As a new data scientist, you’re struggling with a large data set and how you can make it tell a story. The stakes are high; you’ve been told by your manager that this data should inform strategic thinking at your company for approaching new AI methods. Before you get even more lost in your data, a colleague mentions a platform that pairs causal modelling and real-time feedback. Curiously, you dig deeper and realize this tool doesn’t just visualize data nicely—it also models outcomes, displays trade-offs, and highlights when decisions are deviating from long-term goals. With this solution, decision intelligence is to the rescue. 

Decision intelligence (DI) transforms raw data into practical, strategic choices by integrating artificial intelligence, data science, and decision theory.^{1, 2} DI can be seen as a bridge between analytics and action by helping organizations make quicker, more reliable choices. In its repertoire of multi-disciplinary uses, DI is able to consider novelty in company decision-making on several levels, ranging from strategy to operations, with context in mind.

It might be the first time you’ve heard about decision intelligence, but you may recognize its close relative, business intelligence (BI), and undoubtedly have heard about artificial intelligence once or twice. Depending on who you ask, DI and AI have vastly different scopes compared to their neighbor, BI. The lack of a standard definition differentiating these related concepts can make the field of DI feel rather muddy, so let’s make these distinctions before going deeper:^{1, 3}

DI assumes that the choices an organization makes are founded in recognizing that behaviors lead to results. Due to this inherent loop of cause and effect, DI analyses are often transformed into visualizations as a clear way to show what these relationships look like in reality. The data that DI deals with can come from many places, like structured data and unstructured data.³ In our efforts to get the story straight, our mutual friends AI and machine learning supplement how we find data patterns, anticipate trends to come, and make behavioral insights.¹ At its best, decision intelligence makes someone’s data story coherent and compelling without the prerequisite of being a data scientist. 

The key elements of decision intelligence in practice draw from fields related to AI and automation, as well as data and analytics. Some of these elements might include:³

Instead of the fallible gut instinct or the isolated siloing of decision-makers, decision intelligence gives a structured, reusable framework for navigating uncertainty and complexity.^{2, 3} Choices are not so simple nowadays, but they can be made more straightforward thanks to DI. In its many layers of applications, decision intelligence must consider the decision architects themselves—us humans, and how we make up our minds in relation to non-human intelligence.

Blending human and machine judgment

Don’t worry, DI isn’t taking your job: decision intelligence isn’t a replacement for human judgment, but an enhancement of it. Humans can be involved in AI decision-making processes to different degrees, with choice architecture oriented in one of three ways:³

1. Decision Support: Humans are “in” the loop with AI, as it gives us behavioral insights and makes simulated possibilities of paths not yet traveled.
2. Decision Augmentation: Humans are “on” the loop with AI, as it creates advice knowing that context matters and what impacts may occur. When humans are on the loop, we make the last call when it comes to making, amending, or rejecting a decision. 
3. Decision Automation: Here, humans are completely “out” of the loop, AI is acting on its own accord for our decision-making as it behaves within the parameters we’ve established. These behaviors can be audited, so we can keep track while outside the loop. 

This discipline can integrate behavioral science and systems thinking as it accounts for how real people interpret, respond to, and shape decisions in dynamic environments. Rather than replacing human judgment, decision intelligence provides support by making mental models visible or finding options we might otherwise overlook as human users. In this way, decision intelligence acts less like a final answer or a substitution for humans—it’s more like your data-friendly decision-making partner.

Structured Data: Data that is organized into defined fields and formats, like rows in a spreadsheet or values in a database. Structured data makes it easy for decision intelligence systems to detect patterns and simulate outcomes. An example is a DI model that predicts the demand of a company’s sales across regions and time zones.

Unstructured Data: Includes messy, free-form inputs like text, images, or voice. This type of data requires extra processing before it can inform a decision. An example is a DI platform that can look into chat transcripts to identify customer pain points.

Machine Learning: A type of AI where systems learn from data to make predictions or classifications, often used in decision intelligence to recommend actions or flag risks. DI may apply machine learning when predicting supply chain delays relative to a factor like weather patterns.

Choice Architecture: The way options are presented to decision-makers that subtly shape behavior without removing freedom. In DI, choice architecture ensures that AI-generated insights are surfaced in ways that support good judgment.

Decision Analysis: A formal approach to evaluating choices by weighing risks, benefits, and trade-offs. As a common backbone of decision intelligence frameworks, an instance is when a healthcare provider compares patient treatments relative to costs, outcomes, and what resources are available.

Systems Thinking: A holistic way of seeing how decisions are interconnected within a larger ecosystem, helping identify ripple effects and avoid siloed solutions. A city might apply DI and systems thinking to reduce traffic congestion by adding more bus routes and running transit simulations. 

Long before the Internet and other revolutionary technologies of today, innovative thinkers were pondering the science behind our decisions. In the 1960s, Howard Raiffa’s lectures on “Choices under Uncertainty" served as the roots of decision analysis, a predecessor of the later data-driven discipline of decision intelligence.⁴ During this time, decision-makers developed new ways to assess pros and cons through tools like decision trees. These contributions transformed how we think about decisions, their ambiguity, and permutations of outcomes prior to formal subject fields.

Progress in the field of decision science began at a slow pace. It was later in the 1980s when adjacent developments of DI were blossoming, with introductory computer-based decision support systems (DSS).⁴ As monumental tools for decision-making at the time, these programs allowed complex problems and scenario-based thinking to be visualized—acting as a decision-making sidekick long before we could lean on ChatGPT. In the 1990s, bigger leaps and bounds were made, as DSS met data refinement via online analytical processing (OLAP), thanks to Edgar F. Codd.⁴ DSS and OLAP crossover allowed data scientists to capture, analyze, and appraise data from a variety of places for decision-making. 

In the 1990s, the concept of what we now called decision intelligence was put in writing when thought leader James G. March published his book A Primer on Decision Making: How Decisions Happen in 1994.⁵ March spoke to the frustrating experience of pursuing decision intelligence as a complex process that requires patience—it doesn’t simply happen on the spot when we make a decision. At the start of the millenium in the early 2000s, some caught onto March’s message, including Uwe Manning, who described DI as the sum of business intelligence and knowledge management in 2002.⁵ DI’s growth was picking up pace, gaining popularity as people saw its potential for data collection, data analysis, and insight generation. 

The 2010s marked a formative period for DI, as some of its notable proponents, Dr. Lorien Pratt and Mark Zangari, founded “decision engineering” in 2010, which was soon renamed decision intelligence in 2012.⁶ It wasn’t until this time, decades after March’s book, that DI truly caught on. Before the wider endorsement of DI from Google, Pratt made the concept more available to the general public as a discipline where data-driven behaviors transform into real outcomes. Around 2017, DI exploded when then-chief decision scientist at Google, Cassie Kozyrkov, who is often credited with making the field what it is today, stressed the importance of trying to democratize decision intelligence for any decision-maker to make data tell meaningful stories with accountability and ease of understanding.⁷

For a concept that feels so vital to modern operations, what took decision intelligence so long to take off? With Gartner’s naming of decision intelligence as a key business strategy for all companies in 2022, questions remain about how DI fits into the landscape of contemporary AI tools in 2025 (CITE - Forbes). What started as a niche methodology is now seen as a core capability for digital-era decision-making. The role of decision intelligence in a world teeming with artificial intelligence continues to evolve, as decisions may be less and less limited by the bounds of human intelligence. 

Howard Raiffa

A pioneer in decision analysis, Raiffa helped formalize how individuals and organizations navigate uncertainty, blending statistical reasoning with behavioral insights to guide rational decision-making under risk.

Edward F. Codd

A computer scientist who revolutionized data systems by developing the relational database model, laying the foundation for OLAP, and enabling multidimensional analysis that powers modern decision intelligence tools.

Dr. Lorien Pratt
A machine learning and decision intelligence consultant who was a co-founder of DI. Pratt coined the term in 2012 alongside Mark Zangari while working on decision modeling frameworks at Quantellia, a DI software company. Her work focuses on connecting machine learning to human goals through explainable, outcome-driven systems.⁶

Mark Zangari
As co-founder of Quantellia, Zangari helped establish the earliest decision intelligence platforms, emphasizing the integration of causal modeling, systems thinking, and business strategy. His work laid the groundwork for applying decision intelligence across enterprise settings.⁶

James G. March

A social scientist who challenged rational models of decision-making, introducing widespread opinions about the often illogical nature of real-world organizational choices. March was one of the first to use the term “decision intelligence” in his 1994 book, A Primer on Decision Making: How Decisions Happen.

Uwe Manning

In 2002, Manning proposed that decision intelligence emerges at the intersection of business intelligence and knowledge management—an early articulation of DI as a synthesis of data, context, and human judgment.

Cassie Kozyrkov

A leading voice of decision intelligence, Kozyrkov popularized the term while at Google in the late 2010s at the intersection of data science, psychology, and decision science.⁷ Kozyrkov's work argues that AI and data science should move toward more human-centered, decision-driven applications where analytics act as a service to decision-makers over mere technical output.

Decision intelligence isn’t just about making decisions; it’s about making the right ones. Smart decisions are not just faster, they are also in greater alignment with a company’s strategy and behaviors. Further, at the heart of decision intelligence is its ability to help humans and AI work together more closely.

Smarter, faster, and larger-scale decision making

Decision intelligence enables organizations to automate complex choices and respond more quickly to dynamic business environments.^{8, 9} By integrating AI, causal modeling, and real-time data streams, DI reduces the lag between identifying a problem and actually acting on it. This kind of scaled, automated reasoning is especially useful when decision volume is high. Some examples may be managing global logistics, customer interactions, or fraud detection. These are often settings where manual decision-making simply can’t keep pace, but decision intelligence can.^{8, 9} 

What makes an equal impact with decision intelligence is its speed and its quality at scale. DI systems simulate different outcomes and can simultaneously find the best options based on evolving business priorities. For example, a telecom company might use DI to reroute technicians in real time based on weather disruptions and service request urgency. In moving beyond reactive analytics, organizations anticipate change and act proactively, all while staying aligned with broader strategic goals.

Aligning strategy and action

One of the biggest challenges in complex systems is turning strategy into consistent, on-the-ground action. Decision intelligence addresses this gap by embedding goals directly into decision flows. In practice, robust DI ensures that big datasets are organized, context is accounted for, and intent is captured—all at once and in harmony.^{1, 2} When an organization struggles to define a vision in action, DI helps to find the necessary steps. Whether you're aiming to reduce carbon emissions or boost customer retention, DI systems can ensure each small decision nudges a company in the right direction.

Think of a retail brand aiming to improve sustainability. With DI, supply chain decisions like choosing materials, managing inventory, or selecting shipping methods can all be made to align with environmental targets. Employees are no longer guessing how their decisions connect to the bigger picture; they feel coherence in how they’re supported by systems that exemplify trade-offs with long-term goals in focus.

Human and AI Collaboration

The rise of automation does not negate that complex decisions still require human judgment. Decision intelligence is designed with this in mind: it is key to understand that it doesn't replace people, but supports them by making invisible variables visible and gathering team insights they can trust.^{8, 10} In high-stakes domains like healthcare, public safety, or finance, DI creates room for humans to weigh ethical considerations relative to lived experiences, paired with AI-driven advising.

Just like a good startup, what makes this especially powerful is its role in democratizing decision-making. DI platforms might include features ranging from user-friendly dashboards to low-code tools. This enables collaboration between the realistic range of both technical and non-technical stakeholders in any workplace. For example, a financial analyst and a compliance officer can co-develop a risk assessment model without needing to code it from scratch using DI. This co-creation amongst coworkers builds transparency and shared ownership between not just mere human colleagues, but with the DI tool itself—making AI-supported decisions naturally collaborative among their human users.

While decision intelligence holds major promises, it’s not immune to design flaws, implementation gaps, or overreliance on automation. These limitations remind us of the ongoing importance of human judgment, where a careful investigation of both data and context is necessary.

Dependency on high-quality data

Decision intelligence systems are only as strong as the data they’re built on. If the input data is incomplete, outdated, or biased, even the most advanced models can reinforce poor assumptions or produce misleading outputs.^{1, 10} In domains like finance or healthcare, where high-stakes decisions depend on accuracy, the consequences of poor data can be severe as seen in possible consequences like resource misallocation, missed risks, or eroded trust in the system.

Many organizations underestimate the effort required to prepare data in general, and this is especially so for DI use. Real-world data can be easily lost, unstructured, or spread across incompatible systems. The action of cleaning and contextualizing this data takes time as well as cross-team collaboration when decisions rely on sensitive variables like location or behavior. Without strong data governance for quality control, the promise of smarter decision-making can quickly turn into a false sense of confidence.

When silos halt scaling

One of DI’s core strengths is connecting strategy with day-to-day decisions, but it can quickly become a weakness when organizational silos (or, bureaucracy itself) get in the way. Competing incentives, fragmented workflows, and unclear decision ownership often prevent systems from scaling beyond a single use case or department.^{2, 10} For instance, a retail team using DI for inventory might make choices that conflict with sustainability goals tracked by a separate team.

This disconnect is rarely about the technology: it’s about the people and processes around it. Successfully embedding DI requires clear alignment on priorities, open data sharing, and collaborative infrastructure across teams. Without this, models may optimize for local gains while undermining enterprise-wide outcomes. Just like with human teams, decision systems need clear communication and coordination to appreciate the shared context for the best functionality possible.

Not everything can be automatic 

Despite its power, decision intelligence should never be mistaken for a fully autonomous solution. When teams rely too heavily on algorithmic outputs without being skeptical of their assumptions, they risk making decisions that ignore social, ethical, or contextual nuance.^{8, 9} One instance could be an automated loan approval system that may unintentionally reinforce systemic bias if fairness isn’t explicitly included in its decision criteria.

This is why human oversight isn’t just helpful here, it's an essential feature of DI design. True decision intelligence acts as a decision partner instead of a final authority. Teams need the skills and structures to interpret outputs critically, challenge model logic, and adapt recommendations to real-world complexity. In ambiguous or high-impact settings, the intelligent design aspects of humans—like judgment, empathy, and accountability—remain non-negotiable parts of the decision process. Sure, DI can support these values, yet we must not assert that it can replace them.

Clinical decision intelligence for cancer care in Japan

In complex medical environments, making timely, informed decisions can mean the difference between early intervention and missed diagnosis. At Japan’s National Cancer Center, a team of researchers set out to bridge this gap to get diagnoses right with punctuality using decision intelligence. Their system, MedTAKMI-CDI, combines the analytical strength of machine learning with the contextual nuance of clinical expertise to support decision-making across a database of over 7,000 patients.¹¹ Rather than relying on a rigid relational schema, the system uses a flexible metaschema that allows researchers to navigate patient data fluidly—without needing to reload or restructure the dataset.

MedTAKMI-CDI enables what the researchers call clinical decision intelligence (CDI): an approach that merges domain knowledge with empirical learning to extract actionable patterns from messy, time-sensitive health records.¹¹ The system uses a three-layered model of abstraction ranging from basic attribute-value pairs to complex, time-stamped event sequences. This blueprint supports a variety of tasks from demographic profiling to high-stakes analysis of patient treatment pathways. Clinicians can zoom in on an individual’s timeline of care, or zoom out to identify emerging patterns across hundreds of cases—revealing which interventions tend to improve outcomes for particular cancer types or demographic profiles.

What sets MedTAKMI-CDI apart is how it handles uncertainty. Traditional systems require data to be formatted perfectly; MedTAKMI, on the other hand, adapts dynamically to new forms of input. By integrating domain ontologies and abstraction layers, the system makes it possible to explore data intuitively, offering a more human-aligned interface for clinicians who aren’t necessarily data scientists. In practice, this meant oncologists could ask high-level questions—such as which treatment pathways are associated with faster recovery in older patients—and receive evidence-based, interpretable answers in return.

This hybrid approach reflects the broader promise of decision intelligence: the ability to blend machine learning with human reasoning in context-rich environments. Rather than replacing clinical judgment, systems like MedTAKMI-CDI enhance it by offering doctors a clearer picture of complex patient journeys, and surfacing patterns that might otherwise remain buried in spreadsheets or siloed notes. As health systems around the world face increasing pressure to personalize care at scale, decision intelligence offers a pathway that’s both rigorous and responsive to human complexity.

Does Therabot listen? AI-powered mental health support

In 2025, Dartmouth’s AI and Mental Health Lab released Therabot, the first fully generative AI therapy chatbot tested in a randomized clinical trial.^{12, 13} Over four weeks, 106 participants diagnosed with depression, anxiety, or eating disorders interacted with Therabot via smartphone, typing prompts and receiving open-ended, CBT-based guidance. The app led to significant symptom reductions: 51% for depression, 31% for anxiety, and 19% for eating disorder concerns—rivaling outcomes of traditional outpatient therapy.^{12, 13}

A standout finding was the strong therapeutic alliance users formed with Therabot. Participants reported levels of trust and rapport comparable to those with human therapists, frequently seeking the bot late at night for relief and companionship.¹² On average, users spent six hours communicating with the chatbot over the course of the trial—equivalent to eight traditional therapy sessions—demonstrating both sustained engagement and perceived trust in AI-mediated care.¹²

Beyond the numbers, the study underscores critical elements of decision intelligence in action. The system combines real-time analysis of user inputs with CBT-informed response logic, illustrating human-in-the-loop design and context-aware decision-making. AI suggests intervention paths, but clinical oversight ensures only safe and relevant guidance is delivered.¹² This structured feedback loop also supports continuous learning, where the system adapts to user needs and flags when clinical escalation is necessary.¹³

In a sector strapped for resources, Therabot illustrates how decision intelligence frameworks can deliver scalable, personalized solutions while preserving human judgment and oversight. It shows that when AI is purpose-built with transparency, empathy, and safety at the core, it can meet clinical standards and enhance access—without replacing the therapeutic relationship itself.^{12, 13}

Emotional Intelligence

With all this talk of digital-oriented intelligence, we must continue to underscore the value of raw, human types of intelligence. In this piece, TDL writes about emotional intelligence: how it steers our behaviors, its social impacts, and why it is an important type of intelligence to have today. 

Decision Matrix

Decision intelligence isn’t the only avenue for organizing complex, data-based decision-making. In this piece, TDL columnist Isaac Koenig-Workman dives into what decision matrices are, how they can be used to streamline choices, a unique case study on involuntary mental health care, and more!

1. Kerner, S. M. (2024, October 31). What is decision intelligence? TechTarget. https://www.techtarget.com/searchbusinessanalytics/definition/decision-intelligence
2. Hoque, I. (2024, October 22). What is decision intelligence? Quantexa. https://www.quantexa.com/resources/what-is-decision-intelligence-guide/
3. What is decision intelligence? (n.d.). Aera Technology. https://www.aeratechnology.com/what-is-decision-intelligence
4. Delgado, C. O. (2023, May 25). A Brief History: From Decision Analysis to Decision Intelligence. LinkedIn. https://www.linkedin.com/feed/update/urn:li:activity:7067494749548498944/
5. Moser, R. (2021, May 3). Decision Intelligence: A Primer — Part 1. Medium. https://roger-moser.medium.com/decision-intelligence-a-primer-part-1-25c74139eb41
6. What's the future of decision intelligence? (2023, April). HyperFinity. https://hyperfinity.ai/resources/what-is-the-future-of-decision-intelligence
7. Cassie Kozyrkov - Chief decision scientist at Google. (2021, April 8). Barry Kibrick Between the Lines. https://www.barrykibrick.com/blog/cassie-kozyrkov-chief-decision-scientist-at-google
8. Kuttappa, S. (2020, June 7). The rise of decision intelligence: AI that optimizes decision-making. IBM. https://www.ibm.com/think/insights/the-rise-of-decision-intelligence-ai-that-optimizes-decision-making
9. Laluyaux, F. (2023, November 1). Decision Intelligence: A Business Imperative. https://www.nasdaq.com/articles/decision-intelligence-a-business-imperative. 
10. Emini, M. (2022, September 16). Why Decision Intelligence Is The Next Digital Transformation. Forbes. https://www.forbes.com/councils/forbestechcouncil/2022/09/16/why-decision-intelligence-is-the-next-digital-transformation/
11. Inokuchi, A., Takeda, K., Inaoka, N., & Wakao, F. (2007). MedTAKMI-CDI: Interactive knowledge discovery for clinical decision intelligence. IBM Systems Journal, 46(1), 115-133. https://doi.org/10.1147/sj.461.0115
12. Kelly, M. (2025, March 27). First therapy chatbot trial yields mental health benefits. Dartmouth. https://home.dartmouth.edu/news/2025/03/first-therapy-chatbot-trial-yields-mental-health-benefits?utm_source
13. Wel, M. (2025, April 1). AI therapy breakthrough: New study reveals promising results. Psychology Today. https://www.psychologytoday.com/us/blog/urban-survival/202504/ai-therapy-breakthrough-new-study-reveals-promising-results?utm_source