AI projects need a new management approach

Traditional project management methods fail for AI projects. The issue goes beyond adjusting timelines or budgets—it reflects a fundamental difference in how the work progresses.

AI development operates continuously, relying on data and iteration. Unlike conventional IT projects with fixed start and end dates, AI initiatives often run indefinitely. The focus moves from building an application to refining data and models over time. If the data contains errors, the results will too, regardless of how well the code is written.

This change alters leadership roles. Data specialists, rather than application developers, become the key team members. Business users, especially subject-matter experts, must also play an active part in verifying data accuracy. Measuring progress becomes difficult because AI projects evolve instead of concluding. A project manager used to Gantt charts and milestones may struggle without a clear endpoint.

Choosing the right AI model strategy

The first step involves defining the business use case and expected outcomes. The next challenge is selecting the appropriate model development approach.

An AI model is a program trained on data to recognize patterns or make decisions without human input, as described by IBM. The method varies based on the business need. For financial forecasting using existing company data, predefined algorithms may work. For cancer diagnosis, the system might need to learn from global datasets to improve accuracy. Companies without in-house expertise, such as those in customer service, can use prebuilt foundation models and adjust them over time.

Organizations must decide whether to build models from scratch, use prebuilt ones, or combine both. The decision depends on the problem’s complexity, available data, and internal skills. A poor choice can waste resources or produce unreliable results.

Infrastructure and team readiness

Before starting an AI project, companies need to evaluate their IT infrastructure and staff expertise. If existing systems can’t handle the data and processing demands, cloud hosting may be necessary, assuming the budget allows it.

Data analysts excel at cleaning and preparing data, but AI model development requires additional skills in algorithm design and statistical analysis. Data scientists often fill this gap, though traditional IT developers may lack these capabilities. End users, particularly subject-matter experts, must also participate in ongoing model training.

Collaboration between IT and business users must continue long-term. Unlike traditional IT projects that end once deployed, AI initiatives require ongoing oversight. This shift demands changes in how teams work together.

Deploying AI without disruption

The best way to introduce AI into production is gradually. Instead of automating entire workflows at once, companies should target specific steps. This approach minimizes disruption to employees and reduces the risk of failure. It also provides time to identify and fix issues before they escalate.

Human oversight remains essential. AI systems can generate incorrect results if trained on flawed data. Keeping people involved ensures accountability and maintains trust in the system.

The Project Management Institute has outlined a methodology for AI projects, dividing them into six phases: business understanding, data understanding, data preparation, model development, model evaluation, and model operationalization. However, practical guidance remains scarce. Most project management tools still treat AI projects like traditional IT efforts, forcing CIOs and project managers to adapt on their own.

Accountability is critical. Someone must lead the project, update both the team and leadership, and make key decisions. AI projects may never truly finish—at least not until the business use case becomes outdated. Teams and executives must accept this from the beginning.

Training should be part of the project plan. Early AI efforts may produce mixed results, but failures offer valuable lessons. Leadership must recognize that success won’t happen immediately. Small, well-defined use cases with clear goals are the safest starting point.

For now, those leading AI projects carry the responsibility. The methods are still developing, and tools haven’t caught up. What’s certain is that AI requires a flexible approach that values data, long-term collaboration, and gradual progress over rigid deliverables.