The American Arbitration Association-International Centre for Dispute Resolution (AAA-ICDR) has launched a groundbreaking AI arbitrator for construction disputes, developed with QuantumBlack, McKinsey’s AI division. This initiative marks a new era in alternative dispute resolution, promising to cut decision times by up to 75%. The digital arbitrator, which began service in November 2025, is already streamlining documents-only cases that previously took months to resolve.
By combining agentic AI with a human-in-the-loop framework, the system delivers faster, more affordable, and trustworthy awards for an industry that prioritizes predictable outcomes.
How the system was built
Developed with QuantumBlack, the AI system analyzes case filings and evidence from a vast dataset of past awards to generate draft decisions. These AI-generated drafts are then reviewed, modified, or approved by a human arbitrator, ensuring expert oversight while dramatically accelerating the resolution process for construction disputes.
The system’s foundation is a sophisticated AI model trained on over 1,500 historical construction awards, each meticulously annotated by expert arbitrators. QuantumBlack guided the data engineering and iterative testing, a methodology detailed in a McKinsey blog post. Agile development pilots allowed for continuous feedback from the AAA-ICDR Construction Panel to refine the AI’s prompts, reasoning, and interface.
Core Capabilities at a Glance
At its core, the AI arbitrator autonomously parses case filings and evidence, breaking down complex claims into granular issues. It then applies a structured library of legal prompts to generate reasoned draft awards. Each recommendation is benchmarked against historical data for consistency before being sent to a human arbitrator who retains full authority to accept, modify, or reject the AI’s suggestions. This transparency is enhanced by an explanation layer that cites precedents and shows confidence levels for each finding.
Human Safeguards Remain Central
Emphasizing that the system is a “co-pilot rather than an autopilot,” AAA-ICDR President Bridget Mary McCormack confirmed that human arbitrators validate every final decision, as noted in the official press release. These experts can revise the AI’s reasoning, request additional context, or override its suggestions entirely. This human-in-the-loop framework is a critical safeguard and aligns with new guidance from the Chartered Institute of Arbitrators recommending full disclosure of AI use.
Early Performance Indicators
Initial performance data from the pilot program is compelling. For disputes under $1 million, the median time to produce a draft award plummeted from 86 days to just 21. Participants also saw an average 30% reduction in administrative fees, attributed to faster scheduling and more concise records. These metrics are based on internal AAA-ICDR dashboards shared with participating counsel.
Roadmap for 2026
Looking ahead to 2026, the AAA-ICDR plans a phased expansion of the AI arbitrator into new sectors, including energy, commercial leasing, and cross-border supply contracts. The program will also become accessible for lower-value claims down to $50,000. Development is already underway to expand the training data while ensuring strict confidentiality, and a hybrid model for cases requiring oral hearings is currently in beta testing.
Industry Ripple Effects
This pioneering move by the AAA-ICDR is already creating ripple effects across the legal industry. Other major arbitral institutions are taking notice, with the Dubai International Arbitration Centre reportedly evaluating a similar model. The emergence of modular AI tools at events like Paris Arbitration Week suggests a growing ecosystem of legal tech designed to enhance different stages of dispute resolution. The AAA-ICDR’s blend of curated data, explainability, and human oversight is setting a viable new standard for speed and cost-efficiency in arbitration.
