Earlier this year, I watched a demonstration of an AI tool aimed at project managers. In about two minutes, it generated a complete business case from a bare bones brief, a risk register with multiple items assigned to project team members, and a benefits map. The outputs looked professional. They were formatted correctly, used the right terminology, and ticked the boxes that a governance board would expect to see.
The problem was that nobody had scrutinised any of it.
The risk register listed generic risks that could apply to any project in any organisation, alongside some pulled from the details in the business case. The benefits were plausible, again from what the business case said, but unmeasured. The business case made a reasonable argument for the project without engaging with any of the specific context that would determine whether it was actually true. It was documentation that looked like thinking without any of the thinking having happened.
Worse, the only context provided was a short couple of paragraphs as a project brief / mandate. The tool used generative AI to create a business case – and from there, used that as an input to create the risks, and the benefits. This very easily leads to massively compounding errors – particularly if no-one is really thinking about these items, just checking the output for plausibility.
I don’t tell this story to dismiss AI. I use AI in my own work, and I’ve seen it do genuinely useful things. But the current wave of AI tools being sold to the project management market is mostly positioning AI as a way to produce outputs faster – to generate the documents, fill in the templates, write the reports. And that fundamentally misunderstands what those documents are for.
The purpose of project documentation
A business case is not a document. It is a structured argument that forces the people responsible for a project to articulate, clearly and in writing, what they are trying to achieve, why it is worth doing, what it will cost, what the risks are, and how they will know if it has worked.
The value of the business case is not the output. It is the process of producing it – the conversations it requires, the assumptions it forces you to surface, the disagreements it makes visible. A business case produced by an AI in a few seconds hasn’t done any of that. It has given you a document that passes a cursory inspection while leaving all the hard questions unasked.
The same is true of risk registers. The point of identifying risks is not to have a list of risks. It is to force your project team to think systematically about what could go wrong, to have honest conversations about probability and impact, to assign ownership and accountability, and to build contingency into your planning. A risk register generated by AI based on your project description will give you a plausible list of generic risks. It will not give you the project-specific insight that comes from your team actually sitting down and thinking hard about their particular environment, their particular constraints, and their particular uncertainties.
Where AI actually helps
None of this means AI has no place in project delivery. It means we need to be clearer about what it’s good for.
AI is very good at processing large amounts of information quickly and identifying patterns or inconsistencies. That is genuinely useful for project managers. If you give a capable AI model your business case, your timeline, your resource plan, and your benefits map and ask it to identify inconsistencies, it will often find things that a human reviewer might miss – your timeline assumes parallel workstreams that your resource plan shows are staffed by the same person, for example, or your benefits case assumes outcomes that your scope of delivery doesn’t actually include.
That kind of challenge is the question a good programme board chair would ask, if they had the time and the detailed knowledge to read every document closely. AI can give every project manager access to that level of analytical challenge, on demand. That is a meaningful capability improvement.
AI is also useful for drafting. Not for producing final documents, but for getting something onto the page that a human can then review, challenge, and improve. The instinct to start from a blank page is often the instinct to procrastinate; a draft that is 70% right is much easier to work with than nothing. Used this way – as a starting point rather than an endpoint – AI saves time without sacrificing quality.
And AI is useful for summarisation, for research, for generating options when you’re stuck, for reviewing your own writing for clarity. These are productivity improvements, not replacements for judgement.
The distinction that matters
The useful distinction is between AI that helps you think and AI that thinks for you.
AI that helps you think produces a draft business case that you scrutinise and improve. It challenges your risk register with questions your team then debates. It cross-references your documents and asks whether the timeline is consistent with the resource plan. It gives you more to work with and forces you to engage more rigorously with your own project.
AI that thinks for you produces documents that look finished. It removes the friction of having to articulate your assumptions. It generates outputs that pass formal review without the underlying thinking that makes those outputs meaningful. It makes governance faster at the cost of making it superficial.
The project management market is currently being flooded with tools in the second category. They are sold on speed – on how quickly they can produce a project plan, a risk register, a status report. Speed is a genuine benefit when it is applied to tasks that were previously slow for no good reason. It is not a benefit when it is applied to tasks that are supposed to be slow because the slowness is the point. Removing effort from tasks where the point is the effort is a failure, not a win.
What this means for how you use AI
Use AI to make your thinking sharper, not to replace it. Use it to challenge your documents after you’ve written them, not to write your documents for you. Use it to identify what you might have missed, not to give you permission to stop looking.
The projects that will benefit most from AI in the next five years will not be the ones where AI produces the most outputs. They will be the ones where AI helps the project team ask better questions, make better decisions, and anticipate problems earlier. That requires treating AI as a tool for augmenting human judgement rather than automating it away.
Project management, at its core, is about creating the conditions for a group of people to deliver something complex in an uncertain environment. No AI can do that. But the right AI, used in the right way, can make the humans doing it considerably better at their jobs.
That’s the version of AI worth investing in.
Trevor Roberts is a programme and project management consultant and the founder of Dull Industries – a consultancy focused on project turnaround, AI implementation, and digital strategy.