Unlocking the Potential of AI in Enterprises: NTT DATA AIVista's Approach

The rise of artificial intelligence (AI) has prompted businesses across various sectors to allocate significant budgets towards integrating these technologies into their operations. However, many organizations find themselves grappling with a persistent challenge: the 'last mile' of implementing AI effectively. Bratin Saha, CEO of NTT DATA AIVista, recently shared insights at the VB Transform 2026 conference on how enterprises can navigate this crucial phase to unlock the full potential of AI. The conversation revealed that success hinges not merely on technology, but on a holistic approach that incorporates specialized knowledge, domain expertise, and a deep understanding of existing workflows.

Saha emphasized that the last mile is where many enterprises stumble, often failing to achieve the desired outcomes from their AI initiatives. According to him, the key to overcoming these obstacles lies in transforming AI models into tailored solutions that resonate with specific business contexts. This process involves wrapping AI models in a company’s unique data and operational processes, ensuring that they adhere to regulatory standards and align with institutional knowledge. By doing so, enterprises can turn their AI investments into tangible value rather than allowing them to become sunk costs.

The Last Mile Challenge in AI Implementation

Understanding the last mile of AI implementation is crucial for enterprises looking to capitalize on their investments. According to Saha, many frontier models, which are the cutting-edge AI technologies of today, often fall short of production-grade accuracy when applied to complex, regulated environments. For instance, in industries like insurance, workflows can involve intricate tasks such as processing multinational claims, which often feature handwritten inputs and numerous checkboxes. These complexities pose significant challenges for standard AI models, such as Fable 5 or GPT-5.5, which may not meet the rigorous demands of real-world applications.

To bridge this gap, last-mile specialization is essential. Saha indicated that the most effective approach involves customizing not only the AI model itself but also the entire system surrounding it. By integrating client-specific data, workflows, and the often-overlooked 'tribal knowledge'—the unwritten rules and practices that guide employee behavior—businesses can enhance their AI capabilities. This specialization leads to a more reliable and effective AI system that is better suited to meet regulatory requirements and internal standards.

What Does Last-Mile Specialization Entail?

Last-mile specialization is a multifaceted process that requires a deep understanding of the enterprise's unique context. Saha noted that this does not entail merely fine-tuning an AI model, which ranked low among enterprises' priorities in a recent survey by VentureBeat. Instead, the focus should be on leveraging proprietary data to build a supportive system around the model. This system must include specialized guardrails that monitor the AI's performance and rectify errors as they occur.

Implementing this approach involves three critical components:

• Contextualization: Capturing the enterprise’s unique operational context and making it comprehensible for the AI system.
• Model Integration: Employing an ensemble of models to ensure cost-effective operation while maintaining accuracy.
• Guardrail Implementation: Establishing checks that prompt the AI to correct mistakes, thereby enhancing reliability.

The Importance of Domain Expertise

As Saha pointed out, success in implementing AI effectively hinges on three intertwined elements: technology, domain expertise, and change management. While cutting-edge technology is essential, it is often the lack of specialized knowledge and understanding of internal processes that stifles progress. NTT DATA AIVista’s strategy involves pairing AI practitioners with domain experts to ensure that the AI systems mirror real-world workflows accurately.

By engaging with employees who possess in-depth knowledge of their tasks, the organization can encode this expertise into AI agents, ultimately enhancing the system's performance and relevance. This dual approach of leveraging technology and human insight is critical for enterprises operating in highly regulated industries, where compliance and accuracy are paramount.

Transforming AI Budgets into Tangible Value

For enterprises wrestling with the question of how to derive value from their substantial AI investments, Saha has a clear message: the key lies in focusing on the workflows that the AI is designed to enhance, rather than solely on the technology itself. He asserts that the real value of AI is realized when organizations effectively transition workflows from one state to another—moving from point A to point B.

This perspective necessitates a shift in how businesses allocate their resources. Instead of funneling all funds into model development, companies should prioritize investments in domain expertise and change management, which are often the true bottlenecks in AI deployment. By doing so, organizations can unlock a greater share of the potential return on their AI investments.

Furthermore, Saha noted that enterprises do not have to choose between embedding AI into existing workflows and completely redesigning those workflows. Instead, these processes should be viewed as consecutive stages in a comprehensive transformation journey. Starting with embedding AI into current processes allows for smoother change management, particularly for mission-critical operations.

Balancing Bespoke Solutions with Scalability

In the pursuit of effective AI integration, it is vital to strike a balance between creating bespoke solutions tailored to an organization's specific needs and developing scalable systems that can be adapted across different clients. Saha explained that keeping intelligence in the operational framework surrounding the model rather than within the model itself allows enterprises to maintain flexibility and capitalize on evolving technologies, such as open-source models.

NTT DATA AIVista employs a mixed approach, utilizing both frontier and open-source models. This strategy enables organizations to leverage open weights in low-risk scenarios while reserving more sophisticated reasoning for high-stakes situations where errors can have significant consequences. By adopting this methodology, enterprises can enhance their operational efficiency while minimizing risks.

Key Takeaways

• The last mile of AI implementation is crucial for turning investments into tangible business value.
• Success relies on specialization that integrates domain knowledge, proprietary data, and workflows.
• Focus on change management and expertise rather than solely on technology to maximize ROI.
• Balancing bespoke solutions with scalability is essential for adapting to evolving AI technologies.

Frequently Asked Questions

What is the last mile in AI implementation?

The last mile in AI implementation refers to the critical phase where AI technologies are integrated into existing business workflows. This stage often determines whether AI investments yield tangible results or become ineffective. It involves customizing AI models to fit an organization’s unique context, incorporating domain expertise, and establishing processes to ensure accuracy and compliance.

How can enterprises ensure successful AI integration?

Enterprises can ensure successful AI integration by focusing on three key areas: technology, domain expertise, and change management. Engaging with employees who have in-depth knowledge of operational processes helps to tailor AI systems effectively. Additionally, implementing robust governance and accountability structures can help manage the transition and maximize the value of AI investments.

What role does domain knowledge play in AI implementation?

Domain knowledge is critical in AI implementation as it informs how AI systems should be customized to reflect real-world workflows accurately. This expertise helps organizations identify specific needs and challenges, allowing them to develop AI solutions that are not only effective but also compliant with regulatory standards. Without this knowledge, AI systems may fail to deliver the expected outcomes.

How can organizations balance bespoke solutions and scalability in AI?

Organizations can balance bespoke solutions with scalability by keeping intelligence within the operational framework surrounding AI models. This approach allows for customization based on unique organizational needs while also leveraging open-source and frontier models for broader applications. By maintaining flexibility in how AI is integrated, enterprises can adapt to changing technologies and market demands effectively.