Data literacy: Reading project data critically in the age of AI

Project professionals now have access to more data, more frequently and in more detail than at any previous point in the profession’s history. The quality of decision-making has not kept pace. The Wellingtone State of Project Management Report 2026 found that 42% of project professionals spend at least a day manually collating reports and that half of organisations lack access to real-time KPIs, highlighting persistent challenges in project visibility and performance management.The gap between data volume and data understanding has never been wider.

What data literacy actually means

Data literacy is not the ability to read a dashboard. It is the ability to understand what a dataset can and cannot tell you. To identify where assumptions are embedded. To distinguish correlation from causation. To recognise when an apparently precise number is a construction rather than a measurement. And to ask the questions that the data is not designed to surface.

In project governance, these skills matter because the data available is almost always constructed. Earned value calculations depend on how progress is measured. Confidence percentages depend on what historical patterns the model was trained on. RAG statuses depend on thresholds that someone chose. None of these are neutral measurements. They are interpretations. The quality of governance depends on whether those interpretations are understood and challenged, or treated as objective facts.

The data quality problem hiding in AI-generated reports

Behind every AI-generated dashboard is a data quality problem. Reporting data in projects is rarely clean: self-reported progress, inconsistently applied category definitions, financial integrations that may not align and transformation rules that compress complex realities into single numbers. When AI processes this data, it does so at speed and scale, meaning errors and inconsistencies are amplified alongside accurate information. AI systems can produce outputs with such fluency and apparent precision that the underlying quality problem becomes harder, not easier, to identify.

The black box problem

Modern AI systems cannot explain their internal reasoning in any operationally meaningful way. When a predictive analytics tool produces a delivery confidence score, the path from input to output cannot be audited. The same input may produce different outputs on different occasions. When an AI tool recommends a resource reallocation or flags a risk, the project professional cannot trace the reasoning. They can only validate the output against independent evidence.

This is not a reason to avoid AI tools. It is a reason to treat their outputs as the starting point for human analysis, never the conclusion. The accountability for a governance decision remains with the professional who made it, regardless of what the AI recommended.

Correlation is not causation

One of the most persistent data literacy failures in project governance is treating correlation as causation. AI tools are particularly effective at identifying correlations in large datasets. A project analytics platform may surface a strong correlation between team size and delivery delay, between stakeholder meeting frequency and scope creep, or between any number of project variables. Correlations can be useful. Acting on them as if they explain the underlying relationship is a different matter entirely.

Good data governance in project environments requires asking not just what the data shows but why it shows it, and whether the explanation assumed by the AI’s presentation is actually supported by the evidence.

Practice checklist

• Ask what the data cannot show: Before every significant governance decision: what questions is the available data not designed to answer? What is systematically absent from the picture?
• Trace the construction: For every key metric: who defined it, how is it measured, what choices were made in its production and what would it look like if those choices had been different?
• Challenge the confidence: When an AI tool presents a precise number, ask what that precision is based on. High specificity in output does not mean high reliability in the underlying data.
• Separate correlation from causation: When the data shows a relationship between variables, ask explicitly whether there is evidence for a causal mechanism or whether you are looking at correlation without explanation. 
• Keep the black box in view: When you cannot audit the reasoning behind an AI output, document that explicitly in your governance record, alongside the independent evidence you applied to validate or challenge it. 

 

You may also be interested in:

• How to Use Data to Make Better Decisions
• What is artificial intelligence in project management?
• Explore AI and data tools and modules through APM Learning