Harmful behaviors such as violence, substance abuse, and suicidal ideations profoundly affect both individual soldier well-being and military unit cohesion, corroding the foundation of our combat readiness and pulling leaders away from their primary mission: preparing for combat. Too often, leaders have only reactive solutions available, such as ordering safety stand-downs after a harmful behavior has occurred. What if, instead of simply responding to such events, they could predict where it was most likely to occur and proactively intervene?

This question sparked a multiyear grassroots effort called Project Prevention. A brigade data analytics and innovation team in the 4th Infantry Division at Fort Carson, Colorado developed a new data-centric method to proactively identify units at risk and enable timely, preventive interventions. The result was the Unit Risk Forecasting Tool (URFT), which applies predictive data analytics to give leaders an additional tool to keep their soldiers ready to train and fight.

Over the past year and a half, this tool identified the specific companies/batteries/troops within a brigade that were at increased risk of harmful behaviors in a given week, enabling dozens of opportunities for leaders to proactively step in to address the root causes fueling the risks. It was a low-cost investment in combat readiness with a documented empirical impact on the health and readiness of the force. It does not replace, but rather augments and focuses, human intuition—helping highlight otherwise unnoticed risk trends that busy leaders may have missed. By systematically analyzing existing data and applying rapid data modeling, the URFT functions as an advanced early-warning system, monitoring the subtle, almost invisible atmospheric changes within a unit that signal a coming storm.

The tool is currently scaling more widely across Army units, but we have only scratched the surface of its potential to both help soldiers and increase military readiness. Deepening the data architecture, refining the risk algorithm, and adapting the tool to specific unit needs can amplify its future impact.

Finding the Signal in the Noise: The Momentum Effect in Harmful Behaviors

The project began by trying to determine whether harmful behaviors occurred at random or followed some type of logic. To find out, the team built a comprehensive dataset spanning over two years of serious incident reports (SIRs), which include reportable harmful behaviors such as DUIs, weapon offenses, acts of violence, and suicide-related events.

As we sifted through this mountain of information, a stunningly clear pattern began to emerge from the noise—trouble rarely strikes just once. Rather, company-sized units that had experienced a recent surge in SIRs were significantly more likely to see additional harmful events in the near future. We termed this phenomenon the momentum effect.

Specifically, our initial data model indicated that a company that reported two or more serious incidents within the past three weeks showed a risk of another incident in the current week that was approximately three times higher than units without a string of SIRs. This discovery suggested that a cluster of incidents is not merely a string of bad luck, consistent with military folklore that bad news comes in threes. Rather, such trends are a symptom of deeper, unresolved issues, such as stress from a difficult training cycle, a toxic leadership climate, or low unit morale.

Armed with this insight, the team immediately moved to translate it into a practical tool for leaders. They developed a simple but effective automated alert system using Microsoft Power Automate. Now, when a unit’s incident count crosses that statistically significant threshold, its command team automatically receives an email alert. This simple alert acts as a digital tap on the shoulder, compelling leaders to look closer, initiate conversations with subordinate leaders, and ask the hard questions. It also allows them to move beyond blanket solutions and instead focus specialized resources—like chaplains, behavioral health specialists, or substance abuse counselors—where they are needed most.

This targeted approach produced results during a division-level pilot program in 2024. The team’s analysis estimated that URFT use reduced SIRs across the participating brigade compared to control brigades by 10 to 20 percent over the course of the year. This translated into roughly fifty avoided harmful behaviors, including an estimated twenty fewer suicide-related events, ranging from ideations to attempts.

Making Better Predictions: The Leading Indicators of Harmful Behaviors

Predicting risk based solely on past incidents was useful, but doing so is like driving while looking only in the rearview mirror. To make the URFT more forward-looking, we needed to enrich the dataset with leading indicators of harmful behaviors.

Building version 2.0 required us to begin pulling weekly data from the Army Vantage Data Analytics Platform, a massive repository of soldier information. This allowed us to create a much richer, more holistic snapshot of each unit, encompassing twenty-eight weekly observations for over eight hundred data points between July 2024 and January 2025.

We added company-level variables like the total number of officers, the number of noncommissioned officers, the number of junior enlisted personnel, average Army Combat Fitness Test score, the percentage of soldiers medically ready for deployment, and the primary function (simplified to combat, support, or headquarters).

Adding these unit variables allowed us to apply more sophisticated data models to find the hidden connections in the data. We reran the previous model, alogistic regression, which functions like a simple risk flag by weighing the different factors and estimating the probability that a unit will experience at least one serious incident in the next two weeks.

The team also used a second, more nuanced model called a Poisson regression. This tool acts less like a simple flag and more like a detailed weather forecast that predicts not just if it will rain, but how much rain is expected. It forecasts the likely number of incidents a unit might face, giving leaders a much clearer picture of the potential severity of the situation.

To further enhance predictive power, the team implemented a rudimentary form of artificial intelligence through a machine learning model known as XGBoost. We trained the software on 70 percent of the data we had gathered, allowing the algorithm to learn the incredibly complex and subtle interplay between all the variables. Then, we tested it on the remaining 30 percent of the data it had never seen before. The machine learning model showed it could correctly identify almost all units that genuinely experienced an incident—although the relatively high false positive rate suggests need for better data—making its forecasts useful for command teams who need to act on credible intelligence.

Actionable Insights: Engaged Leadership is a Key Prevention Tool

Running the new dataset through the three models not only confirmed the predictive power of the momentum effect but also unlocked another key insight: Leadership is a key factor in reducing harmful behaviors. While the presence of past incidents was a powerful predictor, the data revealed an even more critical factor. The models showed that as the ratio of noncommissioned officers to junior enlisted soldiers increased, the predicted number of harmful incidents significantly decreased.

This result provided evidence supporting a timeless military principle—noncommissioned officers are the first line of defense for their soldiers. They are the sergeants who live and work with young soldiers every day, who are responsible for their training, discipline, and morale. They are best positioned to spot the earliest signs of personal struggle—financial distress, marital problems, or a dip in motivation—and intervene with mentorship and support.

Of note, the model revealed a critical interaction: A large population of junior soldiers was not, in itself, a predictor of high risk, so long as there was a correspondingly strong cadre of noncommissioned officers to lead them. Effective supervision, the data showed, is a protective factor. Interestingly, some factors the team hypothesized would be significant, like a unit’s average fitness score, proved to have little to no predictive power, allowing leaders to focus on what truly matters. Additional data will allow us to uncover additional leading indicators and make more accurate predictions.

From Data to Action: Real-World Impact and the Road Ahead

The practical application of these findings is already transforming the way commanders lead their formations. The Unit Risk Forecasting Tool provides actionable intelligence that allows leaders to strategically allocate their most precious resources: time and attention.

Rather than waiting for a crisis to command their focus, they can now proactively engage with units flagged by the URFT, bringing in the right support systems to address underlying issues before they escalate into harmful behaviors. The next leadership challenge is to refine the design of these interventions to most effectively treat the root causes of risk identified by the data.

Encouraged by the project’s initial success, the team is working to expand the data sources fueling the URFT, incorporating broader information from across Fort Carson and, eventually, additional Army installations. A larger and more diverse dataset will inevitably lead to greater predictive accuracy.

The ultimate goal is to deploy these predictive models on a secure, user-friendly digital platform that can provide automated alerts to leaders Army-wide. The challenges are significant, as this requires robust data architecture and continuous collaboration with data owners across the entire institution to ensure a steady flow of high-quality information.

But by identifying and addressing potential problems before they escalate, command teams can save invaluable time that can be reinvested in preparing their units for complex combat missions. Moreover, this methodology applies far beyond harmful behaviors and could be adapted to predict combat-related events, logistical shortfalls, or maintenance issues.

Project Prevention is more than just a clever use of data; it represents a step forward in data-centric military leadership. Fusing the timeless principles of engaged leadership with the powerful tools of modern data science can achieve significant progress on one clear goal: to prepare for combat and safeguard the Army’s most valuable asset—American soldiers.

Lieutenant Colonel Jon Bate is a US Army infantry officer serving in the Joint Staff J5. He previously commanded 2nd Battalion, 23rd Infantry Regiment, 1st Stryker Brigade Combat Team, 4th Infantry Division. He has served in the 101st Airborne Division, in the 1st Armored Division, and as an assistant professor of economics in the US Military Academy Department of Social Sciences. A Goodpaster Scholar in the Advanced Strategic Planning and Policy Program, he holds a master in public policy from the Harvard Kennedy School and PhD in political science from Stanford University.

Lieutenant Colonel Stephanie Hightower serves as a physician at Tripler Army Medical Center in Hawaii, under US Army Medical Command and the Defense Health Agency (DHA). Prior to this assignment, she served as a practicing pulmonologist and intensive care physician at Walter Reed National Military Medical Center where she also led organizational wellness initiatives within the DHA. A Fulbright Scholar, she holds a master of science in global health from the University of Oxford and an MD from the George Washington University School of Medicine and Health Sciences.

Captain Jonathon “Caleb” Gage is an officer in the US Army, serving as a data systems engineer (FA26B) with 1st Stryker Brigade Combat Team, 4th Infantry Division at Fort Carson, Colorado. A 2019 graduate of the United States Military Academy, he earned a degree in geospatial information science before beginning his career in the infantry and later transitioning to his current technical specialty. His work focuses on the collection, organization, and formatting of data to ensure it is accessible and actionable for commanders and analysts. His efforts enable the Army to leverage data-driven insights to enhance operational decision-making and mission success, contributing to the Army’s ability to adapt and innovate in an increasingly complex and technology-driven environment.

The views expressed are those of the author and do not reflect the official position of the United States Military Academy, Department of the Army, or Department of Defense.

Image credit: Sgt. 1st Class Matthew Chlosta, US Army