main banner image

The Winning Formula: Merging Human Ingenuity with AI in the Race for Excellence

Ari Kaplan: AI's Intersection with Sports and Business

Explore the fascinating journey of Ari Kaplan, Head Evangelist at Databricks, as he shares his unique perspective on blending AI with sports analytics and business strategies. Delve into his insights from his time with the McLaren Formula 1 team and Datarobot, uncovering the impact of AI in high-stakes scenarios.

Pioneering Analytics in Formula 1 and Beyond

From predicting race outcomes to revolutionizing team strategies, Kaplan discusses the crucial role of data-driven decisions in Formula 1 racing and how these methodologies are mirrored in business. Learn about his groundbreaking work with the Chicago Cubs and the role of AI in building championship teams.

“With AI, our greatest legacy will be how we use this technology to empower our lives, not overshadow them. It’s our job as leaders to teach people by leading the way with AI as our tool, not our master.”
Ari Kaplan

Head Evangelist @ Databricks

Welcome, Ari Kaplan! Your journey in merging the worlds of sports, AI, and business analytics has been groundbreaking. We're excited to delve into your insights and experiences in this exclusive interview, and even more excited for your Opening Keynote at the “Leaders in AI Summit” in Austin, Texas at the Circuit of Americas F1 Racetrack, where you probably know every single turn in your journey across F1!

Formula 1 Analytics Applied

Bruce: “Reflecting on your time with McLaren's Formula 1 team and Datarobot, can you highlight a pivotal moment where your analytics approach made a significant impact on a race strategy or outcome? How does this experience mirror the application of AI and analytics in high-pressure business scenarios?"

Kaplan: Formula 1 is one of the most data-intensive events I’ve ever seen, and it was such a fantastic adventure to travel the world with the McLaren race team. Abu Dhabi, Melbourne, London, Saudi Arabia, and even Circuit of Americas in Austin - where our event will be.

There are hundreds of sensors on each car transmitting thousands of times per second, to be processed, analyzed, and sent back to the race team in real-time for decisions: When to do a pit stop? Which tire types to use? What are the top competitive recommendations to the driver right now?

There were so many moments where our analytics helped the outcome, and one example was predicting likelihood of precipitation, surface temperatures of the track, and air temperatures around the track. We were able to predict weather more precisely in 5-minute intervals, way better than the local weather forecasts. This helped in several races, giving McLaren insights on the track cooling down and rain not falling, giving confidence for several extra laps before opting for a pit stop in Silverstone, in Singapore, and elsewhere. Details are in this article. That was huge!

Bruce: How does this experience mirror the application of AI and analytics in high-pressure business scenarios?

Kaplan: People love sports analytics analogies, because it does indeed mirror AI and data-driven insights in every single industry. Being able extract, transform, load data at scale. Harmonizing structured and unstructured data together. Integrating real-time data - whether racing or social media feeds or IoT sensors. And enabling your team to collaborate - McLaren has hundreds of data engineers and data analysts working every single race day. This experience is the same when a business provides their own insights at their own speed of business. Faster detection of airline delays, real-time chatbots, determining fraud while a purchase is being made, just-in-time marketing - all are examples. Formula 1 is high stakes, high visibility, and milliseconds count!

Data-driven Decisions in Racing

Bruce: In the intense world of Formula 1 racing, where decisions need to be made in fractions of a second, how has the incorporation of data analytics transformed team strategies and race-day outcomes? Could you share an example that particularly stands out?

Kaplan: Note: this seems to be a repeat of the first question} In addition to the weather example, there are many other use cases that stand out. Each team has a wind tunnel that use AI to help decide which of the 80,000 car parts should be tested - since only a few tests can be run per day. AI helps run millions of simulations to guide what to test in real life, that will most likely improve the car’s aerodynamics? Another example is looking at past race information and determining causality of how the car performed as expected or not. This is crucial to do quickly because race teams are off to the next country within a day or so, and testing on the next track is only days away!

Moneyball Revolution in Chicago

Bruce: Your work with the Chicago Cubs is often likened to the 'Moneyball' revolution in baseball. Can you take us through the analytics-driven approach that led to breaking the team's long-standing World Series drought?

Kaplan: It has been quite an honor to have worked many decades in the game I love - baseball. Starting as one of a handful of people working with a team in the ‘80s to seeing the industry grow to tens of thousands of people. With the Cubs, I began consulting with their scouting department beginning in the ‘90s, to convert paper-based player reports into self-service databases that non-technical staff could query directly, along with assisting the GMs with player acquisition and salary negotiations. When new ownership bought the organization, I was brought in to create and lead their entire analytics efforts. We wanted to build a more data-driven approach everywhere it would help win, improve fan engagement, or drive revenue: in-game strategy, player forecasting, player development, injury risk, blending subjective scouting observations with objective biomechanics, and also on the business side with pricing, marketing, customer segmentation, and so much more. It took a team to make it happen.

Analytics in Team Building

Bruce: In building a championship team like the Chicago Cubs, what role did analytics play in scouting and team formation? How do these strategies translate to assembling successful teams in the corporate world?

Kaplan: Scouting is the lifeblood of any sports organization, getting the right players with the right strengths for the right roles on your team. This translates to any business - recruiting the right workers based on both their soft and hard skills, assembling a team of people where each individual’s work role aligns with their best abilities. Also, collaboration plays a central role. Having a unified data platform enables more people to contribute and to benefit - unlocking better strategies. It’s democratization so non-technical people can get their own insights: in sports players and coaches can ask questions without programming knowledge ; in the corporate world it’s line-of-business workers. It’s workflow efficiencies so that people are building upon others works and not starting from scratch each time: data engineers doing ETL together with data scientists creating ML models together with visualization experts and storytellers.

AI’s Role in High-Performance Industries

Bruce: In your perspective, what are some groundbreaking instances where AI has perfectly complemented human skill and ingenuity to create a winning formula, particularly in industries that demand high performance?

Kaplan: In sports, there is a perfect harmonization of human observations (scouts watching young players perform, writing text summaries of their sentiment) and data collection (biomechanical techniques of pitching and hitting). Data science can take both the structured and unstructured data together and determine signal in a vast amount of information. For example, when a scout says “their curveball is deceptive” or “they are not afraid to crowd the plate”, and the metrics show the curveball’s velocity and spin is exceptional, or their batting stance is consistent - then forecasts are much more accurate.

There are so many other examples across industries. The healthcare field is one of my favorites because it saves and prolongs life. Medical data (structured such as blood counts, unstructured such as x-rays) helps human doctors (who look at the patient holistically and where empathy and communication matter for healing) work together for the best patient outcomes.

Navigating AI Challenges in the Modern Era

Bruce: Considering the fast-paced advancements and widespread adoption of AI technologies post-CHATGPT, what are the most pressing challenges you see businesses facing today in leveraging AI effectively?

Kaplan: These are exciting times with GenAI innovation. Every businesses I speak with all looking to implement new GenAI use cases. But because of many challenges, enterprises actually going into production is still very low - something like 10%-20% depending on how you define things.

Many challenges include lack of transparency, lack of skills to build and deploy, hallucinations, and difficulty to govern. With transparency there is a lack of trust, so companies want to know what data is underlying the models, what the lineage is, and how models are being computed. With a lack of employees with skills, solutions that automate the RAG, fine-tuning, and fully building LLMs are crucial. Hallucinations will take time to overcome, with humans guiding the prompting and evaluating the model results. And governance is addressed through unified platforms for access control, operationalizing the LLMs, comparing model effectiveness, and chaining models together for the most valuable insights.

AI Safety and Ethical Implementation

Bruce: In your diverse experience across sports and business, what key principles should organizations prioritize to ensure the ethical and safe implementation of AI systems?

Kaplan: For both safety and ethical implementations, data governance and guardrails need to be implemented, along with leadership buy-in and business oversight. For example, there needs to be access control not only on the underlying data driving the models, but also on the entire lineage through to the models and code itself. Privacy of medical and personal information, for example, needs to be protected in models, LLM training, and notebook code itself. For ethics beyond privacy, companies need to evaluate their data policies and ensure things like data bias, training bias, and model bias are detected and if needed addressed.

Effective AI Governance

Bruce: From your viewpoint, what are the critical components of establishing a robust governance framework for AI? How can organizations implement these to ensure responsible AI usage?

Kaplan: Governance frameworks ensure data and AI products are consistently developed and maintained, adhering to precise guidelines and standards. It's the blueprint for architects, bringing their solutions and data vision to life with consistency, guidelines, and standards. It's scale and speed for data engineers with repeatable workflow management. It's collaboratively building and operationalizing AI models for data scientists, with transparency to operationalize at scale. It's security for data managers, ensuring data assets are shared far and wide to benefit all, yet private when needed. It's trust for executives, with transparency of business insights based on their data and AI assets. And it all drives operational efficiency for finance.

Many organizations have seen the importance of governance frameworks for information security, access control, usage monitoring, enacting guardrails, and obtaining "single source of truth" insights from their data assets. As these organizations grow, these governance challenges compound and without a robust governance framework, traditional governance solutions no longer adequately meet their needs. Data proliferates, so new unstructured and streaming data sources are added to their traditional data warehouses; divergent technology from multiple vendors transforms into never-ending and risky patchwork solutions; and of course, their assets devolve into "data swamps".

What are the components of a robust governance frameworks for AI?

  • Mitigate risk and ensures compliance with centralized auditing across platforms, data lineage, data usage tracking, and auditability.
  • Reduce platform complexity/cost by governing all data assets including files, tables, ML models, and dashboards, no matter where they live - across clouds and data platforms.
  • Accelerate innovation with automation of data engineering and AI. This frees staff from an average of 80% of their time-consuming and repetitive tasks to focus on innovation and monetizing value.
  • Democratizate data and AI to extend innovation to business analysts and line-of-business personas.

AI Leadership in the Current Landscape

Bruce: Leadership in the era of AI requires unique qualities. Based on your experience, what are these essential leadership traits, and how can they be cultivated in today's AI-driven organizational environment?

Kaplan: Ensuring you achieve success and get the best value from AI requires a combination of skills: probability and statistics ; coding ; and understanding of the business. The best leadership skills include the ability to collaborate, to listen with empathy, to communicate, and see the “big picture” and “question behind the question”. And no matter what you need to have a natural curiosity and willingness to take constructive feedback, try new things, fail fast, and automate processes whenever possible.

AI Transformation Expectations Management

Bruce: In the rapidly evolving AI landscape, how can companies manage internal and external expectations to align with the realistic capabilities and potential of AI technologies?

Kaplan: This is an important topic, as so many executives are actively pursuing AI as a concept or as an experiment, without clear direction to what constitutes success. For example, what decides if an LLM is production-ready or “not quite good enough”? Who decides that? Is there a metric for accuracy, or a “gut feel” still? Who is allowed to put a model into production? What are the timeframes for your AI project implementation schedule?

I see too often an executive asking their team to “make a chatbot on our website” or “help our customers get answers on our products with LLMs” without any further direction or expectations. In some cases this is understandable, especially for the first GenAI application at a company, but you really need to go to conferences, hear best practice presentations, and see the potential of AI technologies. This will ultimately raise you with “art of the possible” while grounding you with realism.

AI and Emerging Tech in Sports

Bruce: Looking at your experiences with Mario Andretti and the McLaren team, what emerging AI technologies excite you the most, and how do you foresee them shaping the future of sports analytics?

Kaplan: It’s been an honor to have worked with Andretti, McLaren, Megennis, the autonomous race challenge, and others in the racing industry. AI has been helpful with traditional machine learning use cases such as predictive analytics and classifications. In near the future, real-time streaming analytics will become more prevalent, computer vision with innovate the industry. And technologies like “large action models” where AI will help humans take actions more automatically, such as combining what computer vision sees and making real-time recommendations on racing strategy, driver adjustments such as gear shifting and tail angles. Like in non-sports industries, AI will automate many of the real-world human tasks - like RPA (robotic process automation) but with way more contextual capabilities. This will be like a “helpful SkyNet” to reference Terminator. So exciting!

Measuring AI's Financial Impact

Bruce: In your opinion, what are the best practices for organizations to measure the financial impact of AI initiatives? Are there specific metrics or KPIs that you consider most effective?

The most important metric to me is business impact. Having been an exec at Nielsen, we would measure impact such as lift in sales, specifically tied to the action the business takes (factoring in non-causal activity). For example measuring the increase in sales due to a marketing campaign specifically, and factoring in other factors such as discounts, coupons, interest rate changes, macro economic trends, and so on. Sales lift, cost reductions, time to market, avoiding headcount - these are all measurable KPIs.

Kaplan: Beyond business impact, you can measure cost to build (license of software, storage, employee salary) against business value achieved to calculate the KPI of ROI.