How Vertical AI Achieves Defensible Accuracy - with Steve Hasker of Thomson Reuters
Summary
Steve Hasker, CEO of Thomson Reuters, discusses how "vertical AI" achieves defensible accuracy in regulated legal, tax, and audit environments, where errors carry significant regulatory and financial consequences. Vertical AI is defined as industry-specific, purposefully trained with deep domain expertise, and grounded in authoritative, continuously updated content to meet professional standards for trust, accuracy, data security, and privacy. Thomson Reuters leverages centuries of curated legal content (e.g., 250 years in the UK, 150 in the US) and 4,500 domain experts to train AI agents for tasks like litigation research, drafting, and tax return preparation via products such as Westlaw Advantage, Co-Counsel, and Ready to Review. This approach ensures 100% accuracy, which is critical as general-purpose models (even 97% correct) are insufficient for fiduciary work, emphasizing the ongoing necessity of human oversight for judgment and accountability.
Key takeaway
For AI leaders deploying solutions in regulated legal, tax, or audit environments, recognize that general-purpose AI models, even with high accuracy, are inadequate for fiduciary tasks. Your strategy must prioritize "vertical AI" development, integrating deep domain expertise, authoritative data, and expert-trained agents to ensure 100% defensible accuracy. Implement robust audit trails and human-in-the-loop processes to maintain accountability and mitigate severe regulatory and financial consequences.
Key insights
Vertical AI for fiduciary professions must be purpose-trained on authoritative content by experts to achieve non-negotiable 100% accuracy.
Principles
- Fiduciary tasks demand 100% AI accuracy.
- Human oversight provides judgment and accountability.
- Domain experts must train and curate AI agents.
Method
AI agents are trained by domain experts to replicate multi-step professional workflows, like M&A transactions, providing verifiable outputs and highlighting areas for human review.
In practice
- Generate first drafts of legal briefs or tax returns.
- Automate tax source material organization.
- Refine legal arguments iteratively with AI tools.
Topics
- Vertical AI
- Regulated Industries
- Legal Technology
- Tax Automation
- Audit Workflows
- Thomson Reuters
- AI Accuracy
Best for: CTO, VP of Engineering/Data, Executive, Director of AI/ML, AI Product Manager, Consultant
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by The AI in Business Podcast.