Project Management🇬🇧 English
Project Management of AI Implementations
July 2025·7 min read·Brian P.N. Tofft
AI implementations fail more often than they succeed, and the reason is rarely technical. It is the project management that falls short. AI projects place particular demands on governance, stakeholder management and risk management that differ significantly from traditional IT projects. At We Lead Projects, we have led AI implementations from concept to operations. Here are the key lessons.
In a traditional IT project, the deliverable is defined upfront: a system that does X, a feature that does Y. In an AI project, the deliverable is inherently probabilistic. The model provides likely answers, not guaranteed ones. This means success criteria must be defined differently (what is an acceptable accuracy level?), testing and validation are more complex, and stakeholders must be prepared to work with AI output that is not always deterministic. A project manager who does not understand this difference risks steering towards a goal that does not exist.
An AI project typically moves through six phases: (1) Problem definition and business anchoring: what should the AI solve, and who owns the business benefit? (2) Data assessment and preparation: do we have the data required, at the quality required? This phase is consistently underestimated and accounts for more than half of all delays. (3) Proof of Concept: technical validation that the approach works on a limited dataset. (4) Pilot implementation: deployment in a controlled environment with selected users and structured feedback. (5) Scaling and integration: full rollout with integration into existing systems and workflows. (6) Operations and continuous improvement: monitoring model performance, retraining, and ongoing evaluation. Many projects attempt to jump from phase one to phase five. That is the shortest route to failure.
Beyond the classic project risks (scope creep, budget overrun, resource shortages), AI projects carry a set of specific risks: Data quality risk: the model is only as good as the data it is trained on, and data quality often proves far lower than assumed. Model degradation: models deteriorate over time as data and behaviours change, requiring processes for continuous monitoring and retraining. Bias and fairness risk: the model systematically discriminates against certain groups, with potential legal and reputational consequences. Regulatory risk: the EU AI Act classifies certain AI applications as high-risk with heightened requirements for documentation and control. Adoption risk: users do not trust or engage with the AI solution, and the business benefit is never realised.
AI projects typically involve a broader and more heterogeneous stakeholder group than traditional IT projects: business owners expecting rapid returns, IT architects concerned with integration and governance, legal and compliance teams managing regulatory requirements, end users who must change their workflows, and leadership who approved an investment and expect ROI. A competent project manager maps these stakeholders early, understands their individual concerns and success criteria, and communicates differently to each group. Particularly important is managing early concerns among end users. AI is often perceived as a threat to existing roles, and this resistance can torpedo even technically successful implementations.
Based on our experience with AI implementation projects, five factors consistently separate those that succeed from those that do not: (1) A clear and bounded business problem as the starting point, not AI for AI's sake. (2) A solid and well-documented data foundation established before model development begins. (3) A strong executive sponsor with the mandate and interest to realise the business benefit. (4) A cross-functional project team combining business understanding, data expertise, and technical implementation capability. (5) A realistic timeline that respects the fact that AI projects require iteration and experimentation, and that data always takes longer than planned.
We have supported organisations with AI implementations across all phases, from the initial business case and roadmap through to delivery and transition to operations. Our approach combines strong project management expertise with a practical understanding of what AI projects require: time for data validation, space for experimentation in the PoC phase, and close stakeholder involvement throughout. Whether you are assessing whether AI is the right solution, or already have a project in need of professional governance, we can help. Contact us for an informal conversation.
Brian P.N. Tofft
Managing Partner, We Lead Projects
Brian has more than 30 years of experience in project management and IT transformations across industries.
Get in touch and find out how we can help with your next project.
Contact us