AI Project Management — Applying Progressive Project Management in Practice
Updated: Sep 7
1. Insight
Artificial Intelligence projects require established project management disciplines to operate effectively in conditions where important aspects of the future solution may not yet be known.
The challenge is not simply that AI projects contain uncertainty. Most projects do. The challenge is that uncertainty around feasibility, data, technology, operating processes, human involvement, governance, performance, adoption and value may remain material well into delivery.
Orr Consulting’s Progressive Project Management approach uses that uncertainty to structure the project.
Material uncertainties are identified explicitly in an Uncertainty Log. The project determines what evidence is required to reduce, bound or explicitly accept them and builds tangible evidence-producing deliverables into the plan.
Those deliverables are assessed against clear acceptance criteria and used by governance to determine whether further commitment remains justified.
The underlying logic is:
Uncertainty determines what the project must learn
Evidence-producing deliverables create that learning
Acceptance establishes whether the evidence is sufficient
Governance determines whether the evidence justifies further commitment
Progressive Project Management operates through an explicit project lifecycle:
Initiation → Discover → Design → Deliver
Initiation — establishes sufficient control to begin Discover
Discover — determines what may be viable and seeks to converge on a credible candidate proposition
Design — develops that proposition into an approved candidate solution that current evidence indicates should work and is sufficiently defined to justify implementation
Deliver — implements and tests that candidate to determine whether it actually works sufficiently under the proposed operating conditions
The objective is not to create artificial certainty. It is to progressively reduce, bound or explicitly accept material uncertainty before increasing commitment.
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, Progressive Project Management provides the practical structure for progressing through Initiation → Discover → Design → Deliver.
2. Why This Matters
AI projects can create pressure at two extremes.
At one extreme, organisations may attempt to specify the complete solution before enough evidence exists to support it.
At the other, uncertainty can become an excuse to begin building without sufficient convergence, leaving important design, operating-model and governance questions to be resolved during implementation.
Neither position provides strong project control.
Progressive Project Management instead treats delivery as a convergence funnel.
The project begins with a range of possible answers and progressively narrows them as evidence increases:
Many possibilities → Credible candidate proposition → Approved candidate solution → Accepted operating capability
Each stage must therefore earn the next level of commitment.
If Discover cannot produce a sufficiently credible candidate proposition, progression to Design should not be automatic.
If Design cannot converge on a sufficiently evidenced candidate solution, Deliver should not proceed simply because the planned Design stage has ended.
Further evidence, targeted Discover, refinement, pivot or stopping may all be legitimate governance outcomes.
This creates a simple principle:
Time spent does not earn progression. Evidence does.
It also gives Project Managers a practical way to reconcile two competing requirements.
Technical and delivery teams need sufficient freedom to explore, iterate and learn.
Governance needs sufficient convergence to determine what it is actually being asked to fund, implement and accept.
Progressive Project Management provides both.
3. What This Means
Progressive Project Management begins by making material uncertainty explicit.
Orr Consulting typically applies a generic uncertainty lens across eight project themes/workstreams:
Process — current state and future operating process
Organisation — roles, resources, capabilities, responsibilities, ownership and support
Technology — solution architecture, allocation between AI, rules and humans, infrastructure and integration
Information — information architecture, data viability, reporting and management information
AI Governance & Assurance — accountability, controls, monitoring, assurance and continuing governance
Business Change, Readiness & Benefits — value, adoption, organisational readiness and transition
Testing, Acceptance & Assurance — how the capability will be tested and what evidence will demonstrate that it satisfies the required standard
Project Management & Governance — whether the project itself remains deliverable within its authorised constraints
These provide a structured starting point rather than a mandatory checklist.
Individual projects will create additional uncertainties arising from their organisational, regulatory, commercial, supplier, sector and technology context.
For each material uncertainty, the project should determine:
what evidence is required
what tangible deliverable or supporting evidence will provide it
what acceptance criteria will establish whether the evidence is sufficient
which other workstreams must review or accept material interfaces and dependencies
what assumptions, dependencies, risks, materialised issues and residual exposure must also be controlled through RAID
what governance decision the evidence will support
The purpose is not to create a separate document for every uncertainty.
Related uncertainties should be consolidated into a manageable set of evidence-producing deliverables that provide the evidence required across a workstream or stage.
A single deliverable may resolve several uncertainties.
Equally, an accepted deliverable does not require a new successor deliverable at every later stage. Where sufficient evidence has already been established, that deliverable may become an approved control baseline for subsequent work.
New or revised deliverables should be created where further uncertainty requires additional evidence — not simply because the project has entered another stage.
For example, evidence-producing deliverables might include a validated problem baseline during Discover, an approved candidate operating model or technical solution during Design, and implemented capability, testing and acceptance evidence during Deliver.
The specific deliverables should remain proportionate to the project, its delivery mode and its material uncertainties.
Progressive Project Management plans and governs these tangible outputs — the what — while allowing delivery teams proportionate flexibility over the workshops, analysis, technical spikes, prototypes, configuration, development and other activities used to produce them — the how.
4. In Practice
Progressive Project Management applies across Use, Configure and Build delivery modes.
The underlying uncertainty-management method remains the same.
What changes is where uncertainty is concentrated, the depth of evidence required and the level of commitment involved.
4.1 Initiation
Initiation establishes the conditions required to begin structured Discover.
Typical project controls include:
Project Governance Board Established
Approved Project Initiation Document and Initial Project Plan
Workstream Leads Appointed and Workstreams Established
Initial Uncertainty Log Established
RAID Controls Established
Project Reporting and Control Framework Established
The PID does not need to contain every eventual answer.
It should establish the project, identify its initial material uncertainties and provide a credible evidence-producing plan for beginning Discover.
The Initiation Gateway asks:
Is the project sufficiently defined, governed, resourced and controlled to begin Discover?
4.2 Discover
Discover establishes what may be viable.
It validates the problem and current-state baseline, investigates important uncertainties, develops the emerging future operating capability and tests whether a credible candidate proposition can be formulated.
Evidence may include:
current-state and process analysis
use-case and value assessment
information and data assessment
technical feasibility work
prototypes or technical spikes
organisational and operating-model analysis
governance and assurance assessment
change and benefits baselines
initial testing and acceptance strategy
Discover does not need to prove that the eventual solution will work.
It needs to establish whether sufficient evidence exists to justify further commitment to a credible proposition.
The Discover Gateway asks:
Is there a sufficiently evidenced and coherent candidate proposition to justify Design?
If not, additional targeted Discovery, refinement, pivot or stopping may be appropriate.
4.3 Design
Design develops the credible proposition into an approved candidate solution.
The word candidate is important.
An approved design is not a declaration that every technical or operational uncertainty has disappeared.
It represents the best sufficiently evidenced and governed solution baseline for the level of implementation currently being authorised.
Design should establish sufficient convergence across Process, Organisation, Technology, Information, AI Governance & Assurance, Business Change, Readiness & Benefits and Testing, Acceptance & Assurance to determine what should work.
Material workstream deliverables should also contain evidence of appropriate inter-workstream review and acceptance.
Clear workstream ownership should not create siloed delivery.
Process decisions may affect technology. Technology choices may create information requirements. Governance controls may alter operating processes. Organisational arrangements may affect testing, adoption and benefits.
The project therefore needs evidence that the individual workstream designs form a coherent future operating capability.
The Design Gateway asks:
Is the integrated candidate solution sufficiently evidenced, coherent, controlled and defined to justify implementation at the proposed scope?
4.4 Deliver
Deliver implements the approved candidate and determines what the evidence actually demonstrates.
The operating process, technology, information capability, organisational arrangements and governance controls are implemented and tested together.
Testing should establish not simply whether individual technical components function, but whether the integrated capability performs sufficiently under representative operating conditions.
Deliver therefore follows an evidence cycle:
Implement → Test → Operate → Observe → Evidence → Accept
Evidence should be assessed against the acceptance criteria established during Design.
Where the candidate does not perform sufficiently, refinement remains legitimate.
Deliver is not the mechanical execution of an answer proven during Design.
It is the point at which the approved candidate is exposed to increasingly representative operational evidence.
Governance can then determine whether the evidence supports continued delivery, refinement, transition, further investment or stopping.
Across every stage, the principle remains the same:
Produce sufficient evidence before increasing commitment.
5. Benefits
Applying Progressive Project Management in practice provides:
Greater visibility of uncertainty — material unknowns become explicit and traceable
More purposeful project plans — uncertainties are translated into evidence-producing deliverables
Reduced deliverable proliferation — related uncertainties are consolidated into proportionate outputs
Better integration — workstream ownership is supported by cross-workstream review and acceptance
Stronger governance decisions — progression is based on accepted evidence rather than elapsed time or completed activity
The result is a more controlled route from uncertainty to evidence, decision and justified commitment.
6. Risks
Poor application can weaken the approach.
Key risks include:
False certainty at Design — treating an approved candidate solution as proof that it will work
Deliverable proliferation — creating separate documents for every uncertainty rather than consolidating evidence proportionately
Workstream silos — allowing clear ownership to fragment the integrated future capability
Weak acceptance criteria — producing outputs without sufficient evidence that they are fit to support governance decisions
Analysis without convergence — continuing investigation without reaching the decisions required to progress, refine, pivot or stop
Project Managers must therefore apply the approach proportionately, maintaining enough structure to create evidence and control without introducing unnecessary process.
7. Final Thoughts
AI projects do not need a fundamentally new form of project management. They need established disciplines applied to an environment where important answers emerge progressively.
Progressive Project Management provides that structure:
Initiation establishes control
Discover identifies what may be viable
Design establishes the candidate that should work
Deliver determines whether it actually works sufficiently
Throughout this progression, uncertainty shapes the evidence required, evidence-producing deliverables structure the plan, acceptance criteria determine sufficiency and governance decides whether further commitment remains justified.
Acceptance does not remove all uncertainty. Where relevant, residual uncertainty should transition into ongoing monitoring, assurance and revalidation.
The objective is simple: each increase in commitment should be supported by sufficient accepted evidence.
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.
