that happens to every project manager: the stand-up meeting begins and suddenly a critical ticket is blocked, a developer calls in sick, a dependency slips or a key feature is delayed. In an instant, your carefully planned timeline starts to collapse, and you’re scrambling for solutions.
In this article, we’ll explore how a machine learning model predicted 41% of project delays before they hit the timeline, cutting costs and reducing last-minute firefighting.
The problem: 62% of IT projects miss their deadlines in 2025
As a Project Manager working with Agile teams, I’ve often dealt with delays and blockers, they quickly became part of everyday life. But when I came across the 2025 Wellington State of Project Management study revealing that in 2025, 62% of IT projects miss their deadlines, it shocked me into action. It’s an increase compared to the 2017 PMI Pulse of the Profession study where it was 51% in 2017. Project delays are reaching a critical level.
I know delays are common, but I hadn’t imagined it would be that high. But today, we have tools to anticipate and better understand these risks. Using Python and Data Science, I built a model to predict project delays before they happened.
This statistic highlights two critical points: delays often stem from recurring causes, and they carry a massive business impact. In this article, we’ll explore how data-driven approaches can uncover these causes and help project managers anticipate them.
With this knowledge, we can choose the best course of action.
This is where we can use data science. Surprisingly, the 2020 Wellington State of Project Management report announces that only 23% of companies use project management software to manage their projects, even though these tools generate a wealth of valuable data.
By analyzing information from project tickets, we can build predictive machine learning models that highlight potential risks before they escalate.
That’s exactly what I did: I analyzed more than 5,000 tickets, not only from my current project but also from past initiatives.
Project management software, it turns out, is an incredible source of data waiting to be leveraged.
The Data Gap in Project Management
In traditional project management, reporting plays a central role, yet few reports offer a comprehensive, detailed retrospective of the project as a whole.
For example, in Scrum, we track our velocity, follow the trend of our burndown chart, and measure the number of story points completed.
Traditional reporting still fails to give us the complete picture. Data science can.
As Project Managers, we may know from experience where the critical points lie, but validating those assumptions with data makes our decisions far more reliable.
Building the Dataset
To explore this idea, I analyzed 5,000 Jira tickets—one of the richest sources of project data available.
Since real project data can’t always be shared, I generated a synthetic dataset in Python that mirrors reality, including key variables such as priority, story points, team size, dependencies, and delay.
Code by author
Having built a realistic dataset, we can now explore the different ticket profiles it contains. This sets the stage for our exploratory data analysis.
Most tickets are of low or medium priority, which is consistent with how project backlogs are usually structured. This initial distribution already hints at where risks might accumulate, a point we’ll explore further in the EDA.
While high and critical priority tickets represent a smaller share of the total, they are disproportionately more likely to be delayed.
This bar plot confirms the phenomenon: high-priority tickets are strongly associated with delays. However, this can stem from two different dynamics:
- High-priority tickets are inherently more complex and therefore at greater risk of delay.
- Some tickets only become high priority because they were delayed in the first place, creating a vicious cycle of escalation.
With this simulated dataset, we now have a realistic snapshot of what happens in real projects: tickets differ in size, dependencies, and complexity and some inevitably end up delayed. This reflects the everyday challenges project managers face.
The next step is to move beyond simple counts and uncover the patterns hidden in the data. Through Exploratory Data Analysis (EDA), we can test our assumptions: do higher priorities and more dependencies really increase the likelihood of delay? Let’s find out.
Exploratory Data Analysis (EDA)
Before moving to modeling, it’s important to step back and visualize how our variables interact. Exploratory Data Analysis (EDA) allows us to uncover patterns in:
- How delays vary with priority.
- The impact of dependencies.
- The distribution of story points.
- The typical team sizes handling the tickets.
This chart confirms a key intuition: the higher the priority, the greater the probability of delay.
Dependencies amplify this effect, the more there are, the higher the chances of something slipping through.
Once a delay or risk of delay appears, escalation mechanisms push the priority even higher, creating a feedback loop.
Finally, ticket complexity also plays a role, adding another layer of uncertainty.
Most tickets fall into the medium-risk category. These demand the most attention from project managers: while not critical at first, their risk can quickly escalate and trigger delays.
High-risk tickets, though fewer, carry a disproportionate impact if not managed early.
Meanwhile, low-risk tickets usually require lighter monitoring, allowing managers to focus their time where it truly matters.
We also notice that most tickets have small story point sizes, and teams are usually around five members.
This suggests that agile practices are generally being followed.
Now, we will go further and look at the distribution of risk scores across tickets.
We see that only a small portion of tickets carry a very high risk, while most sit in the medium zone. This means that by focusing early on the riskiest tickets, Project Managers could prevent many delays.
To test this assumption, let’s now explore how complexity per person and priority interact with risk scores.
We cannot observe a clear trend here. The risk score does not seem to strongly depend on either ticket complexity or priority, suggesting that other hidden factors might drive delays.
Technical Deep Dive: Predictive Model
The raw data provides a solid foundation, but domain knowledge is essential to building a truly robust model. To better capture the dynamics of real-world projects, we engineered new features that reflect project management realities:
- Complexity per person = story points/team size.
- Has dependency = whether a ticket depends on others (dependencies > 0).
- Priority story points interaction = priority level multiplied by story points.
Code by author
We chose a Random Forest model because it can handle non-linear relationships and provides insights into feature importance.
Our main focus is on Recall for the positive class (1 = delayed). For instance, a recall of 0.6 would mean the model correctly identifies 60% of all truly delayed tickets.
The objective is not perfect precision but early detection. In project management, it’s better to flag potential delays, even with some false positives, than to miss critical issues that could derail the entire project.
Code by author
The model achieved a recall of 0.41, meaning it successfully detected 41% of the delayed tickets.
This may seem modest. However, in a project management context, even this level of early warning is valuable. It gives Project Managers actionable signals to anticipate risks and prepare mitigations.
With further refinement, the model can be improved to anticipate more delays and help prevent issues before they materialize.
We will use a confusion matrix to better understand the model’s strengths and weaknesses.
Code by author
The model correctly identifies 169 delays, but it also generates 373 false alarms, tasks flagged as delayed that actually finished on time. For a Project Manager, this trade-off is acceptable because it’s better to investigate a few false positives than to miss a critical delay. This is a part of risk management.
However, the model still misses 245 delayed tickets, meaning its predictions are far from perfect.
Overall, this model works best as an early warning system. It provides valuable signals but still needs further training and refinement. Most importantly, it should be complemented with human expertise, the judgment and experience of Project Managers, to ensure a complete and reliable project overview.
Model Interpretability, Scoring, Business impact, Dashboard & model validation
To truly understand why the model makes these predictions, we need to look under the hood. Which features drive the risk of delay the most? This is where model interpretability comes in.
Code by author
We can observe that complexity and the priority-story points interaction are the strongest drivers of prediction accuracy.
Scoring tickets: Identifying What’s Really at Risk.
Why does this matter for Project Managers? Because we can go one step further.
Calculate a risk score for each ticket.
This score highlights which tasks are most at risk, allowing PMs to focus their attention where it matters most and take preventive action before delays escalate.
Code by author
Business Impact Analysis.
The tickets with the highest risk scores confirm the trend: only high and critical priority tasks carry the greatest risk.
This insight matters not only for managing the project timeline but also for its financial impact on the business. Delays don’t just slow down delivery, they increase costs, reduce client satisfaction, and consume valuable team resources.
To quantify this, we can estimate the business value of the predictions by simulating how much cost can be avoided when we anticipate risks and take preventive action.
Code by author
Our baseline shows that 27.6% of tickets are delayed. But what if Project Managers could focus only on the riskiest 20%? We will now simulate this targeted intervention and see how much impact it makes.
Code by author
We identified 1,021 high-risk tickets, representing about 20% of all tasks. Among them, 516 (50.5%) are actually delayed. In other words, these few tickets alone drive roughly 10% of total project delays.
To make this more concrete, we translate the impact into business terms with a medium-size project valued at $100,000. By applying preventive actions on these high-risk tickets, we can estimate the potential cost savings.
Code by author
By taking early actions, we could save $9,270, nearly 10% of the total project cost. That’s not just risk mitigation; it’s a direct business advantage.
PM Dashboard
To make these insights actionable, we can also build a Project Management Dashboard. It gives a real-time view of sprint health, with all the key KPIs needed to track progress, anticipate risks, and maintain a complete project overview.
Code by author
Model validation
We tested the robustness of the model with a 5-fold cross-validation. Recall was chosen as the main metric, because in project management it’s more important to catch potential delays than to maximize overall accuracy.
Code by author
The recall scores across folds ranged from 0.39 to 0.42. This means the model is far from flawless, but it consistently flags around 40% of delays, a valuable early warning that helps project managers act before issues escalate.
Conclusion
In conclusion, this article showed how data science can help make projects smoother by providing a clearer understanding of the causes of delays.
Data doesn’t replace the intuition of a Project Manager, but it strengthens it, much like giving a pilot better instruments to navigate with precision and with a better view of what is happening.
By predicting risks and identifying at-risk tickets, we can reduce delays, prevent conflicts, and ultimately deliver more value.
Project Managers should embrace data science. Today, there are two types of PMs: traditional ones and data-driven ones. They don’t compete in the same league.
Finally, these skills are not limited to project management. They extend to product management and business analysis. Learning SQL or Python enhances your ability to collaborate with developers, understand product performance, and communicate effectively across all levels of the business.
Lesson for Project Managers
How many of our project decisions are based on so-called “best practices” that are, in fact, unverified assumptions? Whether it’s about meeting schedules, team structure, or communication methods, data can help us challenge our biases and uncover what truly works.
Depending on the organization, the analysis can also go deeper: grouping tickets by project phase, field, or stakeholder may reveal hidden bottlenecks and systemic issues.
For example, velocity often drops during the QA phase. Is it because QA engineers underperform? Not at all. They do their job very well. The real issue is the constant back-and-forth with developers: clarifying tickets, figuring out how testing should be performed, or asking for missing information.
To solve this, we introduced a simple process: developers now add clearer testing details in the ticket and spend five minutes on a quick handover call with QA. That small investment of time boosted team productivity and velocity by more than 15%.
Who am I?
I’m Yassin, an IT Project Manager who decided to learn Data Science, Python, and SQL to bridge the gap between business needs and technical solutions. This journey has taught me that the most valuable project insights come from combining domain expertise with data-driven approaches. Let’s connect on LinkedIn