Overview
ROSS Intelligence built a plain-language legal research engine. Lawyers asked questions in ordinary English, and the system searched judicial opinions for relevant passages and cases. The company sought to reduce the time and cost of legal research in a market dominated by Westlaw and LexisNexis.
ROSS raised about $13 million and sold to major law firms. Its independent life ended after Thomson Reuters sued in 2020, alleging that ROSS's training process infringed copyrights in Westlaw editorial headnotes. ROSS denied the claims but announced in December 2020 that the litigation had blocked new financing and exhausted operating cash. The platform closed on January 31, 2021.
The corporate entity remained alive to defend the case. In February 2025 the Delaware district court granted Thomson Reuters partial summary judgment and rejected ROSS's fair-use defense for the headnotes at issue. ROSS's history therefore combines an early, useful legal-AI interface with a terminal failure to secure durable rights and financing around a critical data source.
Founding Story
Andrew Arruda, Jimoh Ovbiagele, and Pargles Dall'Oglio founded ROSS in Toronto in 2014. The team combined legal practice and computer science. Arruda had worked at a litigation boutique; Ovbiagele and Dall'Oglio developed the technical product.
They entered Y Combinator's Summer 2015 batch and moved to the United States. Dentons-backed NextLaw Labs became an early investor and customer. Other reported users included BakerHostetler and Latham & Watkins.
The founding premise was access. Legal research subscriptions and associate time were expensive, and keyword search required specialist technique. ROSS would let a lawyer state a legal issue as a question and receive ranked passages from cases. The team used legal-domain embeddings and other search methods rather than presenting the product as a general chatbot.
Timeline
- 2014: Arruda, Ovbiagele, and Dall'Oglio founded ROSS Intelligence.
- Summer 2015: ROSS joined Y Combinator and moved from Toronto to the United States.
- 2015–2016: Dentons' NextLaw Labs backed the company, and major law firms announced use of the product.
- October 2017: ROSS raised an $8.7 million Series A, bringing announced funding to $13 million.
- Following years: The company expanded legal coverage and search capabilities.
- May 2020: Thomson Reuters and Westlaw sued ROSS in federal court in Delaware.
- December 11, 2020: ROSS announced it was stopping operations because the litigation prevented fundraising and cash was running out.
- January 31, 2021: The customer platform became unavailable after a transition and export period.
- February 2025: The court granted Thomson Reuters partial summary judgment on infringement and rejected ROSS's fair-use defense for the disputed headnotes.
What They Built
ROSS indexed US legal material and accepted questions in natural language. It returned ranked case passages intended to answer the research issue, with links to source opinions. Word embeddings helped the system match context and meaning beyond exact keywords.
The product targeted the first stage of legal analysis: finding authority. It did not replace a lawyer's duty to read cases, determine whether they remained good law, and apply them to facts. Its promise was faster discovery and fewer missed cases.
ROSS needed both public law and editorial structure. Judicial opinions are generally public records, but commercial legal databases add headnotes, classifications, citation treatment, and other editorial work. The lawsuit centered on whether materials derived from Westlaw headnotes were used to create ROSS's search system and whether that use was lawful.
Market Position
Target Customers
ROSS sold to law firms, legal departments, academics, students, librarians, solo lawyers, and self-represented users. Large firms offered credibility and contract value; smaller practices and public users fit the access-to-justice mission.
Market Size
Legal research is a recurring professional requirement. ROSS described an $8 billion market, though the reachable opportunity depended on jurisdictional coverage, citation reliability, workflow integration, and lawyers' willingness to add or replace a trusted research subscription.
Competition
Westlaw and LexisNexis owned deep collections, editorial systems, citators, customer relationships, and research habits. Bloomberg Law, Fastcase, vLex, Casetext, and specialist tools offered alternatives. ROSS differentiated through conversational retrieval, but incumbents could add similar interfaces to better-established content libraries.
The data boundary was decisive. Public opinions could support a competing search product, while proprietary editorial enhancements required separate rights or independent creation. A research interface could not compensate for uncertain provenance in the underlying corpus.
Business Model
ROSS sold subscription access to legal professionals and organizations. Enterprise adoption required security, jurisdictional coverage, training, support, and confidence in results. The company also needed continuing investment in data ingestion, search quality, citation treatment, and sales.
Funding covered the long path toward competing with entrenched research providers. When litigation made a new financing round unavailable, subscription revenue was not sufficient to sustain operations. Insurance covered the defense, but not the operating company.
Post-Mortem
ROSS's shutdown was directly explained by its founders. The company denied Thomson Reuters' allegations and called the suit meritless, yet acknowledged that the legal battle prevented fundraising and consumed the runway needed for development and marketing. It stopped taking customers and closed the product.
The case exposed a structural risk that should have been treated like infrastructure. Search quality depended on training and evaluation material. If the company could not prove the rights and independent provenance of that material, the entire product and financing story remained vulnerable.
The February 2025 order did not retroactively cause the 2021 closure, but it confirms that the underlying legal risk was material. The court ruled for Thomson Reuters on copyright infringement involving selected headnotes and rejected ROSS's fair-use defense. This report does not generalize that fact-specific ruling to every form of artificial-intelligence training.
ROSS correctly anticipated demand for question-based legal research. Comparable interfaces later became common. The company failed to survive long enough to benefit because litigation around how it built the system became financially terminal before the merits were resolved.
Key Lessons
- Data rights are part of product architecture. A core corpus needs documented provenance, permissions, and replacement options.
- Public source material does not erase editorial rights. Court opinions and commercial headnotes are different assets.
- Litigation can kill before judgment. A startup may win arguments later and still lose access to operating capital now.
- Accuracy needs source traceability. Legal users must be able to inspect the underlying authority rather than trust a generated answer.
- A good interface is not a durable moat. Incumbents can add natural-language search when they already control trusted collections and distribution.