# EXL Completes Acquisition of iMerit to Strengthen AI Model Training and Evaluation Capabilities Across Enterprise Industries
The artificial intelligence industry is entering a transformative phase where the quality of models depends heavily on the expertise and data used to build them. In a significant move, ExlService Holdings Inc. (EXL), a global provider of digital solutions and analytics services, has finalized its acquisition of iMerit Technology, a specialized firm focused on AI model training, evaluation, and reinforcement learning. The deal marks a strategic effort to bridge the gap between raw AI development and real-world enterprise deployment, particularly in sectors where trust, accuracy, and reliability are non-negotiable.
## Background of the Two Companies
EXL Service Holdings, founded in 1 Centre99, operates worldwide with approximately 68,000 employees and serves a broad range of industries including insurance, healthcare, banking and capital markets, retail, communications and media, and energy and infrastructure. The company is headquartered in New York and has built its reputation by integrating technology into complex business operations on a large scale.
In 2012, iMerit Technology was founded with a specific focus on data annotation and model preparation for robotics, autonomous mobility, healthcare AI, and other high-stakes digital industries. Based in San Jose, California, iMerit differentiates itself through a proprietary platform called Ango Hub, which allows clients to collaborate on complex multimodal data, generating highly curated and validated training artifacts for mission-critical AI models. Its network of subject-matter experts works across geographies to label, evaluate, and fine-tune data for organizations that cannot afford errors in their AI pipelines.
## Why the Acquisition Matters
EXL’s Chairman and CEO, Rohit Kapoor, emphasized that this acquisition connects traditionally separate stages of the AI lifecycle into one unified platform for enterprise clients. iMerit brings expert-led model training, multimodal evaluation capabilities, a proprietary collaboration platform, and access to a global network of domain experts. EXL contributes deep industry knowledge, enterprise-scale data resources, and extensive experience in integrating technology into business workflows. Together, these capabilities aim to create a continuous pathway from preparing proprietary data to fine-tuning a model, evaluating its real-world behavior, and operationalizing it in business processes.
The acquisition is designed to help organizations transition from AI pilots to production-scale deployments. It also aims to make AI systems more trustworthy by ensuring models are tested against edge cases, unusual conditions, and specialized business contexts where failure is not an option.
One notable example of iMerit’s work before the acquisition was its partnership with Carbon Robotics, where it helped process millions of plant images to build an agricultural model used for robotic weeding systems. This project illustrates how detailed data annotation and evaluation are essential for physical AI applications in agriculture, where environmental variability demands highaccuracy models. The collaboration required digesting vast amounts of visual data to teach robots how to distinguish between crops and weeds in real time.
## Challenges AI Builders Often Overlook
Svaya Ramaswami Pulakula, Founder and CEO of iMerit (now EVP and Head of iMerit at EXL), pointed out models frequently underperform in unfamiliar scenarios or specialized industrial contexts despite scoring well on standard benchmarks. The real bottleneck isn’t just the algorithm; it’s often the lack of high-quality, domain-specific data and human insight required to evaluate and fine-tune models for specific business environments.
Domain experts play an irreplaceable role by challenging models, exposing failure modes, and validating whether a model’s behavior is reliable in specific roles. In healthcare, finance, and robotics, a dependable system needs more than general competence; it requires specialization. Human experts help build realistic scenarios, identify rare edge cases, and evaluate whether systems responded appropriately in complex, high-context situations where mistakes carry severe consequences.
## Physical AI and Robotics: A Growing Priority
Radha Basu further noted that the success of physical AI systems, including autonomous vehicles and robotics, depends less on scale and more on data quality. These systems must interpret noisy, multimodal inputs and reason in real time to identify correct actions in unpredictable environments. The challenges evolve from perception to planning and decision-making, requiring multiple layers of validation.
In autonomous driving, AI is trained using vision, lidar, and audio data, then tested through simulated scenarios like collisions. Experts evaluate not just what the system perceives but how it behaves and explains its decisions, ensuring it can act safely when encountering real-world edge cases. This is especially important in safety-critical and regulated industries.
## Safety and Compliance in AI Development
Safety and compliance must be designed into every process, not treated as a final check before launch. Organizations must demonstrate that their AI systems perform well on average and are also tested against situations where errors could create irreversible clinical, financial, or operational harm.
The concept of trace analysis is becoming more important in this space. By tracing data and model behavior, organizations can better understand where errors or unexpected outcomes originated, improving accountability and transparency in regulated industries.
## The Future of Enterprise AI
Most enterprise AI projects still struggle to achieve consistent business value. The challenge isn’t simply picking a model but training, evaluating, adapting, and governing systems to perform reliably in specific operations over time. As AI economics shift, the value moves from model-building to making AI usable, trustworthy, cost-effective, and outcome-driven.
Rohit Kapoor predicted competitive advantage will come from combining multiple layers: proprietary data, domain context, model evaluation, computation, and governance. Foundation models will improve, but the key question for enterprises is which AI system delivers the most reliable outcome under regulatory and operational constraints in a specific workflow. Over the next several years, proprietary enterprise datasets will become one of the most valuable assets. Companies will want specialized models trained on their own data and also need continuous evaluation and reinforcement learning to keep those models accurate as conditions evolve.
The companies that succeed will combine data, context, AI processes, evaluation, and execution into one operating model. The addition of iMerit strengthens EXL’s position significantly by giving it a direct role in how AI is built, governed, and deployed.
## Frequently Asked Questions (FAQ)
**What is iMerit Technology?**
iMerit is a firm specializing in data annotation, model training, evaluation, and reinforcement learning for AI systems used in robotics, autonomous mobility, healthcare, and other high-tech industries. It operates a proprietary platform called Ango Hub that enables clients to work on complex, multimodal data for mission-critical AI models.
**What is EXL Service Holdings?**
EXL is a global digital solutions and analytics company founded in 1999, serving industries including insurance, healthcare, banking, retail, media, and infrastructure. It employs roughly 68, widespread team members and has deep experience in integrating AI into business operations.
**Why did EXL acquire iMerit?**
To bridge the gap between AI development and enterprise deployment by unifying model training, evaluation, and operationalization into one platform. The acquisition gives EXL direct access to expert-led training, multimodal evaluation capabilities, domain-specific data networks, and a proven collaboration platform.
**What are some real-world applications mentioned?**
Physical AI for agriculture, such as robotic weeding powered by image analysis; autonomous driving using vision, lidar, and audio data combined with simulation testing; and healthcare and financial AI requiring high accuracy in regulated environments.
**What is the role of human experts in AI development?**
They identify rare edge cases, build realistic testing scenarios, validate model behavior in specific business contexts, and ensure trustworthiness. In robotics and autonomous systems, experts help evaluate whether models respond safely in unpredictable conditions where mistakes have severe consequences.
**What is the future outlook for enterprise AI?**
Model-building is shifting toward usability, trustworthiness, and outcome-driven systems. Proprietary datasets and specialized models trained on company-specific data will become more valuable. Continuous evaluation and reinforcement learning are essential for maintaining accuracy under changing conditions.
## Conclusion
The integration of AI model training and evaluation into enterprise practices requires a holistic approach that combines data expertise, automation, and human insight. EXL’s acquisition of iMerit reflects a broader understanding that building better models is only part of the solution; deploying them safely, tracking their behavior, and ensuring they generate reliable outcomes in real workflows is what drives business value. As industries adopt more AI systems, those that invest in comprehensive evaluation and governance will lead the way.
Thank you for reading
