Technology, AI Economics
Unlocking Agentic AI: Transforming Processes and Operations at Scale
How agentic AI transforms enterprise processes and operations at scale, with real case studies and EU AI Act, CRA, and NIS2 compliance guidance.
Publication
15 min
This whitepaper was co-authored with Vishal Vadabhat and Sagar Mulay, and originally published externally by LTM.
Executive Summary
Agentic AI is redefining how organisations approach process automation and operational efficiency. By enabling autonomous, goal-driven decision-making, it goes beyond traditional AI systems to deliver self-healing capabilities, proactive actions, scalability, cost savings, and improved accuracy.
This paper explores how agentic AI can transform business processes and operations at scale, with a particular focus on the European Union’s regulatory landscape — where the EU AI Act introduces stricter requirements for transparency, accountability, and compliance that businesses must navigate to responsibly adopt high-risk AI systems.
Drawing on real-world case studies across IT, sales and marketing, and after-sales service, the paper demonstrates measurable business outcomes and provides a roadmap for organisations to unlock the full potential of agentic AI as part of a broader enterprise AI transformation.
Introduction
AI is undergoing a significant shift, from reactive systems that respond to inputs, to proactive systems capable of autonomous action and collaboration. Generative AI, which focuses on producing content from user prompts, has paved the way for agentic AI, where intelligent agents independently set and achieve goals, interact with other systems, and drive complex workflows without constant human oversight.
An earlier paper, “Architecting Agentic AI,” explored the core principles, implementation challenges, and emerging trends surrounding this technology. Building on that work, this paper focuses on two practical applications: process-based AI, which automates individual tasks and workflows, and operations-based AI, which optimises broader business functions to deliver AI in business operations at scale.
With the European Union taking a leading role in AI governance through the EU AI Act, organisations face both opportunities and challenges in adopting these advanced systems — setting the stage for how agentic AI can be deployed responsibly as part of an enterprise AI transformation strategy.
European Context for AI Use
EU AI Act
The European Union has introduced the EU AI Act, a key regulatory framework governing the use of artificial intelligence within the EU. The Act classifies AI systems into risk levels — unacceptable risk, high risk, and minimal or no risk — based on the potential harm they may cause. Under Article 3(2) of the Act, risk is defined as “the combination of the probability of an occurrence of harm and the severity of that harm,” which has several implications for agentic AI systems.
Due to its autonomous and complex nature, agentic AI is likely to be classified as high risk, potentially affecting how it is adopted and deployed.
In sensitive sectors such as healthcare, transportation, and education, where decisions directly impact human lives and involve personal data, agentic AI will face stricter regulatory scrutiny.
The Act places strong emphasis on transparency and accountability for high-risk systems — agentic AI solutions designed with clear auditability and ethical practices will be better positioned for approval and may face reduced regulatory barriers.
Agentic AI in Action: Automated Insurance Claims Processing
A practical example of agentic AI in action comes from a major Dutch insurer that automated up to 90% of its automobile claims workflow using custom AI agents. These autonomous systems review and assess claims without human intervention, leading to significant operational efficiency gains, faster claims processing times, and the ability for employees to focus on creative and strategic tasks rather than routine administrative work.
Cyber Resilience Act (CRA)
Two other major regulatory frameworks that can affect the adoption of agentic AI systems in Europe are the Cyber Resilience Act and the NIS2 Directive. The CRA ensures that digital products — including hardware and software — meet cybersecurity requirements, promoting a “cybersecurity by design” approach to minimise the likelihood and impact of cyberattacks. Connected technologies such as IoT devices, AI-driven tools, and analytics platforms fall within its scope, requiring manufacturers to build secure products from the outset, provide ongoing security support, and deliver regular software updates to address vulnerabilities.
NIS2 Directive
The NIS2 Directive expands on the original Network and Information Systems Directive, extending its scope to a broader set of sectors: energy, transportation, healthcare, finance, water management, digital infrastructure, public communications and digital services, waste management, postal and courier services, and critical product manufacturing and public administration. It requires enhanced incident reporting and stricter security measures, ensuring that essential service providers maintain a high standard of cybersecurity.
Impact of CRA and NIS2 Directive on Agentic AI Adoption
Noncompliance with either regulation carries significant penalties — fines under the CRA can reach 2.5% of global annual turnover. For organisations deploying agentic AI, this means factoring in cybersecurity and compliance from the earliest stages of development.
Higher Development Costs: investments are needed to ensure secure design, proactive monitoring, and robust incident reporting mechanisms.
Regular Audits: frequent, rigorous audits must be conducted to identify and mitigate cybersecurity risks.
Cross-Functional Collaboration: legal, cybersecurity, and AI teams must work closely from design to deployment to maintain compliance and reduce exposure to risk.
By addressing these requirements early, organisations can reduce compliance hurdles and strengthen trust in their AI systems.
Process-Based AI
The next sections examine two core applications of agentic AI: process-based AI and operations-based AI. Process-based agentic AI focuses on automating individual workflows and tasks. These systems break processes into multiple sub-processes, with specialised AI agents managing each step autonomously.
A comprehensive knowledge base, such as a large language model, supports decision-making by providing access to historical data, standard operating procedures, rules, and workflows. This enables the system to perceive its environment, process information, and act independently to streamline and optimise operations.
Example: IT Service Management (ITSM) Using Agentic AI
Building the Knowledge Base for Ticket Resolution
In ITSM, agentic AI can significantly enhance problem-solving accuracy, speed up ticket resolution, and automate various repetitive sub-processes. The foundation of this system lies in a robust knowledge base, built using historical ticket data and resolutions, records of previously implemented solutions, categorisation of issue and incident types, and knowledge and resource data from connected systems such as CRM and ERP. Multiple AI agents use this information to simplify sub-processes and automate various steps in the ticket management process.
How Agentic AI Resolves Issues
When a new ticket is raised, agentic AI performs several functions autonomously: it categorises the issue by determining its type, severity, and potential impact; searches for solutions by referencing historical data, knowledge bases, and predefined rules; and acts on insights by automatically resolving the issue, suggesting potential solutions for user selection, or assigning the ticket to the most qualified resource based on historical performance patterns.
Intelligent Issue Prioritization
The agentic AI-based ITSM system can also identify and segregate multiple issues and incidents, ranking them by priority based on factors such as the user raising the issue, the impact on workflows, and the severity of the problem. It addresses these issues in priority order, and can recognise duplicate service requests and suppress them — focusing only on the original request — which reduces the overall number of tickets and shortens the time required to resolve issues.
Success Story: Optimizing Configurations and Quotes for CHVAC Products
A global leader in building cooling solutions faced challenges in their sales process, seeking to improve price prediction accuracy for new product configurations while increasing the volume of quotes analysed. LTM deployed an agentic AI-powered multi-agent system using advanced machine learning and deep learning models, alongside historical customer data, to recommend optimal configurations and generate accurate quotes.
Accurate Price Predictions: achieved 82% accuracy for new product configurations using 18 ML models and three deep learning models.
Boosted Sales Potential: enabled a 7x increase in sales opportunities.
Enhanced Quote Analysis: processed nearly three million quotes post-implementation.
Improved Sales Conversions: the platform incorporated market-based pricing logic and tracked pricing decisions to support better lead conversion.
Operations-Based AI
How Agentic AI Supports Business Operations
Operations-based AI applies agentic AI to optimise and automate broader business operations, rather than focusing solely on isolated processes. This approach enables real-time decision-making, adaptability, and self-healing capabilities — a hallmark of AI in business operations. Common applications include autonomous vehicles, predictive asset maintenance, and smart city energy management systems. By dividing large-scale operations into manageable subtasks, synchronised AI agents can collaborate to ensure continuous optimisation and rapid response to changing conditions.
Example: Predictive Maintenance
For predictive maintenance of assets, operations-based AI systems use machine learning, data analytics, and real-time data to forecast equipment failure and take appropriate action to prevent it — shifting from a reactive approach to an initiative-taking one, helping reduce unplanned downtime, improve equipment lifecycle, and increase operational efficiency.
The system requires real-time operating data such as pressure and temperature from equipment, collected from sensors. This data is cleansed, and appropriate parameters are selected through data analytics. A machine learning algorithm then identifies anomalies and compares them with historical operational data to detect patterns and predict equipment failure, after which the agentic AI system selects an appropriate resource based on availability and schedules maintenance to avoid failure.
Success Story: Predictive Maintenance for Elevators and Escalators
A leading global elevator company partnered with LTM to implement a “Transportation as a Service” predictive maintenance model, enhancing both service quality and customer experience. The solution used IoT sensors and agentic AI to monitor elevator and escalator health in real time and provide proactive maintenance alerts.
Revenue Growth: increased service contract revenue by 2%.
Cost Reduction: reduced field service costs by 33%.
Faster Turnaround: cut service turnaround time by 50%.
Data-Driven Monitoring: collected IoT data from 284,000 units globally, deploying 14 AI models per unit.
Proactive Alerts: real-time monitoring and automated notifications via a mobile app, supported by historical service data for predictive analytics.
Necessity of Agentic AI in Process-Based and Operations-Based Systems
The nature of agentic AI systems makes them highly suitable for both process-based and operations-based applications. These systems not only outperform non-AI solutions but also deliver distinct advantages over traditional or standalone AI, which often depend heavily on manual oversight and static, pre-trained data.
Self-Healing and Autonomous Nature: agentic AI systems operate independently, completing multiple tasks without human intervention, and can auto-correct when a defined goal is not met — unlike traditional AI, which requires ongoing human monitoring and manual correction.
Proactive Approach: unlike traditional AI, which is reactive and waits for input, agentic AI anticipates issues and initiates solutions proactively, making it especially effective in dynamic, fast-changing environments.
Accuracy Improvement: goal-driven by design, agentic AI consistently improves output accuracy over time, minimising human error in high-precision areas such as financial forecasting and medical diagnosis.
Scalability and Cost-Effectiveness: these systems manage large workloads and datasets without compromising speed or quality, reducing reliance on manual effort and cutting costs, time, and resource requirements.
Challenges in Adopting Agentic AI Systems
While agentic AI offers transformational benefits, its adoption comes with technical, regulatory, and operational challenges that are essential to address early for smooth implementation.
Stringent Regulations: the EU AI Act and other frameworks emphasise transparency and accountability for high-risk AI, and the complex algorithms behind agentic AI may produce outcomes that are difficult to interpret, creating trust and compliance issues.
Cybersecurity Risks: because agentic AI systems interact with third-party applications and handle sensitive data, they require robust cybersecurity measures to prevent breaches and unauthorised access.
Data Readiness: high-quality, diverse datasets are critical for accurate outcomes — poor or incomplete data can lead to errors or bias, making a thorough data readiness assessment essential before deployment.
Operational Integration Challenges: implementation can involve integrating with legacy systems, scaling across environments, performance monitoring, and ensuring access to clean, relevant training data.
User Acceptance and Trust: agentic AI may be perceived as a threat to jobs or as a “black box” system due to its complexity, so building user trust through transparency, education, and pre-adoption strategies is crucial for long-term success.
LTM offers a structured approach to help businesses adopt agentic AI responsibly and effectively, including end-to-end assessments of current IT landscapes to evaluate system maturity and identify gaps. Its AI practice includes over 500 data scientists and more than ten accelerators to expedite AI rollouts, supported by eight Centres of Excellence serving over 100 customers globally.
Conclusion
Agentic AI represents a paradigm shift in how organisations manage processes and operations. By enabling autonomous decision-making and optimising workflows with minimal human intervention, it drives adaptability, precision, and scalability at a level traditional AI cannot match.
Its full potential can only be realised by carefully addressing regulatory, ethical, and operational challenges. Building trust through transparency, ensuring data quality, and planning for seamless integration are essential steps for successful adoption.
As AI continues to advance, agentic AI will play a pivotal role in shaping the future of autonomous systems across industries. Organisations that develop a clear roadmap and well-defined implementation strategy today will be best positioned to lead in this emerging landscape.
CONTINUE EXPLORING
More writing on technology, AI economics, systems thinking, leadership, and mentoring.
CONNECT