Kenya’s SMaRT Experiment Could Reshape How Courts Match Mediators with Thousands of Cases

Kenya’s SMaRT system could improve civil dispute resolution by matching cases with mediators using performance, workload and accreditation data. Simulations show agreement rates of up to 61.7%, but a year-long randomized trial will determine whether these gains translate into real-world judicial efficiency.

Kenya's push to modernise court-annexed mediation is moving beyond basic digitisation towards data-driven decision-making. A September 2026 World Bank Policy Research Working Paper, developed through World Bank research and governance teams with researchers affiliated with Virginia Tech and University College London and collaboration from the Kenyan Judiciary, introduces SMaRT—Selecting Mediators that are Right for the Task. The system is designed to match civil cases with appropriate mediators while balancing successful settlements, mediator workloads and efficient use of judicial resources.

The initiative could have wider implications for Kenya's justice system and economy. Civil disputes that remain unresolved consume court time, increase costs for litigants and can create uncertainty for businesses and households. Court-Annexed Mediation offers an alternative by attempting to resolve disputes before they proceed to full judicial trials.

30,633 Cases Reveal Why Better Matching Matters

The research uses Kenyan Judiciary data covering 30,633 cases between 2016 and 2025, involving roughly 2,100 mediators, 87 court stations and 13 case types. Mediators are accredited for particular locations and categories of disputes, while the mediation secretariat prefers them to handle no more than three concurrent cases.

This makes assigning cases more complicated than simply choosing whoever is available. Historical analysis indicated that lower-performing mediators could receive cases ahead of stronger performers and workload limits could be unnecessarily exceeded.

The researchers estimated individual mediator effectiveness using a value-added model that accounts for differences including case type, court station, referral mode and time period. Their results suggest that a one-standard-deviation increase in mediator value added corresponds to a 13.36 percentage-point increase in the probability of successful case resolution.

For policymakers, this means mediator assignment should be viewed as part of judicial performance rather than a routine administrative exercise.

Higher Settlement Rates, but Workloads Need Careful Balancing

SMaRT uses an optimisation system that considers mediator performance, accreditation, current workload and likely future demand. Instead of automatically selecting the mediator with the strongest record, it can preserve that mediator's capacity for future cases where fewer alternatives are available.

Simulations demonstrate the potential benefits. When mediator performance is assumed to be known, assigning cases according to the lowest workload produces a 46.7% agreement rate, while a Greedy performance-based approach reaches 56.9%.

Depending on how strongly the algorithm penalises mediator overload, SMaRT produces agreement rates ranging from 57.4% to 61.7%. A theoretical upper-bound approach reaches 65.9%, but it ignores practical capacity constraints and heavily concentrates assignments.

The numbers reveal an important policy trade-off. Aggressively targeting settlements could overload high-performing mediators, while focusing too heavily on equal workloads could reduce expected resolution rates. Judicial authorities therefore need transparent rules determining an acceptable balance between performance, workload and fairness.

Faster Mediation Could Strengthen the Business Environment

Under Kenya's Court-Annexed Mediation system, civil cases can be referred to mediators before proceeding to trial, with mediation operating within a 70-day window. Cases reaching agreement are resolved, while unsuccessful disputes return to court.

If smarter matching increases successful settlements in real-world implementation, it could reduce the number of cases returning to conventional trials and ease pressure on the judiciary. Businesses could also benefit from more efficient dispute resolution, particularly smaller companies that have fewer resources to absorb lengthy legal proceedings.

For international development partners, Kenya offers a potentially useful GovTech experiment. Rather than simply financing the digitisation of paper-based processes, development programmes could examine how administrative data can improve the allocation of scarce public-sector resources.

Private-sector opportunities could emerge in case-management platforms, secure cloud infrastructure, data analytics, algorithm auditing and digital public-sector solutions. But risks must also be addressed. Judicial information can be sensitive, and poorly governed automated recommendations could create concerns about transparency, privacy, accountability or unfair treatment.

Human oversight therefore remains important. SMaRT is intended to support mediator recommendations through the Cadaster digital mediation platform, with Court-Annexed Mediation officers remaining involved in assignment decisions.

Kenya's Real Test Will Come Outside the Simulation

The system appears technically capable of handling substantial complexity. In full-scale experiments, the median optimisation involved 52 cases and 701 eligible mediators, creating 3,632 variables and 14,326 constraints. Despite that scale, the median solution time was only 0.25 seconds. The 95th percentile was 1.23 seconds and the 99th percentile 2.68 seconds.

However, these results largely come from simulations using historical judicial data. They should not yet be interpreted as evidence that SMaRT has improved court performance nationwide.

The algorithm has been approved for a year-long randomized controlled trial, providing an opportunity to test whether predicted improvements translate into actual settlements and manageable mediator workloads.

For Kenya and its development partners, the priority should therefore be rigorous evaluation before wider adoption. Authorities should track settlement rates, case duration, mediator workloads, distribution of assignments and performance across different courts and case categories while maintaining human oversight and data safeguards.

If real-world evidence supports the simulation findings, Kenya's experiment could provide useful lessons for governments seeking to allocate scarce skilled personnel more efficiently. Its wider value will depend not simply on whether an algorithm works, but on whether technology can improve public services while preserving fairness, transparency and accountability.