Is AI helping us make better decisions or just faster bad ones?

The argument around the quality and speed at which we can now access data is something I've been pondering since we completed our research into digital transformation and AI in project management (published by APM as Digital Transformation and the AI Imperative in Public and Private Sectors).

To put this in perspective, a person in the Middle Ages would absorb a lifetime of data through stories, conversations and experiences. Today’s project managers face that same lifetime volume every single day, delivered through dashboards, reports and emails. We're talking about up to 500 million daily data points in modern mega-projects.

We've shifted from information famine to information feast, and I'm not convinced our brains have caught up. In behavioural economics, the theory of bounded rationality explains our cognitive limits in processing information. Take chess. Most people can anticipate two to four moves, while grandmasters manage five to 10. We all have limits, and they have not vanished simply because AI now feeds us insights at lightning speed.

The acceleration trap

We’re in danger of a speed trap. Artificial Intelligence (AI) excels at accelerating access to information and automating routine decisions. It can scan those 500 million daily data points and highlight patterns that would take human analysts months to find. But acceleration is not the same as improvement.

Our interviews revealed persistent concerns about data quality and the fundamental challenge of distinguishing between AI recommendations that genuinely enhance decisions and those that simply accelerate poor choices. Fundamentally, the problem with low quality at high speed is what the host of HBO’s Last Week Tonight with John Oliver refers to as “AI slop”, an ever-expanding deluge of AI noise.

The quality shepherd

This brings me to a fundamental question about data discernment. Can we tell apart AI recommendations that genuinely improve decisions from those that simply accelerate poor ones? I believe our profession needs people who know when to trust AI and when to override it.

The idea of the quality shepherd means having the ability to ask awkward questions. Why is the AI recommending this approach? What might this recommendation be missing? What data could it be overlooking? What human factors aren’t captured in the algorithm? Most importantly, does this recommendation actually serve our project goals, or does it just feel impressively data-driven?

Our research identified six key digital competencies that create effective quality shepherds. Digital and IT skills provide basic fluency to navigate AI tools, while data literacy and critical thinking enable you to question what algorithms are telling you. AI integration capabilities help you understand when to trust predictive modelling versus human expertise. Digital collaboration skills ensure you can effectively challenge AI recommendations within your team. Agility allows quick adjustments when AI insights prove to be wrong, while strategic digital leadership provides confidence to override algorithmic recommendations when human judgement suggests a different path.

Leading through digital complexity

Being a quality shepherd isn't just about technical competency, it's about leadership. The role requires strategic thinking to judge when human insight should override algorithmic recommendations. When do you trust AI-driven scheduling versus your own knowledge of team dynamics? How do you balance predictive analytics with stakeholder concerns that algorithms can't capture? Quality shepherds will need to navigate organisational resistance and foster cultural change, showing teams that digital transformation is not simply about adopting technology, but about evolving how they collaborate and make decisions alongside it.

I've spent much of my career studying why projects succeed or fail. Tools alone do not deliver projects; people do. AI can make us faster, more efficient and even more confident. But whether it makes us better depends on how we choose to use it.