March 2026 Summaries

In 2026, business analysts are focusing on building a comprehensive stack of AI tools rather than relying on a single solution, as these tools cater to recurring reporting, ad-hoc questions, auditable analysis, and documentation, which many teams often overlook. The guide identifies top AI tools for business analysis, highlighting their features, use cases, and technical requirements, and emphasizes the importance of selecting tools that integrate seamlessly into existing workflows while being robust enough to handle complex data scenarios. Key tools like ChatGPT, Julius AI, Zerve, Databricks, Power BI, Tableau, Looker, Notion AI, and Google Sheets are categorized based on their strengths in areas such as natural language processing, reproducibility, scalability, and documentation. Each tool is suited for specific roles, from business analysts needing tools like Zerve and Tableau for detailed analysis, to founders who require quick, actionable insights. Ultimately, the right combination of tools addresses specific workflow challenges such as reproducibility and auditability, often requiring a multi-tool approach to effectively manage business analysis tasks.

Predictive analytics leverages historical data, statistical models, and machine learning algorithms to forecast future events, enabling organizations to make informed decisions and reduce reliance on guesswork. This approach has become more accessible thanks to open-source tools and cloud computing, allowing businesses to predict outcomes like customer churn, product demand, and equipment failures across various industries such as finance, healthcare, retail, supply chain, marketing, and energy. Key techniques include regression models, decision trees, time-series forecasting, survival analysis, and deep learning, each suited to specific types of data and problems. Despite challenges like data quality, overfitting, and organizational adoption, modern platforms streamline predictive analytics by integrating data preparation, model training, deployment, and monitoring. Zerve AI exemplifies these advancements by offering an Agentic Data Workspace that enhances collaboration, reproducibility, and efficiency in predictive workflows. As companies increasingly rely on predictive analytics, it becomes crucial to maintain high-quality data, robust validation, and continuous monitoring to ensure reliability and accuracy in decision-making processes.

The evolution of computational notebooks reflects the broader trajectory of data science, transitioning from academic exploration tools like Maple and Mathematica to enterprise-scale applications. Early tools such as IPython and its successor Jupyter, launched in 2014, integrated code, text, and visuals, making them popular in academia despite limitations in reproducibility and deployment for business use. As data science entered the corporate world, contributions like SciPy and Matplotlib expanded Jupyter's applicability, but challenges in transitioning from exploratory analysis to production remained, often requiring handoffs to engineers for further development. Modern platforms like Google Colab introduced collaborative features but retained Jupyter's single-kernel limitations, highlighting the need for solutions that bridge the gap between notebooks and integrated development environments (IDEs). In response, tools like Zerve have emerged, offering language interoperability, parallelization, and cloud optimization to meet demands for speed, reproducibility, and scalability. Looking forward, the future of data science environments will likely involve platforms that merge the flexibility of notebooks with the robust features of IDEs, facilitating faster transitions from idea to production while supporting AI advancements.

Zerve is a new, code-first Data Science Development Environment designed to enhance the productivity and impact of data scientists by addressing inefficiencies in traditional data science workflows. Built to enable exploration, stability, and collaboration, Zerve separates storage from compute and organizes code as a directed acyclic graph (DAG) to ensure stable and consistent execution without the need for restarting kernels. The platform supports real-time collaboration, language interoperability between Python, R, and SQL, and seamless multi-processing through serverless technology. By serializing outputs and making artifacts persistent, Zerve allows for efficient multi-language usage and simplifies the transition from development to production. With its novel architecture, Zerve aims to overcome the limitations of low-code/no-code solutions and traditional notebooks like Jupyter, offering a robust environment for serious data science and AI projects while being freely accessible with generous resources on its free plan.

Zerve is a data collaboration platform that enhances teamwork by allowing multiple data scientists to work simultaneously within the same project without interfering with each other's contributions. Unlike typical data platforms that offer basic sharing capabilities, Zerve supports real-time co-editing, comprehensive version control, and integrated review workflows. This design addresses common issues like merge conflicts and inefficient review processes by isolating code blocks with their own compute resources, enabling seamless collaboration akin to Figma or Google Docs. Zerve's platform automatically logs version history, keeping track of code execution and results, while also facilitating in-platform code reviews with live data access, eliminating the need for external tools like Slack. Furthermore, Zerve simplifies onboarding by providing new users immediate access to a consistent environment, complete with pre-configured dependencies and connections, and introduces an AI agent to assist with tasks such as data cleaning and code analysis, enhancing productivity without requiring human intervention.

Greg recaps his experience at the ODSC 2026 where he demonstrated agentic coding by building a cloud cost optimization simulator live, using a single prompt without any pre-written code. He describes the complexities of managing cloud infrastructure costs and how his simulator effectively balanced the use of reserved, on-demand, and spot instances to optimize costs while maintaining reliability. Greg highlights the advantages of the Zerve agent, which operates within the development environment, allowing it to adapt dynamically to data changes and code errors, thereby enhancing the efficiency of data science workflows. The simulator's adaptability was tested by increasing workload volatility, revealing that the optimal resource allocation strategy remained largely unchanged. He also discusses the architecture of Zerve, which separates code execution into individual kernels to prevent state sharing issues common in Jupyter notebooks. Greg emphasizes the effectiveness of short iterative prompts over lengthy ones and mentions the potential for Zerve to be used beyond prototyping, with capabilities for deploying models as APIs and building applications. He invites users to explore Zerve with a free tier offering and encourages feedback from the community.

FeatureByte, co-founded by Razi Raziuddin and Xavier Conort, addresses the challenges of feature engineering, which is often a manual and time-consuming part of the data science workflow. By automating the process of analyzing metadata, generating and evaluating features, and ensuring point-in-time correctness, FeatureByte significantly reduces the time needed for model development from months to days while achieving model performance improvements. Despite the hype around feature stores, they are beneficial mainly for real-time use cases like fraud detection, with simpler solutions sufficing for most other workflows. Automation is becoming crucial as data teams face increasing pressure to deliver more models in production, and those not incorporating automation risk falling behind. Razi emphasizes that tabular data presents unique challenges distinct from language models, as its context is highly dependent on specific variables, requiring a different approach for effective analysis and prediction.