The Stewardship Lens: Connecting AI, Sustainability and Ethics


The Stewardship Lens: Connecting AI, Sustainability and Ethics

Most organisations now have an AI policy, a sustainability commitment and a code of conduct. On many projects these arrive as three separate sets of expectations with three different owners. The data protection lead wants to know which tools the team is using. The sustainability team wants a carbon figure for the business case. The ethics statement sits somewhere in the induction pack. A project manager could reasonably conclude that modern delivery has simply acquired three more compliance layers, each with its own template.

That conclusion is understandable, but it misses the more useful connection. AI, sustainability and ethical judgement raise the same underlying question in different forms: what does this project do to people and resources beyond the deliverable, and who is paying attention to it? Stewardship is a good name for that question. It means looking after something on behalf of others, knowing you will hand it on. Project managers are in that position all the time. They hold an organisation's money, data, reputation and relationships for a limited period before passing the results to someone else. Seen that way, the three topics stop competing for attention and start reinforcing each other.

The shared shape of the problem

Look at what each topic actually asks of a project manager and a common pattern appears. An AI tool can produce outputs faster and more cheaply than the process it replaces. Its errors, its use of personal data or its hidden assumptions still land on customers, applicants or residents who never see the tool. A sustainability decision about materials, energy or suppliers may make little difference to the project's own budget. Its effects arrive years later, in someone else's operating costs or in a community the project team will never visit. Ethical questions, meanwhile, tend to surface precisely where no rule settles the matter and the convenient option is not obviously wrong.

In all three, the consequence is usually displaced, delayed or both. It falls on someone who is not in the room when the decision is made, often after the project has closed. It is also largely invisible to the measures projects are normally judged by. A schedule, a budget and a scope baseline can all report green while a project quietly exports cost, risk or unfairness to people outside it. And in each case the shortcut is locally rational. Nobody chooses harm; they choose the option that makes this month's numbers work.

Readers who studied the Seventh Edition of the PMBOK® Guide will remember stewardship as one of its twelve named principles. The Eighth Edition consolidates the principles into six, and the word no longer has a heading of its own, but the concern has not disappeared. Read Sections 3.3 to 3.7 of The Standard for Project Management as a group and they describe a single discipline from five directions: a holistic view of what the project touches, attention held on value and not only output, quality built into what will actually be used, leadership that accepts accountability for consequences, and sustainability integrated into decisions instead of checked at the end. That reading is Omega's, not PMI's, but it is a practical one. A project manager who applies those principles seriously is already practising stewardship, whether or not a policy has asked for it.

Four questions that make the lens usable

A lens is only useful if it changes what you notice. The four questions below are an Omega device, not an official framework. They work equally well on an AI adoption, a materials choice or an awkward request from a sponsor.

The first is who carries the consequence if this goes wrong. This is not about who approved it or who would be blamed, but who actually experiences the effect. If the honest answer is a group with no voice in the project's governance, that is a signal to look harder, because nobody else in the room is placed to do so.

The second is when the consequence will show up, and whether the project will still exist by then. Effects that arrive after closure are the easiest to discount, because they will never appear in a project report. That does not make them less real. It makes the project manager one of the few people positioned to see them coming.

The third is whether the people affected would see the decision as reasonable if it were explained to them plainly. This is a practical fairness test, not a philosophical one. A decision that only holds up while nobody asks how it was made rarely holds up once somebody does.

The fourth is who could explain how the decision was reached. With AI this becomes pointed, because a tool can shape a decision without anyone being able to say why it produced a particular answer. The same test applies to a supplier's carbon claim or an estimate built on assumptions nobody has examined. If the explanation stops at "the system said so" or "the supplier assured us", accountability has moved somewhere it cannot be held.

None of these questions tells you what to decide. They tell you whether a decision is being made with its consequences in view, and by people with the authority to make it.

A routing model, a carbon target and the residents who never chase

Consider an original, fictional example. A housing association is replacing the system that schedules work for its repairs engineers. The project is hybrid: it has a fixed go-live date and a formal business case, but the platform is configured and tested in short iterations alongside the supplier's team. The supplier offers an optional module that uses AI to prioritise jobs and plan engineers' routes. Its projected benefits are attractive and credible. It promises fewer van miles, which supports the association's carbon reduction commitment, and faster average response times.

During testing, a repairs supervisor seconded to the project mentions that jobs from certain residents seem to sit lower in the queue than she would expect. The project manager does not switch the module off, and she does not dismiss the observation as a quirk of test data either. The first step is to find out whether the pattern is real. The team pulls a sample of past requests that arrived by post or through support workers and runs them through the test environment.

The pattern holds. The model has partly learned what counts as urgent from how often residents chased a job in the past. Residents who report once, by letter or through someone else, rarely chase, and many of them are older or more vulnerable. The residents most likely to wait longer were the ones least likely to complain. Nothing on the project dashboard would have shown it, because the dashboard reports only averages and mileage.

Run the four questions across that situation and it sharpens quickly. The consequence falls on residents who have no seat on the project board. It would arrive months after go-live, possibly as a serious disrepair case. It would be hard to defend if it had to be explained plainly to the people affected. And nobody on the project can currently say why an individual job ranked where it did.

What the project manager does next matters as much as what was noticed. Whether to accept this exposure is not her decision alone, because it trades a declared benefit against the association's obligations to its residents. Her stewardship contribution is to make the consequence visible and decidable. That means taking the sponsor a clear account of what testing found and what it would mean for residents, together with the realistic options:

  • go live with a human review step for vulnerability flags and certain job categories;
  • add response times by reporting channel to the acceptance criteria and benefits measures;
  • phase the rollout;
  • defer the module.

How the chosen response is recorded depends on which part of the hybrid project it touches. In the iterative configuration work, it becomes backlog items with explicit acceptance criteria. Against the fixed business case, it becomes a recorded change with its effect on benefits stated.

The carbon benefit was genuine, and nothing in this reasoning requires giving it up. The lens does not rank sustainability above fairness, and it does not treat AI as inherently suspect. It stops one legitimate goal from being achieved by quietly spending something else.

Where stewardship sits in ordinary project work

For PMP® candidates, the connection matters because the current exam reflects the same shift. The 2026 PMP Examination Content Outline names artificial intelligence and sustainability among the emerging trends that informed the job task analysis behind the exam. Sustainability then appears explicitly in enablers covering planning, quality, compliance and risk, and ethics is listed among the elements of project governance.

Ethics also has a longer history in the credential. PMI's Code of Ethics and Professional Conduct, which PMP holders agree to uphold, rests on responsibility, respect, fairness and honesty, and it predates the current generation of AI tools. The useful preparation habit is to read past the surface of a scenario. A situation involving an AI tool or a carbon target may turn on a familiar judgement about who is affected, what is actually known and who has the authority to decide. That is the kind of reasoning we develop in PMP® Exam Preparation, where working out what a situation is really asking matters more than recalling which topic it belongs to.

On a live project, stewardship rarely needs a new document. It needs familiar artefacts used with a wider field of view. When success measures are agreed, ask what they cannot see, since averages, totals and headline savings are exactly where displaced consequences hide. Widen the stakeholder list to take in people who will never attend a meeting: the residents, the future operations team, the people whose data a tool will process. Treat delayed and displaced consequences as risks with named owners, so that they are reviewed on the same cycle as everything else.

Supplier claims deserve the same scrutiny as any other input to a decision. When a supplier describes what its AI does or what its product saves, test that claim as rigorously as you would test an estimate. Then be explicit at handover about what operations inherits, including who will monitor a model once the project team has gone.

Two cautions keep the lens honest. First, stewardship is not a personal veto, and it does not license a project manager to substitute private values for the organisation's decisions. It works through governance. That means making consequences visible, recording decisions, and escalating when a legal, safety or ethical concern goes beyond the project manager's authority.

Second, it is not new. Experienced project managers have long thought about the operator who inherits the asset, the neighbours who live with the works and the safety obligations that outlast the project. That is especially true of those from construction, engineering and regulated environments. AI and sustainability extend that discipline into new territory rather than replacing it.

That is the real connection between the three topics. They are not three specialisms bolted onto project management. They are three places where an established professional obligation has become harder to ignore: looking after what you have been trusted with, on behalf of the people who will live with the results.

Andre Malowney

Interested in going further?

Seeing when a technology or environmental question is really about who carries the consequences takes practice against unfamiliar situations. Our PMP® Exam Preparation course builds that judgement across the 2026 outline, so the same reasoning holds whether the scenario involves an AI tool, a carbon target or a sponsor under pressure.

The principles behind reading AI, sustainability and ethics together are set out in full in the PMBOK® Guide, Eighth Edition.

Ad · Amazon affiliate link.