Why AI Delivery Refuses to Behave Like a Normal Project
Updated: Sep 1
1. Insight
AI initiatives are increasingly being delivered through established project management structures.
However, many project managers are finding that AI projects do not behave in the same way as traditional digital or business change projects.
Many AI developments do not fit neatly into waterfall or even agile delivery in the way many project managers are used to. It is often more fluid, experimental and iterative, with requirements, solutions and delivery paths becoming clearer as work progresses.
Requirements may not be fully defined at the outset. Solutions may not be fully understood before delivery begins. Understanding often develops as work progresses.
At the same time, project managers are still expected to:
Define scope
Build delivery plans
Manage timelines and resources
Provide confidence to stakeholders
This creates a tension.
In practice, the problem is that AI delivery often refuses to behave like a normal project. The work may still need structure, plans and assurance, but the conditions are more fluid than many project managers are used to, with understanding emerging progressively rather than being fixed upfront.
Experienced project managers are being asked to deliver outcomes while important uncertainties around requirements, feasibility and solutions are still being resolved.
This is not a failure of project management discipline.
It is a natural response to working in a delivery environment where certainty emerges progressively rather than being established upfront.
That is why AI projects can feel so different in practice. The challenge is not lack of discipline, but the need to apply discipline in a delivery environment that does not behave in a familiar way.
2. Why This Matters
AI delivery often operates under conditions that are less stable than traditional project environments.
This means AI projects can resist the normal assumptions of project delivery. Scope may be harder to stabilise early, feasibility may need to be proven during delivery and the route to a workable solution may only become clear through iteration.
Project managers may find that:
Requirements evolve as understanding improves
Feasibility needs to be proven during delivery
Solutions are shaped through iteration rather than defined upfront
Stakeholder expectations change as the opportunity becomes clearer
Without recognising this, there is a risk of:
Attempting to lock down scope too early
Interpreting iteration as loss of control
Applying delivery approaches that do not match the nature of the work
Underestimating the differences between types of AI implementation
In this context, maintaining delivery confidence becomes more challenging — even for experienced project managers.
This can make AI projects feel less predictable, even when they are being managed well.
3. What This Means
AI project delivery requires a shift in how control is established and maintained because the work does not always behave like a normal project.
The objective is not to remove uncertainty at the outset, but to maintain control while uncertainty is progressively reduced through delivery evidence.
The challenge is not to abandon project management discipline.
It is to apply it in a way that accommodates:
Evolving understanding of requirements
Emerging solution design
Iterative delivery approaches
Varying delivery modes with different risk profiles
In practice, this often means the project cannot be managed as a simple sequence of defined tasks moving toward a stable end-state. Instead, the project manager may need to manage evolving scope, emerging learning and shifting stakeholder understanding at the same time.
Without this adjustment, projects can become:
Difficult to plan and re-plan
Subject to ongoing scope tension
Misaligned with stakeholder expectations
Hard to baseline and control in the way stakeholders may expect
Technically delivered but not successfully adopted or used
In practice, this can create a delivery environment where multiple dimensions are evolving at once. Scope may still be forming, solution options may still be emerging, and stakeholder expectations may still be adjusting as understanding develops.
The role of the project manager becomes less about executing a fixed plan and more about maintaining control, alignment and momentum while uncertainty is progressively resolved.
4. Maintain Control
There are five practical steps that project managers can take to maintain delivery control and confidence.
4.1 Build AI Orientation
Project managers do not need deep technical expertise.
However, they do need enough orientation and understanding to:
Recognise different types of AI capability
Understand how they are typically applied
Assess relative complexity and risk
This enables more effective:
Planning
Stakeholder engagement
Expectation management
Explore this further in the Insight here: The Orr Consulting AI Universe — The Current AI Landscape.
4.2 Recognise the Delivery Modes
AI projects are not delivered in a single, uniform way.
In practice, project managers will encounter different delivery modes, including:
Adoption of generative AI tools (use-mode)
Configuration of AI-enabled vendor solutions (configure-mode)
Custom AI development (build-mode)
Each mode has:
Different levels of uncertainty
Different timescales
Different technical and delivery risks
Understanding these distinctions is critical to:
Selecting appropriate delivery approaches
Setting realistic expectations
Managing delivery effectively
Explore this further in the Insight here: AI Project Management — Progressive Project Management for AI Delivery.
4.3 Test Assumptions Early
AI delivery can depend on critical assumptions and dependencies that need to be tested early. Data availability, quality, accessibility and suitability are often among the most important.
In more traditional projects, data may be treated as an input to delivery.
In AI projects, data can determine whether the proposed solution is feasible at all.
This is especially important in custom AI development and in configuring AI-enabled vendor solutions, where delivery may depend on whether the required data exists, whether it is accessible, whether it is sufficiently reliable and whether it is suitable for the intended use case.
Project managers should ensure that critical assumptions and dependencies, including those relating to data, are tested early.
This helps avoid late discovery of feasibility issues and supports more realistic planning, risk management and stakeholder expectations.
4.4 Manage Uncertainty
AI projects still require effective initiation to establish governance, ownership, controls and an initial delivery plan. Where significant uncertainty remains, delivery can then progress through Discover, Design and Deliver stages and cycles, with understanding and commitment increasing as evidence develops.
These stages and cycles use evidence-producing deliverables to:
Refine requirements
Test assumptions and feasibility
Develop and validate solutions progressively
Support informed governance decisions
This does not replace project management discipline. It changes how planning and control are applied.
Governance checkpoints can then use the evidence produced to determine whether the project should continue, be refined, pivot or stop.
Project managers should expect:
Plans to evolve
Scope to be refined
Learning to be built into delivery
In AI delivery, iteration is not simply a response to change. It is often a mechanism for progressively resolving uncertainty and generating evidence about what will actually work.
This also means that decision points, control mechanisms and delivery assurance may need to be applied differently from more traditional project work.
Explore this further in the Insight here: AI Project Management — Progressive Project Management for AI Delivery.
4.5 Plan for Adoption
Even where AI solutions are technically successful, value depends on effective adoption, use and the realisation of intended benefits. This requires stakeholder understanding, trust in outputs and effective adoption of new ways of working.
Project managers should ensure that:
Changes are clearly communicated
Confidence in AI-supported decisions is built
Users are supported in adopting new ways of working
Changes to roles, decisions and ways of working are actively managed
Without this, projects may be delivered but fail to realise their intended benefits.
Explore this further in the Insight here: AI Benefits Realisation — Ensuring AI Delivery Translates into Measurable Value.
5. Benefits
When project managers adapt their approach to AI delivery, the impact is significant:
More realistic and adaptable delivery plans
Better alignment between expectations and delivery
Improved management of uncertainty and iteration
Stronger stakeholder confidence
Increased likelihood of successful adoption and value realisation
Projects become more resilient, better understood and more likely to succeed in practice
6. Risks
If AI delivery is approached using only traditional project assumptions, common patterns can emerge:
Unstable plans as requirements continue to evolve
Tension between fixed scope and emerging understanding
Late discovery that required data is unavailable, inaccessible or unsuitable
Misalignment between delivery outputs and stakeholder expectations
Underestimation of differences between AI delivery types
Solutions delivered but not adopted, trusted or used effectively
AI investments that fail to unlock their intended value
These challenges are often not caused by poor project management.
They arise from applying familiar approaches in unfamiliar conditions.
In other words, the problem is not simply that AI projects are difficult. It is that they often refuse to behave in the more predictable way that traditional project structures expect.
7. Final Thoughts
AI is becoming part of the project delivery landscape, but it does not always arrive in a form that behaves like normal project work. Requirements may move, solutions may emerge through iteration and delivery confidence may need to be maintained in conditions of greater fluidity than many project managers are used to.
Experienced project managers already have many of the skills required. The challenge is ensuring those skills are applied in a way that reflects how AI delivery actually behaves in practice.
The objective is not to replace project management discipline.
It is to apply it in a way that reflects how AI solutions are developed, tested and adopted.
By building AI orientation, recognising the delivery mode, testing critical assumptions and dependencies early, managing uncertainty progressively and planning for adoption and benefits, project managers can maintain delivery confidence while working with fluid requirements and evolving solutions.
This Insight is part of the Orr Consulting AI Transformation Library — structured thinking for AI transformation leaders and decision makers.
8. Call to Action
If AI delivery in your organisation is starting to feel harder to control than more traditional project work, the next step is not to force premature certainty onto it, but to strengthen the structure, planning, evidence and assurance around how AI projects actually behave.
If this Insight reflects your organisation’s experience, we would welcome a conversation.
Subscribe to Orr Consulting to receive occasional emails with practical AI Insights and updates.
