Four Minutes Saved Per Case: Brazil’s MIDAS Offers a New Blueprint for AI-Powered Justice

AI-powered MIDAS increased judicial clerks’ productivity in Fortaleza by 37.6%, saving about four minutes per case file and highlighting AI’s potential to tackle repetitive court workloads. For governments and development partners, the bigger challenge is ensuring these efficiency gains translate into smaller backlogs, faster justice and lower costs without compromising quality or human oversight.

Artificial intelligence is beginning to show measurable productivity gains inside one of the most difficult public institutions to modernize: the judiciary. A 2026 study by the Inter-American Development Bank (IDB), involving the Court of Justice of Ceará (TJCE) and research participation from FGV Law School in Rio de Janeiro, finds that an AI-supported system deployed in Fortaleza increased judicial clerks' daily productivity by about 37.6%. The findings suggest that governments do not necessarily need to begin their judicial AI journey by automating complex legal decisions. Significant gains may instead come from using technology to remove repetitive administrative work while keeping humans firmly responsible for legal judgments.

Brazil demonstrates why such improvements matter. The country's State Justice system ended 2025 with nearly 58 million pending judicial processes, despite adjudicating and closing more than 30 million during the year. Ceará alone received 692,630 new processes and finished the year with more than one million pending. Judicial congestion is therefore not simply an administrative problem. Slow dispute resolution and contract enforcement can affect credit, investment, entrepreneurship and wider economic activity.

Can AI Turn Repetitive Court Work Into Faster Justice?

The technology tested in Fortaleza is MIDAS, the Similar Acts Identification Mechanism, developed under Ceará's PROMOJUD judicial modernization programme. The IDB-supported modernization project amounts to US$35 million and forms part of the broader US$1 billion Brasil Mais Digital credit line.

MIDAS uses natural language processing and unsupervised machine learning to identify judicial rulings containing similar text and organize them into clusters. Clerks can then prepare administrative documents, including summons letters, attachment orders and notifications in batches rather than repeating essentially the same process case by case.

Importantly, MIDAS does not decide cases. Judges retain responsibility for judicial decisions, while clerks remain responsible for administrative processing. This makes workflow automation a potentially lower-risk starting point for governments seeking to introduce AI into sensitive public institutions.

A 37.6% Productivity Gain Offers a Powerful Policy Signal

The pilot ran from June 30 to August 14, 2025, involving 62 clerks at Fortaleza's First-Degree Judicial Secretariat. The treatment group contained 37 clerks, compared with 25 in the control group. Researchers analyzed 43,614 case files originating from 30,024 distinct judicial processes.

The results were significant. Clerks with access to MIDAS produced about 10 more case files per day, a 37.6% productivity increase relative to the control-group average.

Assuming an eight-hour workday devoted to file preparation, control-group clerks produced about 30 files daily, or about 16 minutes per file. MIDAS reduced the implied processing time to about 12 minutes, saving roughly four minutes per file and enabling clerks to process about one-third more files within the same working time.

The finding also remained strong after researchers removed unusually productive observations. Depending on the specification, MIDAS continued to generate approximately 7.9 to 9.5 additional files per day.

Governments Must Look Beyond Files Processed Per Day

For policymakers, the Fortaleza experiment offers an important lesson: governments should identify repetitive workflow bottlenecks before investing heavily in more ambitious AI systems. Courts contain thousands of procedural tasks that may be suitable for automation without transferring legal authority to algorithms.

But higher employee productivity does not automatically mean faster justice. If clerks process documents more quickly but judges or other court departments cannot absorb the additional workload, AI could simply move the bottleneck elsewhere.

Governments should therefore measure success using broader indicators such as total case-processing time, backlog reduction, error rates, cost per completed case and service quality, rather than focusing only on documents processed per employee.

Quality represents another unresolved issue. The pilot measured the speed and volume of case-file production but could not determine whether MIDAS affected accuracy. Future evaluations should therefore examine whether productivity improvements are achieved without increasing mistakes or reducing procedural quality.

From Development Finance to a New Market for GovTech

For international development institutions, the results demonstrate why digital-government financing should increasingly combine technology investment with rigorous impact evaluation. Success should not simply be measured by how many AI systems governments purchase, but by whether those systems reduce processing times, lower costs and improve public services.

The private sector could also benefit from growing demand for natural-language processing, legal-document management, workflow automation, cybersecurity, cloud infrastructure and system integration. Technology providers capable of demonstrating measurable public-sector productivity gains may find expanding opportunities as governments modernize justice systems.

However, this market also carries risks. Governments will need strong rules covering data protection, cybersecurity, auditability, interoperability and human oversight. Technology vendors will increasingly need to demonstrate not only that their systems work technically, but that they produce measurable improvements without weakening institutional accountability.

Ceará is now moving toward a broader statewide randomized controlled trial, expected to examine MIDAS alongside another tool designed to assist judges in drafting decisions. That expansion could answer the central question left unresolved by the Fortaleza pilot: whether productivity gains ultimately reduce judicial processing times and case backlogs.

For governments, development partners and private-sector technology providers, Fortaleza provides a promising but cautious signal. A 37.6% productivity improvement shows that targeted AI can deliver substantial administrative gains. The bigger test is whether minutes saved at clerks' desks can eventually translate into faster, cheaper and higher-quality justice for citizens and businesses.