Agenda Machine Learning Week 2025

June 3-4, 2025 l Conference
June 2 and June 5, 2025 l Workshops
Sheraton Phoenix Downtown Hotel, Phoenix, AZ

People won’t use what they don’t trust and understand – and they don’t trust or understand ML and AI. Concerns about bias, hallucinations, compliance and explicability – about trust – are holding many companies back. James will show how adding decisioning to ML and GenAI let’s you design, build, deploy and use trusted ML/AI solutions.

The greatest tool is the hardest to use. Machine learning is the world’s most important general-purpose technology – but it’s notoriously difficult to launch. Outside Big Tech and a handful of other leading companies, machine learning initiatives routinely fail to deploy, never realizing value. What’s missing? A specialized business practice suitable for wide adoption.

In this presentation, Machine Learning Week Founder Eric Siegel presents a six-step practice called bizML for ushering machine learning projects from conception to deployment. This disciplined approach, which is the topic of Siegel’s new book, The AI Playbook, serves both sides: It empowers business professionals and it establishes a sorely needed strategic framework for data professionals.

The media and entertainment world are increasingly looking to AI to cut some of the cost-intensive processes involved in making and producing content. Warner Bros. Discovery is a leading global media and entertainment company, renowned for producing content such as Harry Potter, Game of Thrones, Big Bang Theory and Friends. 

In this case study, Sravanthi will share how Warner Bros. Discovery is improving the cost efficiency of the caption generation process for its Max Streaming Service using an internally built product, “Caption AI.” Caption AI cuts costs relating to generating captions by up to 50%, while creating new file captions also takes up to 80% less time. Caption AI combines custom models for Speech to Text, Speaker Segmentation, and On-Screen placement of Caption Blocks. Caption AI outputs are edited and reviewed by manual caption transcribers to ensure accuracy and context, maintaining high quality in the final captions.

This presentation moves beyond the conventional view of code generation, exploring powerful capabilities for analyzing and enhancing existing codebases. With a fixed output length, how should you scaffold out a new application? How can you leverage large context windows to effectively package an entire codebase in a prompt?

Together with Karl, you will explore a variety of applications from generating architecture diagrams, identifying code quality issues, implementing minor features, and even generating documentation. You’ll learn about useful prompting techniques and integrations with the responses. You’ll leave this session understanding how to apply Generative AI to help you build better software faster, even with an existing codebase.

There’s a fundamental problem with the typical model development process: It evaluates models in terms of technical metrics like AUC/precision/recall without also including business metrics like profit/ROI/savings – the stuff that actually matters to the company.

This is a serious problem – if you aren’t measuring business value, you’re not pursuing business value. Further, those technical metrics fail to provide your client/stakeholder meaningful visibility – she doesn’t care about AUC. How is she supposed to authorize deployment?

That’s why Machine Learning Week founder Eric Siegel recently co-founded Gooder AI. It addresses this fundamental issue by way of its SaaS product, the first full-scale platform for machine learning validation – to maximize the value of models by testing and visualizing their business performance.

Spoiler alert: Unlike technical measures, a model’s business performance (profit, savings, etc.) depends on how you use it. So assessing the business value requires a specialized visualization solution, one that allows you to interactively try out what-if deployment scenarios. This includes setting various business inputs, which are subject to change, and moving the decision threshold to estimate the potential deployed value.

Come watch Eric demo Gooder AI and show how it drives ML deployment to maximize business impact.

Based on Andreas Welsch’s “AI Leadership Handbook,” participants will learn to transform AI hype into actionable business outcomes. Drawing from over 60 interviews with AI leaders and hands-on practitioners, this session provides strategic insights into navigating the complexities of AI implementation. Attendees will gain practical knowledge on fostering a culture of innovation, driving human-AI collaboration, and leading their organizations to new heights. This comprehensive framework covers everything from strategy to leadership, culture, collaboration, and security, providing senior leaders with the tools they need to bring their teams along on the AI journey.

By the end of the session, senior business and technology leaders will be equipped with actionable steps to increase AI literacy across their teams and foster an environment ready for future AI advancements. They will discover the secrets to turning technology hype into measurable business success, ensuring their organizations are well-prepared for the next wave of AI innovation. This program offers a roadmap for senior leaders to effectively lead AI programs, bridging the gap between AI potential and real-world business impact.

Customer interactions are a vital source of information for Paychex. With data coming from diverse sources such as phone calls, emails, chats, and surveys, deriving valuable information becomes challenging. The large volume of data complicates the information extraction process. Paychex has developed a model that combines clustering algorithms to identify the most similar groups of interactions. The model uses GenAI to determine representative topics in each group and employs neural networks for ongoing topic predictions to address this issue. This method allows Paychex to better understand and respond to customer needs, thereby driving success.

In this session, Nathan will show how nonprofits are harnessing the power of AI to serve missions rooted in social benefit, rather than profit. Discover the frameworks and ethical guidelines these organizations are developing to align AI innovations with core values like equity, transparency, and compassion. He’ll discuss the unique challenges nonprofits face in balancing innovation with integrity, sharing examples of how AI can amplify impact without compromising mission-driven goals. This session invites participants to learn how nonprofits set a gold standard in AI ethics, creating models that not only meet but elevate the values underpinning their work for the greater good.

Concepts are sacred, Tools will change, but Data still remains the emperor. Collecting professional, industrial experiences across engineering domains from over 36 years, this talk emphasizes the importance of data quality, adequacy and relevance in finding robust ML outcomes. This talk covers stellar data examples across broad industries such as nuclear, aerospace and power – where the speaker had hands-on experience in developing critical ML models – emphasizing the need for addressing data quality upfront. While highlighting what could go wrong in making crucial go-no-go decisions, the talk in thought-provoking pedagogic style, offers examples to get reliable data, right from the sensors, up to feeding the model.

Hahn is an analytics services firm—but they’ve also developed a unique PR solution called FuseLight that combines predictive AI with genAI: You give it the name of a brand and an influencer and it both predicts the media outcome and generates proposals for a joint PR initiative.

Turning a predictive model into a market-ready product involves more than just accuracy—it’s about delivering actionable insights, intuitive usability, and market traction. In this talk, Michael will show FuseLight’s journey from a predictive analytics model to a product that empowers brands to forecast public fascination with brand partnerships. Along the way, he’ll tackle three key questions: 

• How do you make predictions truly actionable for clients? 
• What’s the trick to turning complex data into a simple, intuitive interface? 
• And how do you acquire market share in a competitive space? 

Join Michael for an inside look at the technical, strategic, and practical steps that transformed a model into a scalable solution driving real-world impact.

Most ML models never deploy. Industry research shows that most predictive analytics projects fail in that respect. What makes the difference? What are the best practices to resolve the organizational issues that so often preclude deployment?

How do Table Discussions work?

You are surrounded by fellow data scientists and there is no better place to discuss and share our common problems – off the record, and in a small group. These are your people – they understand your situation. Often rated the best part of the event, sharing your problems with like-minded professionals is your path to answers, a little empathy, and a stronger professional network.

Come prepared with an issue you are facing, a problem you are solving, or a question that needs answering. And then, be ready and willing to help others to overcome their challenges.

The data science world has largely split in two. Predictive AI and generative AI projects are so different, solving different business problems in very different ways. However, they’re both use case categories of ML and they involve many of the same personnel. And, oftentimes, they combine to become more than the sum of their parts. What are your best practices on this front?

How do Table Discussions work?

You are surrounded by fellow data scientists and there is no better place to discuss and share our common problems – off the record, and in a small group. These are your people – they understand your situation. Often rated the best part of the event, sharing your problems with like-minded professionals is your path to answers, a little empathy, and a stronger professional network.

Come prepared with an issue you are facing, a problem you are solving, or a question that needs answering. And then, be ready and willing to help others to overcome their challenges.

Healthcare data pose unique challenges and opportunities; they can be longitudinal touchpoints for healthy people and potentially indicative of behavior but require careful handling to ensure accuracy and confidence. STChealth leverages machine learning to identify data patterns that reduce human workload and decision fatigue in “deduplication” (removing duplicates and merging matched records), increasing both accuracy and confidence in immunization registry datasets. Additionally, they developed a machine learning classification algorithm that identifies vaccine sentiment on an individual level by maximizing the same immunization data, helping jurisdictions deliver targeted interventions to increase disease protection in vulnerable communities.

The New York Times (NYT) launched its paywall in March 2011, beginning its journey as a subscription-first news and lifestyle service. From its inception, this metered access service was designed so that non-subscribers could read a fixed number of articles every month. With the NYT’s technological transformation into a data-driven digital company, we now successfully develop models that intelligently gateway the most worthwhile pageviews in real time within milliseconds. This decision is made to maximize a combination of business objectives such as subscriptions, user engagement, and search visibility, while balancing the tradeoff between them.

This talk will begin with historical context on the digital transformation of The Times and its subscription funnel strategy with users moving from unregistered to registered and subscribed states. Following an introduction of the data science group, we will draw on differences between predictive and prescriptive machine learning models. We will then discuss how Randomized Control Trials (RCTs) can be designed to train causal machine learning models for prescriptive decision making. Modeling details about our real time models for registered and unregistered users will be discussed, highlighting the delicate balance of multiple objectives using Pareto optimization techniques. We will also present engineering technologies utilized at The Times to deploy these models for high throughput inference.

When businesses base their strategies on the idea that customers can be segmented by Demographic features, they often miss a more powerful factor: behaviors. In this session, Maria will show innovative business practices to better segment audiences and how AI/ML can empower clustering and predictive techniques from a behavioral approach. She will also introduce the frameworks and the principles that will enable business to extract deeper, more valuable insights into consumer personas.

By incorporating these next-generation methods, businesses can evolve their marketing practices, enhance targeting precision, and significantly improve customer acquisition and retention strategies.

This talk is designed to empower business leaders in their measurement systems strategies, offering forward-looking AI/ML techniques that generate growth. Will explore  how Top companies leverage audience segmentation and predictive models to stay competitive in an ever-evolving market.

Dive into the intricacies of building robust Retrieval-Augmented Generation (RAG) systems. This session will explore leveraging Large Language Models (LLMs), OpenSearch, and DeepEval to design and evaluate end-to-end RAG systems. Attendees will gain insights into integrating these technologies to enhance information retrieval, improve response accuracy, and streamline evaluation processes. Join Chris to uncover best practices and innovative approaches in the evolving landscape of AI-driven solutions.

This session offers a practical exploration of AI-driven image generation, focusing on real-world applications, use cases, and the limitations of current technology. Karthik will delve into the fundamentals of generative models, examine the leading tools available today, and discuss future trends in the field. The session will include a hands-on coding demonstration using an open-source model, showcasing practical applications. Additionally, he will address critical topics around incorporating safety measures to prevent misuse, along with ethical considerations for responsible use of this powerful technology.

In today’s competitive consumer markets, determining whether new product sales are incremental or cannibalistic is crucial for informed decision-making. However, traditional machine learning approaches focus on correlation, falling short in offering the causal insights required for real-world business interventions.

This talk introduces a novel approach to sales analytics by integrating causal modeling and Bayesian methods, moving beyond black-box predictions. Evan will showcase a real-world case study in developing advanced causal sales analytics tools. The solution aimed to quantify how new product launches affected both internal and competitor sales, helping the company make data-driven decisions to optimize its product portfolio.

You’ll explore the limitations of standard statistical models and why causal and counterfactual thinking is essential for accurate sales attribution. Attendees will learn how to build models that not only predict outcomes but also answer “what if” scenarios, offering valuable insights into the trade-offs between sales incrementality and cannibalization.

By the end of this session, you’ll have a clearer understanding of how causal models can better guide product strategies, and Evan will even throw in a few award-winning statistics jokes to keep it light! Whether you’re a data scientist, business leader, or machine learning practitioner, this talk will equip you with actionable tools to transform your sales analytics.

This presentation will explore the cutting-edge advancements in the rapidly evolving field of Generative AI. You’ll delve into one or more of the latest developments, which may include:

• Evaluating state-of-the-art Large Language Models
• Investigating challenges and solutions in Retrieval-Augmented Generation (RAG)
• Comparing capabilities of the newest AI-powered image generators
• Analyzing AI-driven search technologies
• Examining emerging applications of Generative AI in various industries

In this session, CIBC Sr. Director Advanced Analytics & AI Chris Rohoman will discuss how the organization is leveraging both open and closed source models to change the bank. As he puts it, “We have pursued many use cases, from knowledge central to financial advisors using our custom model, which increases efficiencies, revenues and build consistency.”

The crises are:

1. Most research results are false.
2. Most discoveries never get implemented

(Perhaps, in light of the first, the second is less bad? 🙂

The shocking extent of these problems may lead to despair.

But hope should triumph as attainable solutions to the crises are explained.

Marketing mix models, channel attribution models, and recommender systems can harmoniously collaborate to gain insights about channel value, ROI, and lift. Join Jayda as she unravels the technicality of marketing models in production, provides recommendations on model selection, and shares proven communication tools for stakeholders in the world of CRM and marketing across industries.

So, you have created your first AI application. Congratulations! But how do you scale, operate, maintain, and grow it within your enterprise? This is the challenge that many organizations face today. Machine learning models cannot live on their own and have to be incorporated into the production environment. 

Generative AI solutions: To make things worse, programming frameworks, tools and infrastructure are evolving at an enormous pace. Luckily, new architectures, tools, and design pattern have arrived to work with these new technologies. One important field of research is MLOps (with speciality LLMOps for large language models), which has evolved into a way of working and set of best practices to deploy, test, manage, and monitor machine learning models in production. 

In this session, Bas will present a reference architecture that can be applied to any machine learning or generative AI project, with some examples from the real world and practical tool advice. Based on this information you should be able to create your own solution architecture and reason about the technology options. Join this session to learn how to build a robust and scalable AI platform that meets the needs of your enterprise!

Generative AI offers powerful applications for clinical document summarization in utilization management (UM). At its core, these systems leverage Natural Language Processing (NLP) to analyze and interpret complex medical texts. Advanced machine learning models, such as transformer-based architectures, can be trained on vast corpora of medical literature and clinical notes to understand context, medical terminology, and relationships between different pieces of information.

These AI systems can then automatically extract key clinical data points, including diagnoses, medications, treatment plans, and relevant patient history. Beyond mere extraction, generative AI can synthesize this information into concise, well-structured summaries tailored to UM requirements. For instance, the AI might generate a summary highlighting specific elements that impact care decisions, such as severity indicators, failed previous treatments, or contraindications.

These summaries can be formatted to align with standardized UM review criteria, making it easier for reviewers to quickly assess care appropriateness. Additionally, some advanced systems can generate targeted follow-up questions or flag potential areas of concern, further streamlining the UM process.

By automating these complex cognitive tasks, generative AI not only saves time but also helps ensure consistency and comprehensiveness in clinical document analysis for utilization management.

Priyanka will show you how to build an end-to-end customer churn prediction pipeline using MLFlow. She’ll cover every step of the process—from data preprocessing and feature engineering to tracking experiments, building ML pipelines, and training high-performing classification models. The entire workflow will be managed within MLFlow, allowing us to build, track, and deploy pipelines seamlessly. To make predictions accessible, a user-friendly interface using Streamlit for real-time visualization of churn predictions will be developed. This session offers a practical and approachable way to implement customer churn prediction for both beginners and experienced data practitioners.

Thought leaders in machine learning Dean and Karl, field questions from the audience about strategies for machine learning projects, best practices, and tips, drawing from their decades of experience as consultants and company executives.

In India, CPG brands like P&G and Unilever rely on sales forces to gather inventory orders and expand retail networks. Apart from taking the orders, one of the KPI for these salesman is to add new retail shops (Mon-n-Pop stores) to increase business. However, some sales personnel create fictitious retail shops in databases, fueled by KPI pressures, submitting random images to fulfill mandatory photograph requirements. Currently, no effective automated solution can verify store authenticity or detect duplicates.

Collaborating with a leading CPG brand, Rohit and his team processed one million images to enhance their data quality and remove duplicates. Leveraging GenAI multimodal algorithms, they distinguished store images and identified duplicates using embeddings. This presentation will showcase how GenAI Visual Question Answering addresses this real-world challenge