AI moves from pilot to production: Meet the third MINDS cohort

Published Jun 2, 2026 · Updated Jun 2, 2026

A woman is on a computer surrounding by gear: The third MINDS cohort demonstrate how AI is being embedded in production systems and producing tangible results

The third MINDS cohort demonstrate how AI is being embedded in production systems and producing tangible results. Image: Unsplash/ThisisEngineering

Cathy Li

Head, Centre for AI Excellence; Member of the Executive Committee, World Economic Forum

Maria Basso

Head, AI Applications and Impact, World Economic Forum

Artificial intelligence (AI) has left the pilot phase and is now embedded in production systems while already delivering tangible results.
AI deployment is shifting beyond digital workflows into factories, power grids, mines, hospitals and supply chains, involving a combination of autonomous AI agents with human oversight.
The World Economic Forum's MINDS initiative has selected 16 organizations to join its third cohort, representing innovative AI use cases that deliver measurable operational, societal and sustainability outcomes.

The pace at which artificial intelligence (AI) is being adopted has outrun the evidence of what it actually delivers. The World Economic Forum’s MINDS programme exists to close that gap by spotlighting deployments where AI has moved from pilot to delivering measurable results in production.

Our first cohort was announced at last year’s 'Summer Davos' in Tianjin, China, with 18 cases spanning from optimizing train operations to improving access to diagnostic care. A second cohort of 15 followed at the World Economic Forum's 2026 Annual Meeting in Davos, Switzerland, earlier this year.

The third cohort, unveiled today, adds 16 cases across 26 organizations and reveals three patterns that weren't yet visible a year ago.

First, AI is moving out of the screen and into the physical world. This cohort's deployments operate inside mines, factories, power grids and hospitals and not just inside data and decision tools.

Second, agent-based systems are arriving with governance built in. Tiered human oversight, operational safeguards and rollback paths are now standard from day one.

Third, sustainability outcomes such as lower energy use, less material waste and fewer kilometres driven are showing up as operational results even when emissions reduction was not the system's primary purpose.

We’ve mapped 16 cases across six themes: productivity, sustainability, healthcare, learning, trusted information and resilient infrastructure. What follows are the ways organizations are applying AI to solve these important overarching macrosystem-level challenges.

Cohort 3: In numbers

selected cases deployed in 28 countries across the globe

50%

of selected MINDS have headquarters in Greater China, 25% in Europe, 15% in North America and 3.5% across South America, Africa, Australia and New Zealand.

55%

of verified applicants improved accuracy, defected detection or reduced errors with AI.

30%

of verified applicants reduced the energy consumed by their operations after deploying AI.

50%

of verified applicants increased output or throughput per person or unit using AI.

Selected MINDS by theme

1. Accelerate growth through productivity and innovation

To raise productivity, quality and speed across operations so organizations deliver more value with fewer inputs.

Valterra Platinum and Dunia Innovation GmbH

Problem: Industrial catalyst and materials research and development (R&D) is slow, fragmented and reliant on manual experiments, with limited comparable data. In emission control and industrial systems, this delays development, drives costs and exposes projects to up to $300 million in risk.
Approach: A closed-loop autonomous platform that combines autonomous robotics, AI experiment selection, simulations and real-time learning to accelerate discovery, delivering experimental capacity in one day that is equivalent to the weekly output of 15 scientists.
Result: Increased industrial experimentation throughput by up to 5,500% – approximately 56 times higher – months-long R&D timelines reduced to weeks and lower operating costs from up to $2,000 to less than $300 per experiment.

Molecular Universe Pte. Ltd. (SES AI Corp)

Problem: Developing better batteries remains slow and costly, as scientists must navigate vast chemical design spaces using fragmented data and time-intensive testing, limiting progress in performance, safety, charging speed and lifespan.
Approach: The company has developed an AI-driven materials discovery platform that combines scientific literature, simulations, laboratory data and autonomous experimentation to identify and optimize battery materials using chemistry-focused AI models and continuous experimental feedback.
Result: Reduced electrolyte discovery timelines from approximately two years to three months, while physical experimentation has been cut by up to 70%, helping accelerate R&D efficiency, reduce costs and speed up the development of safer energy storage systems.

Qingdao Hisense Hitachi Air-conditioning Systems Co., Ltd.

Problem: Air-conditioner manufacturers face rising demand for customization and quality but fragmented data and manual workflows still dominate design, implementation and optimization, slowing launches, limiting automation and reducing quality.
Approach: An AI-enabled workflow unifies process design, implementation and optimization. Simulations reduce physical testing, software and documentation are generated automatically, while AI models inspect products in-line and production feedback continuously improve design.
Result: Factories using the AI-workflow achieved: 37% faster product development, 49% higher production efficiency per person, full AI quality inspection coverage, 30.4% fewer first-pass yield losses and 83% faster anomaly resolution.

KUKA and Mech-Mind Robotics Technologies Co. Ltd.

Problem: Modern manufacturing lines must handle dozens of variants with sub-millimeter tolerances, often in hazardous environments. Traditional robots only work when all its parts are in the same place, so each new variant demands custom fixtures and reprogramming, forcing manual operation that becomes harder to scale.
Approach: An intelligence platform, combining AI-driven 3D vision, world models and standardized software, turning any existing robotic arm into an adaptive, intelligent agent. It is compatible with over 1,000 robot models, enabling retrofit deployment without bespoke engineering, giving robots the “eyes, brain, and grippers” to handle real-world complexity while reducing fatigue and workplace risk.
Results: The partnered assembly line delivered 50% increased single-shift capacity and 30% improved labour efficiency. Hazardous operations were fully automated to avoid human harm and the line incorporated flexibility around new variants without downtime or re-engineering.

Xuchang Pang Donglai Trade Group Co. Ltd and Mettler Toledo (Changzhou) Measurement Technology Ltd.

Problem: Supermarkets rely on disconnected manual workflows across packaging, fresh food and checkout. Constant readjustment of machines, wasted resources and manual lookups also slows service, increases costs and reduces customer satisfaction.
Approach: A cloud-edge AI system connects packaging, fresh food and checkout. Using images, weight and dimensional data, AI-powered devices identify products with 99% accuracy, while a central platform monitors devices and continuously retrains models across stores, filtering faulty data.
Result: Checkout time per item dropped 60%, film waste dropped 20%, packaging pass rates rose 18 percentage points and pre-pack machine downtime fell 30-60 minutes per store daily.

2. Build a sustainable planet, energy and food systems

To cut emissions and waste, improve efficiency and grid reliability, and protect nature as economies grow.

Bota Biosciences

Problem: Biological R&D is bottlenecked by slow design-build-test-learn cycles; research proposals that should take weeks often take much longer. Ambiguous jargon and inconsistent language adds to the hold-up, stalling the scaling of biomanufacturing.
Approach: A self-driving lab for synthetic biology that automates the full pipeline from experiment planning to execution in the biofoundry. This is achieved by combining large language model with over 10 million proprietary experimental data points and a novel structured language (biology protocol language) that translates AI-generated plans into unambiguous, machine-executable laboratory protocols.
A self-driving synthetic biology lab automates the full pipeline from experiment planning to execution in the biofoundry, combining large language models, over 10 million proprietary data points and a structured biology protocol language that translates AI plans into machine-executable workflows.
Result: A reduction in planning-to-execution time in excess of 70% and a two- to three-fold increase in experiment throughput per scientist.

China Southern Power (CSG) Guangxi Power Grid, in collaboration with CSG AI Technology Co., and China Southern Power Grid

Problem: Critical power transmission routes can overload when renewable generation, extreme weather or grid topology shift electricity flows. Operators have relied on manual checks, on-site experience and recalculations within short windows, leading to slower, more conservative and less scalable decisions.
Approach: The system continuously reads real-time grid data, forecasts and network conditions. A reinforcement learning agent recommends safe adjustments to generators, storage and controllable loads, each checked against operating constraints, tested in simulation and reviewed by operators before execution.
Result: At the Guangxi control centre, severe overloads were reduced by approximately 90%. Response speed improved by roughly threefold, with strategies generated in seconds, while also delivering cost savings.

TW Solar

Problem: Solar cell manufacturing involves millions of ultra-thin silicon wafers moving continuously through highly automated production lines. Data for each cell is captured and distributed across multiple independent systems, requiring manual correlation when anomalies occur. The traceability systems used are largely passive, resulting in delayed corrections, declines in yield performance and increased manufacturing costs.
Approach: Using a wafer-level traceability platform and AI algorithms, the company built a closed-loop system that pinpoints root causes to specific equipment and processes while continuously monitoring and predicting abnormalities.
Result: Reduced root cause investigation from three days to three minutes, significantly improving problem response and corrective action efficiency, while increasing per-capita production by 7%, improving solar cell yield by 0.5% and raising conversion efficiency by 0.1%.

Shanghai Jiao Tong University and Lenovo

Problem: As AI, enterprise and high-performance computing (HPC) workloads continue to grow rapidly, data centres are facing increasing pressure from rising energy consumption, constrained power availability and the need to scale computing capacity while reducing carbon emissions.
Approach: An AI-driven orchestration platform that continuously optimizes enterprise, AI and HPC workloads across cloud and data centre infrastructure, treating energy as a schedulable resource. Powered by Lenovo xCloud AIOps, it uses a unified decision engine to optimize workloads, compute, cooling and power systems in real time.
Result: Deployed and validated at the Shanghai Jiao Tong University data centre, the solution achieved up to an 11% reduction in energy consumption and up to 310 tonnes less carbon dioxide emissions annually, while enabling greater compute output and infrastructure efficiency under the same power budget.

Occidental

Problem: Before investing millions in drilling, companies need a better understanding of geologic reservoirs and where oil and gas are most likely to be found. Today, geologists and engineers manually interpret vast geological and seismic data. Today, geologists and engineers manually interpret vast geological and seismic datasets – a slow, hard-to-scale process where small differences can materially affect drilling, reserves and capital allocation.
Approach: The company developed an AI toolkit that helps geologists and engineers build a shared first-pass view of subsurface conditions. Using well logs, rock chemistry, sensor data and historical interpretations, it correlates rock layers, predicts lithology before drilling, identifies geologically similar zones and estimates underground oil, gas and water. Integrated into existing workflows, the tools allow experts to review and refine AI-generated interpretations.
Result: Geological interpretation work that previously took months now takes days, with less drilling downtime, fewer drilling-tool changes and reduced horizontal drilling time.

3. Build healthier lives with better healthcare

To tackle late diagnosis, fractured care journeys, long waits and unaffordable treatment to improve outcomes and equity.

SBP Group

Problem: Bringing new medicines from trials to patients can take years. In medical institutions, specialized, compliance-driven decisions remain fragmented across systems, documents and individual expertise, slowing R&D, reducing operational efficiency and delaying patient access.
Approach: A governed enterprise AI operating model that embeds specialized agents across clinical, commercial and enterprise management workflows. A tiered autonomy framework automates low-risk tasks, supports medium-risk decisions with human collaboration and requires human approval for high-risk medical, financial and compliance decisions.
Result: The system supports more than 20,000 enterprise users and has cumulatively completed over 1 million AI-assisted reviews. Patient screening time has been cut by 83%, clinical site investigation cycles are 67% shorter, operational efficiency in core medical affairs scenarios improved by 30% and market analysis cycles reduced from one month to one day.

4. Expand learning, skills, culture and opportunity

To remove barriers of cost, language and location, and connect learning to skills and work across a lifetime.

The Territorial Delegation of Educational Development, Vocational Training, University, Research and Innovation in Granada, and The Mind, Brain and Behavior Research Center (CIMCYC) of the University of Granada (UGR), and ATENXIA

Problem: Around one-in-six children experience attention and literacy-related difficulties, yet current intervention approaches for children and adolescents still rely heavily on standardized, clinician-led models and manual professional supervision.
Approach: An AI-driven gamified platform that continuously adapts exercises and intervention pathways in real time, while keeping professionals in the loop.
Result: Literacy-related cognitive outcomes improved by 25-35%, while the time required for monitoring and manual intervention adjustment decreased by up to 50% over a 16-week intervention period.

5. Safeguard trusted information and governance

To make information reliable, increase cybersecurity and legal accountability, and service fast, fair and secure.

Netsafe and Standard Security

Problem: Email scams not only impact our economy, with reported losses of up to NZD 3 billion, but also affect our loved ones, disproportionately targeting digitally inexperienced users and financially vulnerable households. As scam prevention remains a proactive system, scammers continue to operate at almost no cost or friction.
Approach: Changing scam reporting from a passive intake process to an active adversary-disruption workflow, through a platform that creates personalized AI personas to autonomously engage scammers in long conversations, wasting their time and disrupting operations at scale.
Result: The platform mobilized over 40,000 citizens who forwarded scam emails for AI engagement, consuming more than 7,500 hours of scammer time across 297,490 emails and 9,510 unique scammer accounts.

Quipu

Problem: Over 80% of Latin America's informal workers – from corner shops, street vendors and delivery drivers – have no access to formal credit because traditional scoring relies on past borrowing history. With no record, you get rejected. The alternative is loan sharks run by criminal organizations, charging up to 800% interest.
Approach: A multimodal AI scoring engine that reads financial SMS notifications (capturing real-time income and spending patterns), combined with business videos, product photos and social media activity; treating all of these as credit signals. The system extracts over 535 variables and uses visual embeddings to assess business viability from images alone, adding 25% predictive power.
Result: The platform has processed more than 300,000 individual scores and supported around 49,000 loans. It reports an 80.1% renewal rate, a reduction in clients’ use of informal “gota-a-gota” lenders from 38.6% to 28.5% and 13.11% of clients reporting at least 15% income growth.

6. Build resilient infrastructure, communities and supply chain

To increase public safety and mobility, anticipate and absorb shocks, keep essential services running, and speed response and recovery.

TCL Industries Holdings Co. Ltd.

Problem: Home appliance logistics is challenging because products vary widely in size, weight and fragility. Container loading still relies heavily on manual planning and staff experience, leading to wasted space, higher costs, slower fulfillment and greater risk of damage and safety issues. This includes challenges that intensify as volumes and order complexity grow.
Approach: An AI-powered container loading system that integrates with enterprise systems to ingest real-time product, inventory and transport data. Using 3D optimization, it identifies the safest, most space-efficient loading plan, validates compliance requirements and generates visual packing instructions for warehouse teams.
Result: The system cut container planning time from more than 30 minutes to under 30 seconds, improved space utilization, reduced shipments and transportation costs, lowered more than 700 tonnes of carbon emissions and supports 7,000+ monthly loading operations.

Shougang Mining Corporation and EACON Mining Group

Problem: Mining haulage is often hazardous, labour-intensive and difficult to staff, particularly in remote or high-altitude operations, where drivers face dust, vibration, noise, harsh weather and long shifts. Open-pit mining environments are also highly dynamic, with unpaved haul roads, changing conditions, limited visibility and unexpected obstacles creating additional operational complexity.
Approach: The solution combines L4 autonomous driving, high-precision positioning, 360-degree perception, vehicle-to-vehicle communication and an integrated dispatch system designed for mining operations. It supports mixed fleets and more than 70 truck models, coordinating both autonomous and human-operated equipment throughout the loading, hauling and dumping cycle.
Result: A 90% reduction in haulage cost per tonne, 13.3% lower energy consumption and an 82.5% reduction in human workload associated with high-risk haulage tasks.

A year after cohort 1: Foxconn case study

In 2025, Foxconn, in partnership with BCG, was selected for Cohort 1 of the MINDS programme for Project Genesis, an AI transformation initiative reshaping electronics manufacturing.

Since joining the programme, Foxconn has successfully rolled out an updated version of the solution at their Hongfujin Precision Electronics (Chengdu) Co. Ltd factory in its Chengdu campus. It integrates AI ready data, orchestrates multi agent reasoning, reusable machine skills and digital twin validation into a unified adaptive operating system.

The platform also enables a "self thinking" factory, where multiple AI agents dynamically collaborate to optimize operations, respond to anomalies and continuously adapt to changing conditions.

The initiative has been deployed across more than 300 production lines and supports over 5,000 personnel, reducing anomaly response times by 47% and improving energy efficiency by 30%.

In addition, Foxconn has digitized more than 5,000 expert rules and manufacturing practices, achieving a 78% automation rate in knowledge reuse and creating a scalable foundation for software defined manufacturing across more than 200 business scenarios.

A signage at Foxconn booth at the International Automobile & Motorcycle Parts & Accessories Show (AMPA) trade show in Taipei, Taiwan, April 14, 2026 Image: REUTERS/Edgar Su

MIND cohorts represent much more in AI innovation

Hundreds of companies applied for the third MINDS cohort.

The projects presented today, alongside the previous cohorts, represent only a small percentage of the innovative and transformative use cases that can be embedded through AI and frontier tech applications.

What's becoming clear is that AI adoption is no longer a question of whether but how. Organizations across health, education and research and development are moving beyond isolated use cases toward enterprise-wide transformation; reimagining operations, infrastructure and even physical environments in the process.

Every deployment featured here is the work of dozens of contributors across organisations. Explore the full list of MINDS members behind these AI-driven solutions on our site.

Interested in applying for the fourth cohort? Organisations embedding AI with meaningful impact, supported by strong governance and responsible practices, are invited to apply via the MINDS website. Selected MINDS will be announced at the World Economic Forum Annual Meeting 2027 in Davos-Klosters, Switzerland.

License and Republishing

World Economic Forum articles may be republished in accordance with the Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International Public License, and in accordance with our Terms of Use.

The views expressed in this article are those of the author alone and not the World Economic Forum.