Demystifying Decision Intelligence: Insights from Google Cloud‘s Cassie Kozyrkov

In the rapidly evolving world of data science and artificial intelligence, a new discipline has emerged that is transforming the way organizations make decisions – decision intelligence. And one of the leading voices in this field is Cassie Kozyrkov, Chief Decision Scientist at Google Cloud.

As an expert in statistics, machine learning, and decision making, Cassie has made it her mission to democratize decision intelligence and help businesses harness its power to drive better outcomes. I recently had the opportunity to speak with Cassie and dive deep into her thoughts on this game-changing approach.

What is Decision Intelligence?

Put simply, decision intelligence is a framework that brings together techniques from data science, social science, decision theory, and management science to help organizations make smarter, faster decisions. It leverages AI and machine learning to augment human decision making, but also incorporates best practices and processes to ensure optimal outcomes.

"Decision intelligence is really about extending data science with all of these tools and techniques to support organizational decision making," Cassie explains. "It‘s not just about building models, but thinking holistically about the entire decision making process and how we can use data and AI to improve it at scale."

Whereas traditional data science focuses on extracting insights from data, decision intelligence goes a step further by connecting those insights to specific actions and choices. It provides a structured approach to define decision problems, design and evaluate alternatives, and implement and measure the results.

Some key tenets of decision intelligence include:

• Framing decisions as optimization problems with clear objectives, constraints, and tradeoffs
• Leveraging AI/ML techniques like reinforcement learning, causal inference, and multi-armed bandits to identify optimal choices
• Incorporating behavioral science principles to account for cognitive biases and heuristics in decision making
• Designing human-in-the-loop systems that combine machine intelligence with human judgment
• Continuously measuring and improving decision quality using experimentation and feedback loops

By operationalizing these principles, organizations can make decisions that are more data-driven, transparent, and aligned with their goals and values.

The Evolution of Decision Intelligence

While the term "decision intelligence" has only recently gained prevalence, its roots can be traced back decades to fields like operations research, management science, and statistical decision theory. What‘s new is the confluence of advanced AI/ML capabilities, vast troves of digital data, and a growing recognition of the power of decisioning as a source of competitive advantage.

One of the seminal works in this space is James March‘s 1991 book "A Primer on Decision Making", which laid out a framework for understanding organizational decision processes. More recently, Stanford professor Yoav Shoham‘s 2018 paper proposed decision intelligence as a new academic discipline, crystallizing many of the key concepts.

Over the past few years, decision intelligence has begun to transition from a niche research topic to a mainstream business capability. Analysts like Gartner and Forrester have published extensive reports on the topic, and a growing number of startups and established vendors are building decision intelligence products and services.

Perhaps the most prominent champion of decision intelligence has been Cassie Kozyrkov and her team at Google Cloud. They‘ve not only been pioneers in applying decision intelligence internally across Google‘s operations, but also evangelizing the approach to the broader industry and developing tools to make it more accessible.

Decision Intelligence in Action at Google

Google is renowned for its data-driven culture and use of AI/ML across every facet of its business. But even for a company of its scale and sophistication, implementing decision intelligence has been a transformative force.

One high-profile example is Google‘s use of DeepMind‘s machine learning to optimize cooling in its data centers. By applying advanced reinforcement learning algorithms to control the cooling systems in real-time based on inputs like temperature, pressure, and power usage, DeepMind was able to reduce energy consumption by 40%. In a single data center, this translated to savings of hundreds of thousands of dollars per year.

Beyond infrastructure optimization, Google has applied decision intelligence to virtually every business function, from marketing and product development to HR and finance. Examples include:

• Optimizing ad targeting and pricing in real-time based on user behavior and auction dynamics
• Personalizing product recommendations and search results to maximize user engagement and satisfaction
• Forecasting demand and optimizing inventory and fulfillment across Google‘s global supply chain
• Identifying and mitigating risks like payment fraud, content violations, and account takeovers
• Streamlining customer support operations by dynamically routing cases and automating responses

What‘s notable about these applications is that they go beyond traditional prediction and classification tasks to directly guide and automate decisions. They encode business goals and constraints into AI/ML models and enable real-time optimization based on live data feeds.

Another distinguishing feature of Google‘s approach to decision intelligence is its emphasis on human-centered design and responsible AI. Cassie often stresses the importance of keeping "humans in the loop" and avoiding blind automation. Every decision system at Google is subject to extensive testing, monitoring, and oversight to ensure safety, fairness, and accountability.

For example, when Google developed a machine learning model to help detect breast cancer in screening mammograms, they didn‘t just deploy it autonomously. They first conducted rigorous clinical validation in partnership with academic medical centers, and continue to have the system‘s outputs reviewed by radiologists. The goal is to augment, not replace, human expertise.

The Business Impact of Decision Intelligence

Google‘s successes with decision intelligence are not just anecdotal – they‘re backed up by hard data. At Google Cloud Next ‘21, Cassie shared some statistics on the business impact of their decision intelligence initiatives:

• 2-3X increase in data science productivity by automating routine tasks and enabling self-service analytics
• 10-20% improvement in forecasting accuracy for demand planning and inventory optimization
• 5-10% reduction in customer acquisition costs through more targeted marketing and offers
• 2-5X speedup in time-to-insight for root cause analysis and anomaly detection

And Google is not alone. A 2020 Forrester survey of data and analytics decision makers found that enterprises adopting AI and decision intelligence drove 5-7% improvements in revenue, costs, and risks compared to peers. Gartner predicts that by 2023, more than a third of large organizations will be using decision intelligence for structured decision-making.

As more organizations climb the maturity curve from business intelligence to machine learning to decision intelligence, the competitive bar will continue to rise. Forrester calls this the "AI velocity divide", where laggards may quickly find themselves at a structural disadvantage to peers who can make smarter decisions faster.

Getting Started with Decision Intelligence

If you‘re inspired to start harnessing the power of decision intelligence in your own organization, Cassie has some advice on where to begin:

1. Identify your key decisions: Start by mapping out the most important decisions your organization makes, the frequency, and the business impact. Prioritize those that are high-volume, time-sensitive, and data-rich.

2. Break down silos: Decision intelligence requires close collaboration between data scientists, subject matter experts, and business leaders. Foster cross-functional teams and align on a shared vision and metrics.

3. Build the technical foundation: Ensure you have the necessary data infrastructure, tooling, and governance in place to support real-time, automated decision making. This includes things like data integration, feature stores, model monitoring, and API management.

4. Design for transparency and trust: Invest in techniques like model interpretability, bias testing, and human oversight to ensure your decision systems are fair, ethical, and reliable. Continuously monitor and tune performance based on feedback and changing conditions.

5. Cultivate a decision-centric culture: Ultimately, adopting decision intelligence requires a mindset shift from hindsight to foresight, from intuition to evidence, and from silos to systems. Lead by example and celebrate successes to drive cultural change.

To help accelerate this journey, Google Cloud has developed a suite of tools and services for decision intelligence, including:

• Vertex AI – An end-to-end platform for building and deploying ML applications, including support for AutoML, distributed training, model monitoring, and pipelines.

• Vertex Optimization – A managed service to frame and solve multi-objective optimization problems for use cases like portfolio optimization, resource allocation, and product configuration.

• BigQuery ML – Enables users to create and execute ML models in BigQuery using SQL queries for tasks like demand forecasting, customer segmentation, and anomaly detection.

• Cloud AI Building Blocks – A collection of plug-and-play AI components for common business tasks like document understanding, contact center assistance, and marketing optimization.

But tools are just one piece of the puzzle. Even more important is growing the community of practice around decision intelligence. To that end, Cassie and her team have launched a variety of educational resources and events, including:

• The Decision Intelligence Quickstart guide – A hands-on tutorial to teach the end-to-end process of framing a decision problem, designing a solution, and deploying it to the cloud.

• The Decision Intelligence Practitioners certification – An online training program and exam to equip data scientists and analysts with the skills to apply decision intelligence techniques on Google Cloud.

• The Decision Intelligence Summit – A virtual conference showcasing the latest research, products, and customer stories in decision intelligence, featuring speakers from Google, Accenture, Unum, and more.

By making decision intelligence more accessible and building a vibrant ecosystem around it, Cassie believes we can democratize this superpower for organizations of all sizes.

The Road Ahead for Decision Intelligence

As powerful as decision intelligence is today, we‘ve only begun to scratch the surface of its potential. As Cassie describes it, "Decision intelligence is going to be the defining technology trend of the decade – it‘s like the smartphone of enterprise software."

In the near future, she envisions decision intelligence becoming the default operating system for businesses, orchestrating millions of interconnected decisions across every function and process. It will enable what "black box" AI cannot – transparent, adaptive, and ethical decision making at scale.

Some of the emerging frontiers in decision intelligence that Cassie is excited about include:

• Reinforcement learning: Using techniques inspired by behavioral psychology to train AI agents through trial-and-error to make optimal decisions in complex, dynamic environments.

• Causal inference: Combining statistical methods with domain knowledge to tease out the causal relationships between variables and inform interventions and counterfactuals.

• Digital twins: Building high-fidelity virtual models of physical assets, processes, and systems to simulate scenarios and optimize decisions before deploying in the real world.

• Collaborative intelligence: Designing AI systems that can work seamlessly with human teams, leveraging the complementary strengths of human and machine intelligence.

• Secure multiparty computation: Enabling multiple parties to contribute data and models to make collective decisions while preserving privacy and security.

The unifying thread across all of these trends is a shift from artificial intelligence in the narrow sense of emulating human cognition, to a broader vision of augmented intelligence – enhancing human agency with machine capabilities.

As Cassie puts it, "The point of decision intelligence is not to replace humans, but to empower them. To give them superpowers to make better decisions, faster, in ways that align with their values and goals. That‘s the real promise of this technology, and it‘s what inspires me every day."

While there will undoubtedly be challenges and risks along the way – from data access and quality to model explainability and control – the potential benefits are too great to ignore. By some estimates, applying AI to decision making could drive $13 trillion in economic value annually by 2030.

More importantly, decision intelligence could help solve some of the world‘s most pressing problems, from climate change and public health to education and social justice. By surfacing data-driven insights and enabling better coordination and optimization, it could spur breakthroughs in scientific discovery, policy making, and global development.

As Cassie says, "Decision intelligence is not just a technology, it‘s a movement. A way of thinking that puts decisions at the center and uses every tool in our arsenal to improve them. And I believe it‘s going to change the world."

A Call to Action

So what can you do to be a part of this movement? Whether you‘re a data scientist, business leader, or concerned citizen, there are many ways to get involved:

1. Educate yourself: Take advantage of the growing body of resources on decision intelligence, from online courses and certifications to books and research papers. Follow experts like Cassie Kozyrkov and organizations like the Partnership on AI to stay up-to-date on the latest developments.

2. Experiment and iterate: Start small by identifying a specific decision process to improve and testing out different decision intelligence techniques. Measure the results and continuously refine your approach based on feedback.

3. Advocate for responsible AI: Hold yourself and your organization accountable to the highest standards of ethics, fairness, and transparency in your use of AI and decision intelligence. Engage in public discourse and policy making to ensure these technologies are developed and deployed in service of the greater good.

4. Collaborate and share: Decision intelligence is an inherently interdisciplinary field that requires diverse perspectives and skill sets. Seek out opportunities to collaborate with experts from different domains and share your learnings with the broader community.

5. Dream big: Don‘t be afraid to tackle the big, hairy, audacious problems that decision intelligence could help solve. As Cassie says, "The only way we can create the future we want is by envisioning it and then working backwards to make it a reality."

The age of decision intelligence is upon us, and the opportunities are limitless. By harnessing the power of data, AI, and human intuition, we can make smarter choices, drive better outcomes, and ultimately create a better world. So let‘s roll up our sleeves and get to work – the future is ours to decide.