Many of us wear some kind of fitness tracker, such as a smartwatch or smart ring, or track fitness data using an app on our phones. These trackers will notify us when we need to get up and move, tell us when we should go to bed for optimal sleep, or suggest when we may need to practice some breathing exercises. These are personalized notifications—so how does the tracker know what our bodies need?

Smart devices use predictive analytics to provide us with tailored health recommendations. While we wear them, they track various metrics, including heart rate, steps taken, calories burned, and sleep quality. Through the daily collection of this data, these devices begin to understand our habits and activity patterns. Your tracker may notice that when you get less than seven hours of sleep, your energy tends to dip around 3:00 pm. Using predictive analytics and machine learning, the tracker may notify you to take a short walk or drink water near that time, as it predicts you may be feeling fatigued. It would also recommend that you wind down earlier that night and perhaps practice meditation to ensure you get a good rest. 

Predictive analytics uses statistics and modeling techniques to find patterns in historical and current data and predict future outcomes. It forecasts how trends may continue to provide recommendations on what adjustments should be made. Predictive analytics underlie how we determine weather forecasts, how our phones suggest auto-responses, how streaming services recommend shows, and how Google Maps provides the fastest route. Predictive analytics help to optimize our daily decisions, shaping our experiences in ways we often do not notice. 

Machine Learning: A subset of artificial intelligence where computers use statistical techniques that enable them to learn from data, recognize patterns, and make predictions with minimal human intervention. Machine learning supports predictive analytics by leveraging historical and real-time data to make predictions about future outcomes and refine its statistical models.² 

Data Mining: The process of collecting and analyzing a large amount of raw data to identify patterns and extract information. Data mining happens before predictive analytics, to identify correlations and trends in historical data, which can then be applied to statistical models and forecasting techniques.³ 

Statistical Modeling: The mathematical structure that underscores predictive analytics. Statistical models uncover relationships between variables and use this analysis to generate a visual representation of the data and generate future datasets to make predictions from.⁴
Regression Analysis: A statistical model that predicts the probability of an event occurring based on the analysis of the historical relationship between an independent variable and a dependent variable.⁴ For example, a weather network would predict the chance of rain (dependent variable) based on temperature, wind speed, and humidity (independent variables). Regressions are one of the most common predictive analytics models.²

The first recorded use of predictive analytics can be traced back to 1689. Edward Lloyd opened a coffee house on a busy street in London that quickly became a popular hangout for sailors and ship owners. The coffee house soon evolved into an insurance company, where deals were made to insure ships.⁵ The company used data from past sea voyages to evaluate risk and predict patterns of liability to price the insurance policies.⁶ 

In the 19th century, English statistician Francis Galton introduced the idea of regression after studying the relationship between fathers’ and sons’ heights. He found that sons’ height tended closer to the average of the population instead of following the pattern of their father’s height. This discovery showed that extreme data points move closer to average over time—a phenomenon called regression—which later was expanded to regression analysis, a technique that models the relationship between variables based on historical data.⁷

It wasn’t until the advent of the computer that the practice of predictive analytics expanded to other fields. The Manhattan Project, a program to develop the first atomic bomb during World War II, used computers for a Monte Carlo simulation, a method that generates a broad set of random numbers to simulate the result of an experiment. Scientists used this predictive analytics method to predict how neutrons would move and react inside nuclear material.⁸ 

In 1950, the first programmable general-purpose electronic digital computer, known as the Electronic Numerical Integrator and Computer or ENIAC, leveraged predictive analytics to predict how the airflow in upper levels of the atmosphere impacted weather. This historical analysis was the first time computers were used for weather forecasting.⁸ A few years later, Arnold Daniels, a World War II flight navigator whose battalion did not encounter a single casualty, developed the Predictive Index Assessment, a tool that assesses workplace behavior based on past data and helps businesses hire the right people. Daniels became interested in predictive analytics after a psychologist was sent to study his battalion’s success to see how personality traits may contribute to survival.⁶

Daniels’ Predictive Index led to the adoption of predictive analytics by businesses worldwide, using tools to find the root causes of workplace issues, optimize processes, and maximize profits, all through data-driven decisions. As computers became more advanced and had greater storage and processing capabilities, predictive analytics became more nuanced, able to identify complex correlations and, with additional causal inference techniques, explore potential causal relationships.⁶ 

As we enter the age of big data, more sophisticated AI tools, and machine learning, the power of predictive analytics has increased. Vast amounts of data on people and their behavior can be collected and analyzed for public and private sector organizations to predict future outcomes and adjust their strategies accordingly. 

Edward Lloyd

An English businessman who opened a coffee house in London in 1688 on Tower Street. The coffee house attracted merchants, bankers, sailors, and ship owners, who used the spot as an informal place to discuss business. Underwriters also began frequenting the coffee shop, which led to the first recorded instance of predictive analytics being used to insure ships during sea voyages in 1689.⁹ Lloyd went on to found a newspaper, Lloyd’s News, to share information about the shipping and marine insurance market. Years later, a group of underwriters established Lloyd’s of London, a structured insurance market, which continues to be a global success today.¹⁰

Francis Galton

An English statistician and anthropologist who introduced regression to the research community after realizing that sons’ height would skew towards the population average as opposed to following in their father’s footsteps—sons with tall fathers were slightly shorter, and sons with short fathers were slightly taller. He showed that over time, data showed a regression toward the population mean, which laid the foundation for regression analysis, a broader approach that analyzes the relationships between independent and dependent variables to predict future outcomes. Galton is now infamous for his contributions to eugenics, suggesting the biological superiority of certain races and advocating for the improvement of the human species by limiting who could reproduce.¹¹ 

Arnold Daniels

A flight navigator during World War II who successfully led a battalion that suffered zero casualties. The success of Daniel’s battalion led psychologists to study the personality traits of its members to determine if certain characteristics or leadership qualities could predict success. This experience sparked Daniels’ interest in how psychology could predict performance and led to his development of the Predictive Index Assessment.⁶ This tool could be leveraged by businesses to understand how employee characteristics contribute to success and inform hiring decisions. The tool—which has been validated by over 500 studies—is still used today as a science-backed approach to talent optimization.¹² 

Predictive analytics has transformed industries by enabling organizations to make data-driven decisions, optimize operations, and anticipate future trends. From finance and healthcare to marketing and beyond, predictive models help businesses reduce risk, improve efficiency, and enhance customer experience.

Finance

Predictive analytics are widely used in finance, where there are various levels of risk embedded in each decision. 

Predictive analytics allows underwriters to evaluate the likelihood that they will have to pay the customer out for a future claim in order to accurately price an insurance policy.² For example, if you are looking to purchase car insurance, an underwriter will evaluate the likelihood of a claim by comparing your characteristics—such as age, gender, location, and vehicle type—to historical data of others that match your profile. If they predict you have a high likelihood of making a future claim, they will increase the price of the policy. Similarly, banks will use your credit score, which is based on your historical behavior, to assess the likelihood of you defaulting on a loan for credit risk management.¹³ 

Predictive analytics are also used for fraud detection. Financial services will collect and analyze thousands of data points from their customers to identify when an irregular transaction occurs. This allows them to quickly spot the potential of fraudulent activity so they can freeze the transaction and avoid losses for both the company and customers.² 

Healthcare

In healthcare, predictive analytics helps organizations, hospitals, and doctors to make data-driven decisions, which can greatly improve patient care. By analyzing historical medical data, machine learning algorithms can be used to flag when a patient is at higher risk of developing a disease, as well as indicate what method of treatment is most likely to lead to success.¹⁴ 

For example, a study was conducted by the University of California San Francisco to see how predictive analytics could be used to anticipate patient deterioration in the intensive care unit to indicate the need for interventions to prevent further complications or death. By analyzing electronic health records, vital signs, and other medical history sources, a system was developed to alert healthcare professionals to early signs of sepsis, respiratory failure, and heart attacks. This allowed nurses and doctors to intervene earlier to improve patient outcomes.¹⁵ 

Beyond the individual level, predictive analytics allows healthcare organizations to evaluate when a disease outbreak may happen for a given population. By analyzing data from hospital visits and Google search trends (such as people googling flu symptoms) and comparing this to historical data, healthcare organizations can predict when and where an outbreak may occur. This allows them to take preventative measures, such as allocating more resources, to reduce the impact of the outbreak.¹⁴ 

Behavioral Targeting

By analyzing past user behavior like browsing history, purchase patterns, and social media activity, marketers and businesses can predict which products are likely to appeal to their target audience and what kind of content is most likely to engage them. Predictive behavioral targeting allows companies to anticipate future behaviors based on historical data to offer tailored content and personalized recommendations. For example, if historically, when users visit travel blogs, they then tend to purchase vacation packages, an airline may target individuals who browse these sites with ads for vacation deals.¹⁶ 

Additionally, marketers can look at consumer trends to understand how shifts in the economy might impact a demographic’s spending habits.² For example, if a clothing company uses predictive analytics to identify that consumers tend to spend less on luxury purchases during economic downturns and inflation has just risen, they can develop or promote more budget-friendly clothing lines. 

Predictive analytics relies on historical data to forecast future outcomes, but this approach has limitations. Models showing correlation can be misinterpreted as evidence of causation, the results can reinforce existing biases, and models may struggle to adapt when real-world conditions change. These challenges highlight the risks of relying too heavily on past patterns to make future decisions.

Correlation, Not Causation

Predictive models use historical data to understand how independent variables impact dependent variables. While predictive analytics can detect patterns and show correlation, it does not inherently explain why a certain outcome occurs. As Eric Siegel, consultant and author of Predictive Analytics, writes, “We know the what, but we don’t know the why. When applying [predictive analytics], we usually don’t know about causation…”¹⁷

Sometimes, you can guess about causation using common sense. For example, if a company sees that historically, more people tend to buy umbrellas in early spring, they can assume it’s because of the rainy weather. However, other times, it’s less obvious, and figuring out causal relationships may be crucial for adapting strategies. In healthcare, predictive analytics might identify that someone with high blood pressure is more likely to develop heart disease. While this provides insight into which patients to watch, to really address the issue, practitioners would want to find out what causes high blood pressure and when it is likely to lead to heart disease. Even if the predictive model tracks multiple factors, such as diet, genetics, and stress, it would be difficult to prove causation between any. Experimental studies that manipulate one factor, such as exercise, and compare the results to a control group would have to be conducted to estimate the causal relationship. As Siegel says, “Predictions don’t help unless you do something about them. They’re just thoughts, just ideas.”¹⁷ The call to “do something” is about digging deeper to discover the “why.” 

Perpetuating Bias

Predictive analytics requires historical data to make predictions about future outcomes. Unfortunately, historical data sometimes includes biased and discriminatory patterns that become embedded into future predictions. Data sets are often reflective of social discrimination, so instead of removing human bias, predictive analytics can reinforce it.¹⁸

For example, in 2014, Amazon created a predictive analytics model to analyze resumes and identify top candidates for hiring. The model made predictions based on their current employees and past applicants. However, due to the well-known gender imbalance in the tech industry, the AI model was biased toward male candidates. It would reject resumes with words like “women” (e.g., President of Women in STEM program) and favor resumes with macho-sounding words like “executed.” Amazon tried to tweak the model but was unable to because the historical data reflected biased hiring practices. Amazon let go of the project, but the example demonstrated the potential ethical risks of using AI and predictive analytics in recruiting practices.¹⁹

Shifting Dynamics

As we’ve learned, it’s possible to predict future outcomes based on historical data using predictive analytics. However, our world is constantly evolving. Shifts regularly occur in market dynamics, customer behavior, and our environment. The data used by a predictive analytics model may, therefore, be outdated and not accurately reflect the current reality, which limits its ability to make accurate predictions about the future. 

The financial sector experienced this consequence of predictive analytics in the early 2000s. Financial institutions have long used predictive models to assess the risk of mortgage-backed securities. Essentially, banks sell loans to investors who then make money as homeowners pay back their mortgages each month. Though this allows banks to provide more loans, if too many people default on their loans, banks cannot pay back investors and must foreclose homes. This increases the supply of available homes, reducing their price. This is what happened in the 2008 financial crisis. The models had been built on historical data, which reflected rising house prices and low default rates. This caused banks to assume they could be more lenient with giving out loans and approving mortgages to suboptimal candidates with low income or bad credit. When there was a shift in the mortgage market, too many people defaulted on their mortgages. Many banks and investment firms lost billions of dollars and were declared bankrupt, causing the economy to crash.

The 2008 financial crisis showed that predictive models are a suboptimal tool for assessing risk, as there is deterioration of predictive performance when there is even a small deviation between the current reality and historical patterns.²⁰ 

Uber’s Predictive Edge

Uber has quickly replaced traditional taxi companies in getting customers to their destinations. Uber operates in over 70 countries, serving 25 million trips daily. Part of what makes the company so successful is its use of predictive analytics to enhance the customer experience.²¹ 

Have you ever wondered how Uber matches you with a driver? How do they predict your estimated time of arrival? How do they decide which route the driver should take? How do they determine surge prices? That’s all seamlessly sorted out, thanks to predictive analytics. Uber developed a machine learning platform called Michelangelo that collects and analyzes user data to refine its predictive models and adjust supply, ETAs, and prices accordingly.²² Despite the fact that Uber operates in a dynamic and unpredictable environment, its use of AI and predictive analytics helps it to make data-informed decisions.²³ 

Uber will analyze how many customers have historically requested a ride based on the time of day, the weather, and the area. By predicting how many people will use the app, it deploys a supply of Uber drivers to meet demand so you never have to wait for long.²³ For example, historically, there is likely to be greater demand on a rainy Friday night. Uber can deploy drivers to busy urban areas to ensure customers can quickly catch a ride. Uber will also use both historical and real-time data on traffic to predict how long it will take for your driver to get to you, which is the most efficient route for them to take, and what time you will be at your destination. 

Uber’s use of the sophisticated machine learning platform Michelangelo, which is constantly evolving its model as it gathers more data, has enabled it to make predictions more accurately to ensure an efficient process for both drivers and riders.²³ 

Predicting COVID Severity

We all experienced the “unprecedented” COVID-19 pandemic. Part of the challenge with the pandemic was that it was unlike anything we had experienced before, making it difficult to know how to effectively address it. We were still learning about the severity and risks of COVID-19, making it difficult to choose effective clinical courses for patients who had contracted the virus. However, as time went on, health organizations were able to use historical data to predict which patients may suffer more severe effects of COVID-19 and adjust their treatment course accordingly.

The National COVID Cohort Collaborative conducted a study in 2021 to build a model that would identify which patients were at greatest risk. The collaborative was a centralized electronic health record repository that had access to vast amounts of data related to the virus. They analyzed the data of 1.9 million patients who tested positive for the SARS-CoV-2 infection from 34 medical centers worldwide. They examined how characteristics such as gender, age, and race all related to the severity of their clinical course to identify patterns between demographics and outcomes such as discharge to hospice, invasive ventilatory support, and death. They also analyzed which treatment courses improved patient outcomes.²⁴

The collaborative used linear regression to test for differences in biomarker trajectories between severity groups. The study found that being male, African American, and obese were correlated with more severe outcomes. These findings allowed hospitals to allocate resources effectively and put high-risk patients on more intensive treatment courses.²⁴

How to Predict Mental Illnesses: The Digital Future of Mental Healthcare

People tend to be resistant to seeking help for dealing with mental illnesses, as it requires a lot of effort. As a result, we wait for our mental health to diminish to a dangerous state before we are motivated to get help. In this article, our writer, Sarah Chudleigh, explores how predictive analytics can address this issue. Similar to how fitness trackers can identify a downturn in physical well-being, digital mental health apps can analyze historical trends to find relationships between behavior and mental well-being. If the app notices that someone is exhibiting behaviors related to mental illness, it can nudge them to seek help. 

How Behavioral Science Can Make Airports Less Miserable

Agent-based modeling, which creates multiple independent entities that react in specific ways to stimuli, can be used for predictive analytics to understand and predict the way that individuals behave. Instead of seeing people as one undifferentiated group that behaves in the exact same way, it can predict how individuals may behave in interaction with one another. In this article, our writers, Shreya Jaiswal and Graham Smith, show how agent-based modeling can help improve the airline boarding process and make our travel experience more efficient and enjoyable. 

1. A4Everyone. (2016, July 26). 10 great data analytics quotes. A4Everyone Blog. https://blog.a4everyone.com/2016/07/26/10-great-data-analytics-quotes/
2. Halton, C. (2024, June 27). Predictive analytics. Investopedia. https://www.investopedia.com/terms/p/predictive-analytics.asp
3. Twin, A. (2024, February 23). Data mining. Investopedia. https://www.investopedia.com/terms/d/datamining.asp
4. Joubert, S. (2019, October 29). Statistical modeling for data analysis. Northeastern University Graduate Programs. https://graduate.northeastern.edu/knowledge-hub/statistical-modeling-for-data-analysis/
5. Robby. (2024, February 13). The history of insurance: Lloyd’s of London’s contributions. ActivityCovered. https://activitycovered.com/the-history-of-insurance-lloyds-of-londons-contributions/
6. Predictive Success Corporation. (2019, May 6). A brief history of predictive analytics. Medium. https://medium.com/@predictivesuccess/a-brief-history-of-predictive-analytics-f05a9e55145f
7. Ethington, C. A., Thomas, S. L., & Pike, G. R. (2002). Back to the basics: Regression as it should be. In J. C. Smart & W. G. Tierney (Eds.), Higher education: Handbook of theory and research (Vol. 17). Springer, Dordrecht. https://doi.org/10.1007/978-94-010-0245-5_6
8. After, Inc. (2018, December 28). A brief history of predictive analytics (Part 1). https://afterinc.com/2018/12/28/brief-history-predictive-analytics-part-1/
9. Campbell, H. (2008, February 6). Lloyd’s. Britannica. https://www.britannica.com/money/Lloyds#ref242838
10. Lloyd's Register. (n.d.). Edward Lloyd’s coffee house. https://www.lr.org/en/about-us/who-we-are/our-history/edward-lloyd-coffee-house/
11. The Editors of Encyclopaedia Britannica. (2025, February 12). Francis Galton. In Encyclopaedia Britannica. https://www.britannica.com/biography/Francis-Galton
12. The Predictive Index. (n.d.). The Predictive Index. https://www.predictiveindex.com/
13. HighRadius. (2023, November 24). Predictive analytics in finance: Guide. https://www.highradius.com/resources/Blog/predictive-analytics-in-finance-guide/#
14. Petrova, B. (2025, February 10). Predictive analytics in healthcare. Reveal BI. https://www.revealbi.io/blog/predictive-analytics-in-healthcare
15. Digital Defynd. (n.d.). Healthcare analytics case studies. https://digitaldefynd.com/IQ/healthcare-analytics-case-studies/
16. Dstillery. (2024, June 5). Exploring predictive behavioral targeting: What it is and how it works. https://dstillery.com/blog/exploring-predictive-behavioral-targeting-what-it-is-and-how-it-works/
17. Siegel, E. (n.d.). Predictive analytics: The power to predict who will click, buy, lie, or die. Goodreads. https://www.goodreads.com/work/quotes/21692562-predictive-analytics-the-power-to-predict-who-will-click-buy-lie-or
18. Reddy, M. (2024, September 9). Pros and cons of predictive modeling. Digital Authority Partners. https://www.digitalauthority.me/resources/pros-cons-predictive-modeling/
19. Weissmann, J. (2018, October 10). Amazon’s artificial intelligence hiring software discriminated against women. Slate. https://slate.com/business/2018/10/amazon-artificial-intelligence-hiring-discrimination-women.html
20. Kiefer, H., & Mayock, T. (2020). Why do models that predict failure fail? (CFR Working Paper No. 2020-05). Federal Deposit Insurance Corporation. https://www.fdic.gov/analysis/cfr/working-papers/2020/cfr-wp2020-05
21. Uber. (2024, May 2). From predictive to generative AI. https://www.uber.com/en-CA/blog/from-predictive-to-generative-ai/
22. Uber. (2017, September 5). Michelangelo: Uber’s machine learning platform. https://www.uber.com/en-CA/blog/michelangelo-machine-learning-platform/
23. CodeBasics. (2024, June 12). How Uber uses data analytics to increase supply efficiency. https://codebasics.io/blog/how-uber-uses-data-analytics-to-increase-supply-efficiency
24. Bennett, T. D., Moffitt, R. A., Hajagos, J. G., Amor, B., Anand, A., Bissell, M. M., Bradwell, K. R., Bremer, C., Byrd, J. B., Denham, A., DeWitt, P. E., Gabriel, D., Garibaldi, B. T., Girvin, A. T., Guinney, J., Hill, E. L., Hong, S. S., Jimenez, H., Kavuluru, R., Kostka, K., … National COVID Cohort Collaborative (N3C) Consortium. (2021). Clinical characterization and prediction of clinical severity of SARS-CoV-2 infection among US adults using data from the US National COVID Cohort Collaborative. JAMA Network Open, 4(7), e2116901. https://doi.org/10.1001/jamanetworkopen.2021.16901