In a wide-ranging conversation with The AI Innovator, Bina Hallman, vice president of IBM Technology Lifecycle Services (TLS) shares crucial insights on developing effective AI and data strategies for business operations. Hallman emphasizes that organizations must view AI implementation not as an isolated initiative but within the broader business context.
This holistic approach ensures AI deployments directly address the most critical aspects of business transformation, creating measurable value while navigating the human, technical and strategic challenges of enterprise AI adoption.
What follows is an edited version of that conversation.
The AI Innovator: How should organizations define their AI strategy for operations to ensure alignment with their broader business goals?
Bina Hallman: You can’t look at AI individually, as a silo, but rather in the business context. Before starting this journey, leaders need to understand the organization’s end-to-end business processes and which areas of operations impact the business process the most. Once those are understood and agreed upon by the organization’s senior leaders, they need to define which aspects of the business processes require change or redesign to achieve the desired outcomes. These are the areas to focus on when building the AI strategy for the organization.
What are the key factors organizations should assess before deploying AI in operational workflows?
In addition to evaluating the overall business context, several aspects must be assessed. I would begin by emphasizing the human element. Integrating AI into any workplace represents a cultural shift. There is a widespread misconception that AI will inevitably replace human workers, but this is far from true. Achieving optimal outcomes requires bringing employees along on the journey. Technology can be daunting for non-technical teams, but training and reskilling can demystify it and foster a willingness to explore innovative approaches.
Providing this support has never been more crucial: In 2024, global CEOs estimated that, on average, 35% of their workforce would need to be reskilled, according to the ‘2024 CEO Study’ from IBM IBV. This is more than a billion workers around the globe. You have to educate these groups on how AI can benefit them by automating certain aspects of their daily tasks, thereby freeing up time to concentrate on complex, client-focused work and enabling them to acquire new skills.
Another key factor is ensuring that the organization has a robust data management strategy, covering how data is collected, stored, and governed – who has access, how it is routinely cleansed, and what the retention policies entail. Lastly, it is imperative to comply with all regulatory policies governing data and the industry and to adhere strictly to security and AI ethics guidelines. The ultimate goal is to incorporate AI into processes that will enhance efficiency and value for your operations and your clients as well.
How can businesses effectively measure and demonstrate the ROI of AI deployments in operations?
Our experience in support has highlighted two key dimensions for evaluation: First, AI must deliver business value measured by how effective the support and services processes have become and how fast the organization is delivering answers to clients and solving the clients’ issues. Prioritizing where to implement AI and measuring its impact with the right KPIs is necessary to understand its impact.
Second, AI requires continuous improvement; it is not a static tool or technology, but one that evolves through user feedback and data insights. Establishing a feedback loop that monitors these well-defined KPIs, learns from metrics and incorporates corrective actions is essential for refining the solution and delivering the expected business results. Additionally, client feedback is central to AI’s success since clients ultimately experience the impact of AI on products and services.
Many organizations struggle to move from pilots to full-scale AI deployment. What advice do you have for scaling AI solutions efficiently?
As clients begin implementing AI, the first practical step is to evaluate whether their data centers are ready for this type of technology. This entails enhancing IT infrastructure for power, cooling, network capacity for large amounts of data, and security while ensuring scalability. An effective implementation should prioritize operational efficiency, minimal downtime, prompt responses, and compliance with regulatory standards and ethical considerations.
And this is an intricate challenge – IBM IBV ‘5 Trends for 2025’ report shows that only 25% of executives strongly agree that their organization’s IT infrastructure can support scaling AI across the enterprise. Besides that, having a key partner with in-house AI expertise and the ability to manage the full lifecycle of this underlying infrastructure will help you leverage the benefits of such a technological evolution. For instance, much of IBM’s experience comes from implementing AI in IBM’s own processes and tools, which we bring to client engagements.
With the pace of AI innovation, how can companies future-proof their AI deployments to remain competitive?
AI is an incredibly fast-evolving field, with new models and capabilities emerging constantly, which is why you need to have a mindset of ongoing learning, changing and adapting. Today, the average organization uses 11 generative AI models and anticipates growing its model portfolio by around 50% within the next three years – as noted in IBM IBV’s The CEO’s Guide to AI Model Optimization report.
This expansion is driven by the fact that different tasks benefit from specialized AI models and capabilities. As I’ve mentioned above, the organization also needs to assess its infrastructure to make sure it is ready for AI implementation, considering quick scalability. The company will learn not just from the technology advancements, but from seeing where the opportunities are as it engages its internal teams and external clients.
This happens fast, so shifting rapidly to ensure the business is looking at the highest opportunity and properly investing in those areas is mandatory. Fundamentally, it is not about redeploying a solution overnight but adapting when needed, which will happen more times than you can try to predict.
Data silos and poor data quality are major hurdles. What practical steps should companies take to prepare their data for AI models?
AI is only as good as your data. A robust data strategy is paramount for the successful deployment of AI solutions because data engineering is the backbone of the AI framework. It starts with questions like how to cleanse, gather, search, vectorize and establish relationships between data. AI’s main strength is in its ability to comprehend complex data relationships, making informed decision-making essential. This involves ensuring that data is utilized optimally, procured efficiently, and remains accurate and manageable. Such a strategy is instrumental for aspects ranging from system integration and security compliance to scalability and cost optimization.
What are the key considerations when integrating AI solutions with existing legacy systems in operations?
Integrating AI solutions with existing legacy systems can present several challenges and requires careful consideration to ensure a smooth and effective transition. Compatibility is a primary concern, as I’ve highlighted the importance of the infrastructure for this digital transformation. With that, other concerns will come, such as managing data integration, maintaining system performance, ensuring security, planning for scalability, handling change management, considering costs, complying and leveraging the regulatory requirements, preparing your team to ensure you have the necessary expertise, conducting rigorous testing, future-proofing the solution and securing user acceptance.
How can organizations effectively transition AI models from development to production, and what monitoring practices are essential post-deployment?
Organizations need to start evaluating the use case scenario that would be a good fit to leverage an AI model. Normally, this process requires a team that can articulate the business value, ensuring the architecture and high-level design comply with the company’s technical guidance and security standards. The next step is to identify the roles and responsibilities for an AI project implementation. There may be multiple AI projects being looked at simultaneously, and prioritization based on ROI will be required, therefore a cross-functional team is essential here, with executive, leadership and execution roles.
The AI project implementation involves four key phases: requirement identification, planning and execution, resource management, and KPI/metrics monitoring. Each phase requires specific roles for successful completion. Post-deployment, continuous KPI monitoring is vital. This practice allows not only ongoing evaluation of the AI solution’s performance, effectiveness and impact on business goals but also adjustments to ensure the solution maintains alignment with operational targets, elevating the business value.
What practical steps can organizations take to identify and mitigate bias and other harms in AI systems deployed in operations?
Mitigating bias and harm in AI is an ongoing process, not a one-time task, much like the AI implementation. It requires continuous effort, monitoring and adaptation for improvement. There will hardly be a finish line, but organizations can take several practical actions to identify and mitigate it, such as ensuring that the data used to train AI models is diverse and representative, by conducting regular audits of AI systems and using transparent AI models whenever possible as well.
Establishing clear and ethical guidelines for AI development and use, feedback mechanisms and involving diverse stakeholders in the technology development and decision-making processes are also mandatory steps. Always stay updated on laws and regulations related to AI ethics and fairness, ensuring the systems comply with all the most up-to-date requirements.
For organizations with limited budgets, how can they prioritize investments in AI technologies for maximum operational impact?
A company does not necessarily need to develop its own AI solutions, which require deep expertise in AI. Instead, leveraging products such as IBM watsonx from industry leaders offers a more efficient and faster approach. Ultimately, an AI project must provide business value to an organization, such as providing financial savings, bringing efficiency or enhancements to operations that would result in operational savings, and/or improving customer experience that would potentially drive revenue growth.
With limited budgets, look for low-hanging projects in different departments such as operations or customer support that can be implemented relatively quickly and offer clear business value for the company and its clients. Customer support, for example, is a frequent area that offers a discernible ROI. Businesses that are currently utilizing generative AI in customer service generally report higher customer satisfaction, according to IBM IBV. Notably, these organizations will achieve superior overall business outcomes.
What training or upskilling programs does IBM recommend for teams to successfully implement and maintain AI solutions in operations?
IBM has an internal training portal available to all its employees. This portal provides AI-specific technical education, internal tooling for hands-on experience leveraging diverse AI tools and learning paths leading to certification. I recommend starting by examining people’s existing AI skills, determining what skills are needed, and planning to implement training and upskilling to fill that gap. Some of the new skills that are crucial for AI implementation may include data literacy, AI and machine learning fundamentals, AI ethics, sense and fact check for AI outputs, proprietary information, gen AI tools, and data science.
Companies should plan to incorporate formal and informal learning methods to include mentorship, on-the-job learning, and formal classes, as well as encourage AI knowledge sharing. Depending on the amount of change a company is expecting that will be brought about by using AI in operations, they may also want to wrap some change management around all of this.
What are some of the most common issues you have seen in AI deployments within operations, and how can organizations proactively address them?
Due to the lack of early collaboration with users, some companies struggle to gather requirements, which hinders their readiness and receptiveness toward adopting AI solutions. The omission of KPIs from the outset, along with the absence of automation tools for result collection and analysis, may significantly undermine the success of AI implementations. It is essential to have a comprehensive end-to-end strategy for AI and to communicate it to every department from the beginning. For large-scale companies, creating synergies across various teams while avoiding duplicated efforts can be challenging, as each department aims to progress rapidly.
However, it is crucial to conduct design and architecture sessions for similar use cases across departments. The rules and regulations regarding data security and storage apply to all units. Not having the right skills for implementation is another concern. Everyone involved must undergo mandatory training to establish a common understanding of the basic concepts and methods. Depending on roles and responsibilities, upskilling will be required. This should not be a one-time training, as AI models are continually evolving.
Additionally, a technical support team will likely be involved in solution development, requiring both AI expertise and process and program management skills. It is important to have a clear understanding of the appropriate use cases that could benefit from leveraging AI. As such, strategy, design and architecture teams should evaluate the entire process and identify where AI should be implemented to create business value.
Clearly defining roles and responsibilities for all involved parties is necessary since there are many different roles, each requiring the description of expectations and timelines. Management tools must be established right from the start, ideally utilizing a similar tool across all departments to allow teams to easily access and share learnings while tracking progress. By addressing these areas proactively, organizations can improve AI deployment success and avert potential problems.
Be First to Comment