AI Project Management — Progressive Project Management for AI Delivery
Updated: Sep 1
1. Insight
Artificial Intelligence delivery at project level behaves differently from traditional technology or business change delivery.
While AI programmes are governed through established programme management disciplines, the actual delivery of AI solutions takes place through projects and delivery teams operating in conditions of higher uncertainty, iteration and learning.
Project Managers, delivery leads and technical teams are therefore required to adapt how delivery is structured and managed in practice.
The fundamentals of good project management still apply. The defining difference is that AI projects operate with heightened uncertainty.
Orr Consulting’s Progressive Project Management approach makes material uncertainty explicit and uses it to shape the project. The project identifies what it does not yet know, defines the evidence required and builds tangible evidence-producing deliverables into the plan.
Those deliverables are governed through clear acceptance criteria and decision points. Established RAID controls remain essential, but their role is to manage assumptions, dependencies, risks, materialised issues and residual exposure affecting the plan—not to substitute for a plan designed to create evidence.
Progressive Project Management applies across three primary AI delivery modes:
Use-mode — adopting existing AI tools directly
Configure-mode — configuring and integrating a pre-built AI product
Build-mode — developing a bespoke AI solution
The same uncertainty-led logic applies across all three modes, although the density and distribution of uncertainty will differ. Delivery mode changes where uncertainty is concentrated—not whether it should be recognised and progressively reduced.
Within the Orr Consulting AI Transformation Process, AI Project Management is principally a Deliver-stage discipline and operates within the structure and governance established through AI Programme Management. Within individual projects, however, project delivery begins with Initiation before progressing through Discover–Design–Deliver.
2. Why This Matters
AI projects often begin before important aspects of the solution, feasibility, operating model, adoption and value are fully known.
Agile structures can support iteration, but sprints and backlogs do not by themselves manage uncertainty. Progress depends on identifying the material questions, determining what evidence would answer them and structuring delivery around tangible outputs that produce that evidence.
The concentration of uncertainty differs by delivery mode:
In Use-mode, it often concerns use-case value, governance, user behaviour, adoption and benefits
In Configure-mode, it often concerns product suitability, information readiness, configuration, integration and operational fitness
In Build-mode, it often concerns data, architecture, technical feasibility, performance and maintainability
AI project delivery should therefore increase commitment progressively as evidence increases and uncertainty is reduced, bounded or explicitly accepted.
3. What This Means
AI does not require a new project management discipline. It requires established discipline to be applied in a way that recognises incomplete knowledge and progressive learning.
Material uncertainties should be recorded explicitly in an Uncertainty Log established alongside the project’s other core controls. For each material uncertainty, the project should determine:
what evidence is required
what tangible deliverable will produce that evidence
what acceptance criteria will establish whether the evidence is sufficient
what assumptions, dependencies, risks, materialised issues or residual exposure must also be controlled
what governance decision the evidence will enable
The preferred response to uncertainty is therefore not simply to record it in RAID.
It is to design the project so that the plan produces the knowledge required.
The core uncertainty-management lifecycle is:
Recognise material uncertainty
Define the evidence required
Build evidence-producing deliverables into the plan
Assess and accept the evidence
Manage residual exposure through RAID
Decide whether to continue, refine, pivot or stop
Progressive Project Management plans and governs tangible deliverables — the what — rather than attempting to micromanage every workshop, meeting or activity used to produce them — the how.
Reality 1 – Projects Begin with Questions
Requirements, assumptions and solution options may emerge progressively.
Initiation should therefore establish sufficient control to begin Discover. Discover should then define the problem, intended value, material uncertainties and evidence requirements rather than presenting an assumed solution as certain.
Discover should establish what may be viable and which options justify more detailed design.
Reality 2 – Design Sets Delivery Baseline
Discover identifies credible options. Design converts the preferred approach into an agreed and governed baseline for delivery.
Depending on the delivery mode, this may include:
An Approved Operating Model and Functional Design
A Technical Options and Architecture Viability Assessment
An Approved Technical Solution Design
Information, governance and control designs
An Approved Minimum Viable Capability and Release Baseline
These are not administrative documents. They are primary uncertainty-reduction controls. They establish the governed basis for configuration, build, testing and change.
The minimum viable capability should represent the smallest coherent, governed and testable implementation capable of producing sufficient evidence about end-to-end feasibility, operational fitness and value.
Reality 3 – Technology Alone Isn’t Value
A functioning tool, configured product or working model is not automatically a viable organisational solution.
Projects must test:
Whether the integrated solution performs sufficiently
Whether controls work in practice
Whether users can operate and trust it
Whether exceptions and support requirements are manageable
Whether adoption produces measurable organisational value
Deliver should therefore produce evidence that the solution actually works, can be operated sustainably and creates sufficient value—linking project delivery directly to AI Benefits Realisation and the decision on whether further commitment is justified.
4. Delivery Modes and Cycles
AI delivery does not occur in a single uniform way. Orr Consulting distinguishes between Use, Configure and Build modes.
A project may contain more than one mode, but identifying the dominant profile helps the Project Manager understand where material uncertainty is likely to be concentrated and how deeply evidence must be developed.
Following effective project initiation, the same Discover–Design–Deliver logic applies across all three modes.
The depth of analysis, strength of evidence and level of governance should remain proportionate to the project’s material uncertainty, impact and exposure.
4.1 Use-Mode Delivery
Use-mode enables people to adopt existing AI tools directly in day-to-day work. Technical implementation may be limited, but uncertainty can remain high around governance, user behaviour, information handling, adoption and benefits.
Discover — identifies and validates valuable and appropriate use cases, affected users, risks and constraints
Design — establishes guidance, guardrails, workflows, training and assurance controls
Deliver — enables access, supports users, monitors behaviour and tests whether adoption creates safe and measurable value
The central question is whether the organisation can use existing technology effectively, responsibly and beneficially.
4.2 Configure-Mode Delivery
Configure-mode uses a pre-built AI product or platform and adapts it to organisational requirements, data, knowledge, content, workflows and systems.
Discover — assesses product suitability, information readiness, integration constraints and operating requirements
Design — establishes the operating model, configuration, information inputs, workflows, integrations and controls
Deliver — configures and integrates the product, tests its behaviour and confirms whether it is fit for operational use
The central question is whether an existing product can be adapted sufficiently to meet the organisation’s process, information, governance and value requirements.
4.3 Build-Mode Delivery
Build-mode creates a bespoke AI-enabled solution around organisational data, workflows and requirements. It usually carries the broadest and most interconnected uncertainty, particularly across technology, information and delivery.
Discover — tests the problem, data, feasibility and credible technical options
Design — establishes the approved operating model, technical solution, controls, testing approach and minimum viable capability
Deliver — builds, integrates, tests and pilots the approved capability before any decision to transition, scale or invest further
The central question is whether a viable, valuable and governable solution can be created and operated at all.
Across all delivery modes, Initiation establishes sufficient control to begin Discover. Discover determines what may be viable. Design establishes the approved baseline for delivery. Deliver produces evidence that the integrated capability actually works, can be operated sustainably and creates sufficient value.
5. Benefits
Effective AI project delivery is characterised by:
material uncertainty being made explicit and prioritised
evidence-producing deliverables with clear acceptance criteria
deliverable-based plans focused on tangible results
proportionate use of established RAID controls
effective project initiation followed by progressive Discover–Design–Deliver delivery
governance decisions based on accepted evidence and residual uncertainty
a clear link between solution delivery, adoption and measurable value
Project Managers maintain the structure, accountability and decision discipline required to enable iteration without allowing uncontrolled continuation.
6. Risks
Projects that do not adapt to AI uncertainty commonly experience:
detailed plans built on weak assumptions
material uncertainty hidden across conventional controls
late discovery of feasibility or fitness problems
governance reporting activity rather than decision-useful evidence
continued investment after the original case has weakened
technically successful outputs that are not adopted, operable or valuable
These outcomes reflect a mismatch between the delivery approach and the nature of AI work.
7. Final Thoughts
AI does not replace the fundamentals of project management. It changes how those fundamentals are applied when feasibility, operating requirements and value must be discovered progressively.
Progressive Project Management makes material uncertainty visible and uses it to shape the project:
Uncertainty determines what the project must learn
Evidence-producing deliverables are built into the plan to create that learning
RAID manages assumptions, dependencies, threats, materialised issues and residual exposure
Governance uses the resulting evidence to decide whether further commitment remains justified
The purpose is not to eliminate all uncertainty. It is to reduce, bound or explicitly accept it until the organisation has sufficient evidence to continue, refine, pivot or stop.
For experienced Project Managers, confidence does not come from believing the AI solution will work. It comes from knowing that the project is structured to find out.
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 your organisation is undertaking AI delivery and would benefit from stronger project control, clearer uncertainty management and more evidence-led delivery decisions, we would be pleased to discuss your next AI steps.
Subscribe to Orr Consulting to receive occasional emails with practical AI Insights and updates.
