Harnessing Agentic AI: Strategies for Business Transformation
Technology Leadership: The Key to AI Success
Research from Boston Consulting Group highlights a consistent trend: companies that pioneer technological advancements maintain a competitive edge in leveraging artificial intelligence (AI), particularly agentic AI, to optimize business operations. Jessica Apotheker, BCG’s managing director and senior partner, emphasizes that the frontrunners in AI adoption today are largely the same organizations that led the charge nearly a decade ago.
Moreover, the gap in value creation between these leaders and their competitors is expanding. BCG’s latest findings reveal that firms heavily investing in technology reap significantly higher returns, underscoring the importance of sustained commitment to AI innovation.
Despite widespread enthusiasm, many AI projects fail to deliver tangible business outcomes. BCG’s Build for the Future 2025 report shows that top-tier AI adopters achieve revenue growth rates 1.7 times greater than 60% of companies categorized as stagnant or emerging in AI maturity.
Ilan Twig, co-founder and CTO of Navan, attributes many underperforming AI initiatives to the tendency of layering AI solutions atop outdated legacy systems and inefficient processes, which limits their potential impact.
From Robotic Process Automation to Intelligent Process Orchestration
Building on existing automation frameworks like Robotic Process Automation (RPA) offers a practical pathway toward more sophisticated AI-driven workflows. Bernhard Schaffrik, principal analyst at Forrester, recently discussed at the Forrester Technology and Innovation Summit in London how agentic AI can enhance deterministic RPA by introducing adaptability and nuanced decision-making.
Forrester describes this evolution as “process orchestration,” where AI agents manage complex workflows with greater flexibility than traditional scripted automation. Unlike RPA, which requires anticipating every possible exception during design, agentic AI can dynamically handle uncertainties and unexpected scenarios.
Schaffrik notes that orchestrating intricate processes through conventional automation tools is often impractical due to the sheer complexity of anticipating all variables upfront. Emerging AI-driven orchestration platforms are addressing this challenge by enabling more resilient and intelligent process management.
Establishing a Strong Data Backbone for AI Success
BCG underscores that a robust data infrastructure is fundamental to scaling AI capabilities effectively. Clear governance frameworks and standardized, high-quality data sets are essential prerequisites for deploying agentic AI at scale.
As Ilan Twig points out, “AI’s effectiveness is directly tied to the quality and consistency of the data it processes.” Many organizations struggle with fragmented or inconsistent data, which hampers reliable AI training and deployment. However, building a comprehensive data foundation can be approached incrementally, focusing on cleaning and standardizing data project by project.
Apotheker suggests that organizations set overarching goals for data hygiene while progressively enhancing datasets to support AI initiatives. This approach ensures adherence to data management standards and prepares the groundwork for future AI projects.
Successful agentic AI strategies often rely on a metadata layer that connects AI agents, enabling clear delegation of tasks and decision-making responsibilities between humans and AI. Scott Willson, head of product marketing at xtype, describes AI workflow platforms as orchestration engines that integrate multiple AI agents, data sources, and human inputs through complex, non-linear workflows.
These platforms typically employ event-driven architectures with asynchronous message queues to ensure fault tolerance and scalability. Metadata propagation systems track every data transformation and model inference with unique identifiers, creating immutable audit trails that meet stringent regulatory requirements. Advanced implementations may utilize append-only logs, akin to blockchain technology, to prevent retroactive data tampering.
Reimagining Workflows and Embracing Organizational Change
Alan LeFort, CEO and co-founder of StrongestLayer, emphasizes that the primary challenge in adopting AI workflow platforms is organizational rather than technological. He references Conway’s Law, which states that system designs tend to mirror an organization’s communication structures.
LeFort argues that many companies err by attempting to integrate AI into existing processes designed around human limitations, resulting in incremental improvements. Instead, transformative gains arise from redesigning workflows to leverage AI’s unique capabilities, such as parallel task execution and elimination of knowledge silos.
“Integrating AI into human-centric processes yields marginal gains; redesigning workflows around AI unlocks exponential improvements.”
StrongestLayer exemplified this approach by overhauling its front-end software development process. Traditionally, product development follows a linear path-from customer requirements gathering to design, approval, and implementation-taking 18 to 24 months. Instead of layering AI onto this sequence, the company created new roles and AI pipelines that capture contextual knowledge and enable parallel workflows, dramatically accelerating delivery.
LeFort acknowledges initial resistance but highlights the importance of leadership by example and reframing success metrics around velocity and innovation. He stresses that true speed-to-value requires upfront investment in process redesign rather than rapid technology deployment.
Organizations should prioritize AI platforms that facilitate fundamental process transformation over those that merely automate existing inefficiencies.
Strategic Deployment of Agentic AI for Effective Decision-Making
BCG’s research advocates for a focused approach to deploying AI agents, concentrating on a select number of high-impact workflows and implementing them methodically rather than dispersing efforts broadly.
Ranil Boteju, Chief Data and Analytics Officer at Lloyds Banking Group, highlights the importance of leveraging multiple AI models tailored to specific tasks within a workflow. This modular approach allows different agents to address distinct components of a customer query, enhancing accuracy and efficiency.
Boteju envisions a system where AI agents not only perform tasks but also validate each other’s outputs, functioning as internal reviewers or “second-line coworkers.” This peer-review mechanism helps minimize errors and improves overall decision quality.
Prioritizing IT Security in Agentic AI Deployments
IT professionals understand the criticality of cybersecurity best practices, yet Fraser Dear, head of AI and Innovation at BCN, warns that many AI users lack the developer mindset necessary to embed governance into AI agent creation.
AI agents often access sensitive data repositories such as SharePoint, which may contain multiple document versions, confidential HR records, and salary information. Without proper safeguards, agents may inadvertently access or expose unauthorized data.
This risk escalates when AI agents are shared across teams, potentially granting inappropriate access beyond intended permission levels.
Dear advocates for comprehensive data governance policies that define data boundaries, restrict access based on roles and sensitivity, and specify permissible data sources for AI agents. He stresses that AI agents should be purpose-built following the principle of least privilege.
Like any critical application, AI agents require rigorous testing, including penetration testing to identify potential data exposure risks. Continuous monitoring, auditing, and real-time alerts are essential to detect and respond to abnormal access patterns promptly.
