Managing AI Projects: Why Traditional IT Methodologies Fall Short

Artificial intelligence projects require a fundamentally different management approach than traditional IT initiatives because they are iterative, data-centric, and evolutionary rather than finite. Unlike standard software development, which often follows a linear path to a final release, AI projects rely on continuous data refinement and model training, necessitating a shift in how CIOs and project managers allocate resources and define success.

The Shift from Application Development to Data-Centricity

Traditional IT project management centers on application development, where success is measured by the delivery of functional code. According to the Project Management Institute, AI development instead follows a lifecycle that begins with business and data understanding, followed by data preparation, model development, evaluation, and operationalization.

This shift changes the primary project stakeholders. Because the quality of an AI system depends on the quality of its underlying data, data specialists—rather than application developers—often become the primary go-to people for AI projects. This transition forces subject-matter experts in end-user departments to take a more active role in IT projects, as they are now responsible for determining whether the data is correct.

Evaluating AI Model Development Strategies

Choosing the right development strategy is a challenge in managing AI projects. IBM defines an AI model as “a program that has been trained on a set of data to recognize certain patterns or make certain decisions without further human intervention.” Organizations must decide whether to build a custom model or use prebuilt foundation models.

The strategy depends on the business use case:
* Query-based algorithms: Suitable for internal “what-if” scenarios and financial forecasting using existing company data.
* Machine learning systems: Required for complex tasks like improving cancer diagnosis, where the system must learn from external, worldwide symptoms and data to embellish what is known locally.
* Foundational systems: Ideal for areas where a company lacks internal expertise, such as customer service, allowing the company to customize a preconfigured model over time.

Infrastructure and Staff Readiness

Before launching an AI initiative, CIOs must assess both technical infrastructure and team skills. If on-premises hardware cannot support the intensive processing required for AI workloads, organizations often move to cloud-based environments to scale resources.

A significant gap often exists between the skills of IT developers and the requirements of AI model development, which demands expertise in statistical analysis and algorithm development. Furthermore, model training is not a one-time task. It requires vigilant, ongoing oversight by subject-matter experts to prevent “drift,” where a model loses contextual accuracy over time.

Implementing Gradual Workflow Changes

To minimize disruption, the goal should be automating selected workflow steps rather than entire business processes. This gradual approach allows organizations to manage the impact on human job responsibilities, which can otherwise upset users and derail project progress and trust.

Maintaining a “human-in-the-loop” is essential for mitigating risks such as AI hallucinations or errors resulting from skewed or biased training data. By keeping human oversight at critical decision points—such as triggering a data center failover—CIOs can ensure that AI acts as a tool for efficiency rather than a source of unmanaged risk.

Key Considerations for AI Project Success

* Accountability: Clear leadership is as vital in AI as in traditional IT; one individual must remain responsible for project progress and communication with C-level management.
* Iterative Nature: Stakeholders must accept that AI projects are often evolutionary and may not end in the conventional sense.
* Continuous Learning: Project schedules should explicitly include time for IT and end-user education and training to ensure teams can adapt to the shifting nature of AI performance.
* Scope: Starting with small, tightly constructed business use cases with clear and achievable goals increases the likelihood of early success and provides a foundation for scaling.

As the methodology for AI project management continues to evolve, CIOs must navigate the current lack of specialized software tools by emphasizing transparency, realistic expectations, and deep collaboration between IT and business units.