AI Decision Policy.
The line PCI draws on AI in consequential decisions — including its own.
Overview
Some decisions must never be delegated to a machine. This policy makes explicit where AI may assist and where a competent human must decide, both within project controls and within PCI’s certification process.
At a glance
Where AI may and may not be used
- AI may assist analysis, forecasting, drafting and pattern-spotting
- AI must not make a certification decision
- Consequential outputs must be validated and owned by a competent person
- The basis of a decision is recorded and explainable
Why the line matters
PCI applies this consistently and documents what it does, so the approach can be checked and improved. Final detail is published as the institute matures, and is governed by the related policies below.
Common questions
Can AI make a certification decision?
No — certification decisions are made by people, under the principle that AI proposes and the professional disposes. AI may support parts of a process, but the judgement about whether a candidate has met the standard rests with the certification body and its competent assessors, never with an automated system alone.
Can professionals use AI to analyse a project?
Yes, responsibly, with human validation of the output.
Is AI use explainable?
Yes — conclusions must be understandable and justifiable.
Why this matters
This matters because AI is already changing how project controls is done, and the risk is not that professionals use it — it is that they use it without judgement. The whole point of the standard is to make the responsible use of AI explicit: AI proposes, the professional disposes, and a competent human validates, explains and owns every AI-assisted output.
PCI is deliberately careful here. It does not overclaim what AI can do, nor what its own credential guarantees. It sets a clear, honest expectation for governing AI in controls work and certifies professionals against it — keeping accountability where it belongs.
- Substance over marketing
- Fair, transparent process
- Honesty about our status
- Responsible, governed use of AI
Founding-stage document · Version 1.0 — effective date to be confirmed · Reviewed under PCI governance. PCI makes no claims of accreditation or recognition beyond what is true today.
Four questions before you rely on an AI output
Policies work best as habits. Before an AI-assisted analysis, forecast or draft feeds a consequential decision, run it through four questions. If any answer is no, the output is not yet fit to be relied upon.
- Can I explain it? You can state in your own words how the conclusion was reached and why it is plausible against the project data you already know.
- Have I checked it? The output has been tested against an independent reference — a manual calculation, a prior period, a known benchmark — not simply read and accepted.
- Would I sign it? You are prepared to put your name to the conclusion as your own professional position, with the tool's contribution acknowledged.
- Is the record honest? The decision record identifies where AI assisted, so a later reviewer can see what was machine-proposed and what was human-judged.
This is the same discipline the AI standard expects of certified professionals, and the same discipline PCI applies to itself under its human-oversight commitments.
What a defensible decision record looks like
Because the basis of every consequential decision must be recorded and explainable, the quality of the record matters almost as much as the decision itself. A defensible record is short but complete — it lets someone reconstruct the judgement without the decision-maker in the room.
The question decided
What was actually at stake, stated plainly — an eligibility outcome, a forecast adopted, a sanction applied.
The evidence weighed
The inputs considered, distinguishing clearly which were AI-assisted and which were not.
The validation performed
How AI-assisted inputs were checked before being relied upon, and by whom.
The named owner
The competent person who made the decision and can account for it afterwards.
The reasoning
Why the decision went the way it did, in enough detail for a reviewer to follow the logic.
The route to challenge
Confirmation that the affected person was told how to seek a review of the outcome.
Keep going
Decisions AI may inform but never make
Certification decisions — eligibility, results, sanctions, credential status — are human decisions supported by evidence. AI may assist the evidence layer (statistical flags, anomaly detection, drafting), but a person weighs it, can articulate why, and signs the outcome. "The system decided" is not a sentence this Institute will ever offer a candidate.
Two safeguards make that durable: any AI-assisted input to a decision is identified as such in the record, and affected individuals retain full review and appeal rights against the human decision, whatever informed it.