From the Illusion of Control to Project Foresight: AI-Enhanced Project Controls within the PRINCE2 Method

A practitioner perspective on forecast reliability, early risk visibility, and Project Board decision-making in complex EPC and offshore environments

By Felipe Silva de Oliveira, PRINCE2 Ambassador and Emilly Oliveira, PRINCE2 Practitioner

Introduction

The problem with many complex projects is not always a lack of control. It is the illusion that control still exists.

In complex Engineering, Procurement, and Construction (EPC), offshore, brownfield, shutdown, and decommissioning environments, projects rarely lose control in one single moment. More often, they drift quietly. A late engineering deliverable is treated as manageable. A cost assumption remains open for one more reporting cycle. A known risk has no strong owner. A decision is moved to the next meeting. The schedule still looks credible. The dashboard has not alarmed anyone yet.

This is often how predictability starts to erode.

In this article, project controls refer to the integrated management of schedule, cost, risk, scope, progress, change, and evidence used to support project decisions. I explore how artificial intelligence (AI)-enhanced project controls can support the PRINCE2 method by improving forecast reliability, increasing early risk visibility, and strengthening Project Board decision-making.

My perspective is based on practical experience as a Project Manager in offshore projects, Floating Production, Storage, and Offloading (FPSO) operations, brownfield execution, shutdown planning, and decommissioning governance. In these environments, I have learned that the strongest project controls are not those that produce the most reports. They are the ones that help leadership recognize weak signals early enough to act.

When a project looks controlled but is not

In FPSO decommissioning, a project can look ready on paper while still not being ready in practice.

The asset may have reached the yard. The dismantling sequence may be visible in the schedule. The cost forecast may still appear acceptable. Engineering, procurement, waste management, environmental compliance, and work package readiness may each have their own reports.

The issue is that these control streams are often reviewed separately.

Schedule progress may not be directly connected to waste readiness. Cost exposure may not be linked to contractual responsibility. Environmental constraints may not be clearly tied to the dismantling start date. The risk register may capture several concerns, but it may not show their combined effect on the forecast.

That is where control becomes fragile. One deviation may not look material. Several small deviations, viewed together, can change the project outlook.

As Project Manager, the real question is not only, “Is the activity on schedule?” It is also, “Are the assumptions behind the forecast still reliable?”

That difference matters. There is a major difference between monitoring a problem and confronting it.

A practitioner example from FPSO decommissioning

The following example is based on an FPSO decommissioning project in which I was actively involved as Project Manager. Specific project details, company names, asset identification, commercial information, and sensitive execution data have been kept confidential. My purpose is not to disclose project-specific information, but to show how an AI-supported approach can improve project controls, forecast reliability, and governance in a complex EPC/offshore environment.

From a reporting perspective, the project appeared broadly under control. The FPSO had reached the yard. The next stage of dismantling was planned. The cost forecast remained within an acceptable range. Engineering, procurement, waste management, environmental compliance, and work package readiness were all being monitored.

However, the information was not being read as one integrated picture.

Several small signals were present. Waste treatment assumptions were still open. Cost responsibility for part of the preparation scope was not fully settled. Some work packages were progressing, but more slowly than required. Environmental readiness depended on pending confirmations.

Individually, none of these items seemed to justify major escalation. Together, they showed something more important: the forecast was still formally acceptable, but its basis was weakening.

This is where the AI-supported approach added value. It was not used to replace project judgement or create more dashboards. It was used as a decision-support layer over existing project controls. It connected schedule, cost, risk, waste, contractual, environmental, and readiness data, and helped test whether the assumptions behind the current forecast were still credible.

The early warning was not a single major delay. It was the pattern created by several small deviations.

That changed the management conversation.

Instead of saying, “The asset has reached the yard and dismantling remains planned,” the project team could say, “The asset has reached the yard, but verified readiness for the next stage is not yet confirmed.”

That change in language changed the quality of the discussion.

It allowed me, as Project Manager, to challenge the forecast before the issue became a formal exception. It also made the highlight report more useful. The report moved from describing progress to showing which assumptions could affect the next management stage.

For the Project Board, the decision became clearer. Rather than asking only whether the schedule date was still achievable, the key question became whether the next stage should be authorized without verified readiness on waste treatment, cost responsibility, environmental conditions, and work package maturity.

The AI-supported analysis did not make the decision. It helped the project team ask better questions earlier.

How this connects to PRINCE2

This is where the PRINCE2 method adds discipline.

For me, one of the strongest aspects of PRINCE2 has never been structure or documentation alone. It is the discipline of making clear what is tolerable, what is an exception, and when leadership needs to intervene and decide.

In offshore decommissioning, technical, commercial, and execution issues often move at different speeds. A technical team may see a manageable issue. The cost team may see a small exposure. The planner may see limited schedule movement. The risk owner may see a moderate risk.

The Project Board, however, needs to understand the combined effect.

AI-enhanced controls can support the PRINCE2 method by helping the Project Manager connect these views before tolerances are breached. This strengthens management by exception, improves the quality of highlight reports, and gives the Project Board a clearer basis for decisions.

It also supports continued business justification. In decommissioning, the business case can be affected by longer yard duration, higher waste treatment costs, vessel constraints, subcontractor performance, and scope uncertainty. If the forecast is based on weak assumptions, the Project Board needs to know early.

The same applies to stage boundaries. Moving from preparation to dismantling should be approved only when the project can demonstrate readiness, rather than simply because the planned date has arrived.

That is the practical value of evidence-led governance.

What practitioners can take from this

For project professionals using PRINCE2 in complex environments, the lesson is not that AI should replace project judgement. It should not.

The lesson is that AI-supported controls can help teams test the quality of their own assumptions.

A useful approach starts with disciplined data. The project needs clear ownership of schedule logic, cost assumptions, risk status, readiness criteria, change control, and evidence. Without that foundation, AI will only accelerate weak information.

The next step is to connect the data around decisions instead of reports. The question should be: what does the Project Manager or Project Board need to decide, and what information would make that decision more reliable?

In practical terms, this may include automated workflows for readiness actions, trend-based review of cost and schedule assumptions, early warning indicators linked to tolerances, and clearer escalation triggers for the Project Board.

There are clear limits. AI cannot validate a confined space, confirm a physical isolation, approve a lifting operation, or accept a regulatory obligation. It can support analysis, yet it cannot own an exception, approve a stage, or justify a change to the business case.

Trust in AI-supported controls depends less on the tool itself and more on the reliability of the data, assumptions, and governance around it.

Closing reflection

In complex projects, one of the most dangerous signs is the team's acceptance of deviation as normal, rather than the red indicator. That is why the discussion centers on project control, governance, and decision-making, with technology serving as an enabler. In FPSO decommissioning, a project may look ready because the schedule says it is ready. The more important question is whether the execution system proves it is ready.

That is where AI-enhanced project controls can support the PRINCE2 method. They can help project teams move from fragmented data to earlier insight, from status reporting to forecast challenge, and from assumptions to better decision evidence.

For practitioners, the questions are practical:

• • Is our project data reliable, current, and controlled enough to support AI-enabled analysis?
• • Are our stage boundaries supported by verified readiness evidence, or mainly by optimistic assumptions?
• • Are our reports helping decision-makers act earlier, or only explaining what has already happened?

The opportunity is to make project controls more disciplined, more connected, and more useful for decisions, while increasing automation where it adds value.