks.
- Its customers are investigative agencies and corporate teams; access is sold through software licenses.
- The public record raises questions about data access, inference and the distance between a lead and evidence.
That episode supplies an unusually ordinary entrance to an unusually consequential business. Voyager sells tools for finding patterns in the online material people leave behind. A post, a photograph and a relationship can become pieces of an investigation. The attraction is obvious: give a small team the capacity to examine material that would otherwise overwhelm it. The complication begins when the software supplies an interpretation.
A company built around the overlooked connection
Avi Korenblum founded Voyager in 2012 after two decades in Israel’s intelligence community, Globes reported. In November 2016, it emerged from stealth with a reported $100 million in funding. Backers included Horizons Ventures, Sir Ronald Cohen, Lloyd Dorfman and OCAPAC Holding Company. Early ambitions included financial and ecommerce applications. The underlying wager was that untidy human information could become useful machine-readable intelligence.
Today’s company description emphasizes government, law enforcement and commercial investigators. Its expertise spans machine learning, computer vision and natural language understanding. These disciplines address different parts of the same mess: words without tidy columns, pictures without useful labels, and connections dispersed across records. Voyager combines them around investigative tasks rather than asking every analyst to assemble a toolkit.
The geography has also shifted. Voyager announced a move to the United States in 2022, with Courtney Bromley taking responsibility for global operations. Its current website describes it as UK-headquartered. There is a small lesson here for anyone researching private companies: an old headquarters entry can outlive the corporate arrangement it once described.
Four verbs on the investigator’s desk
The current platform is organized around Check, Analyze, Vision and Verify. Check applies predefined risk criteria. Analyze examines people, topics and networks. Vision works with images, objects and text inside visual material. Verify examines identity claims against digital footprints. The useful distinction is the job each performs: triage, exploration, visual examination and corroboration.
For a corporate security team, that might mean investigating fraud or checking a suspicious profile. For a government analyst, the focus might be a network around an existing lead. These are potential workflows, not proof that the system can establish guilt or predict an attack. A useful output gives the investigator something specific to examine next.
Voyager’s Microsoft marketplace listing describes a cloud partnership and an investigation suite. Its commercial proposition combines several analytical capabilities within one environment. That can reduce handoffs between tools. It also makes the buyer’s demonstration more demanding: the whole chain, from original material to final lead, needs to work.
There is another route into the data. Voyager’s warrant-reader guide describes software for examining social media material obtained through warrants. That workflow begins with records an agency already possesses, rather than treating the web as an endlessly available feed. For an investigator, the difference matters. The useful task may be organizing a large return, finding the relevant exchange and keeping enough context to understand it. Data volume alone does not determine which product belongs on the desk.
The price of a question
A public one-year VoyagerAnalytics listing provides an unusually concrete view of the business model. It quotes $95,000 for up to ten users and 2,500 queries; $140,000 for 4,800 queries; and $200,000 for 9,000 queries. These are published tiers, not a promise about today’s negotiated contract. They show that buyers purchase both seats and a quantity of investigation.
Competition is visible in procurement paperwork. In Panama City Beach’s 2023 evaluation, Cobwebs Technologies scored 97.20 out of 100 and Voyager 57.26. The city recommended Cobwebs, also the lower bidder, for a $50,810 purchase. One contest cannot rank an entire market. It does puncture the idea that a polished AI demonstration settles a purchasing decision.
When access becomes the obstacle
The more consequential interruption came from the platforms supplying the raw material. In January 2023, Meta sued Voyager, alleging fake accounts and unauthorized automated scraping. In December 2024, Meta announced a settlement under which Voyager agreed to a permanent injunction and a monetary payment. This was a settlement, not a trial verdict establishing every allegation. Data access had become a business constraint with consequences beyond the engineering team.
There was also a dispute over interpretation. The Brennan Center’s analysis of historical LAPD sales material questioned claims about ideology and dangerousness inferred from social posts and connections. Its objection reached the central difficulty: religious expression or an acquaintance’s beliefs cannot simply stand in for evidence of intended violence. An understandable risk indicator can conceal a contestable inference.
“Moving our business operations to the U.S. positions Voyager Labs to better serve our fastest-growing markets”
Avi Korenblum · company announcement, October 2022
What survives the demonstration
Voyager continues to pursue the market. An August 2025 company release reported agreements with four global intelligence agencies in the final quarter of 2024. It also promoted a unified interface under new executive management. The useful product idea is integration: an analyst examining a profile could also search its visual material. The release’s performance language should be read as a company claim.
For buyers, the practical lesson is to demand a reproducible path from input to conclusion. Try a known case. Inspect the original record. Test an ambiguous relationship. Ask how conflicting identities are handled. Establish which sources the organization can use and what happens when access changes. A system that accelerates weak inference may merely deliver the wrong lead sooner.
Voyager is most interesting at that boundary. Its tools address a real shortage of investigative attention. Their usefulness depends on available data, relevant criteria and people willing to challenge the resulting connections. The diagram can tell an investigator where to look. The work begins again when the investigator looks.