Pecan AI, an eight-year-old startup, has been a prominent player in the predictive analytics domain for enterprises. Having raised a significant $116 million in funding since its inception, including a notable $66 million round in February 2022, the company has made a name for itself in the industry. The startup’s primary offering has been a predictive analytics platform, tailored for enterprise use, leveraging data to forecast future events and trends.
Recently, Pecan AI has embarked on an ambitious new venture. The company is introducing Predictive GenAI, a groundbreaking tool that synergizes the predictive capabilities of traditional machine learning with the creative prowess of modern generative AI. This innovative blend aims to revolutionize the field by augmenting predictive analytics with the versatility and adaptability of generative models.
The CEO and co-founder of Pecan AI, Zohar Bronfman, shared insights with VentureBeat about this new development. “While we were diligently working on refining predictive analytics solutions using classic machine learning techniques, the gen AI revolution was burgeoning in parallel. Unfortunately, gen AI, for all its advancements, falls short in generating reliable predictions,” he explained.
Gen AI, though remarkable in various applications like chatbots, content summarization, and report generation, isn’t ideally suited for prediction tasks. Bronfman pointed out that the datasets used for training gen AI models aren’t typically formatted for predictive modeling, which demands a specific structure. Predictive models require datasets where each row represents a unique entity, each column signifies a distinct feature, and a separate label column for the target variable. Achieving this format often necessitates substantial data engineering, a task gen AI models struggle with.
Pecan AI’s Predictive GenAI aims to address these limitations by combining the strengths of both technologies. Predictive GenAI is not just about integrating two technologies but also about enhancing user accessibility and experience. Traditionally, predictive machine learning technologies were primarily the domain of data scientists. Pecan AI, however, strives to democratize AI, making it accessible to individuals with business-centric roles within organizations. This approach aligns with the company’s goal of simplifying the adoption of machine learning and AI technologies in the most user-friendly manner possible.
Predictive GenAI comprises two main components. The first is Predictive Chat, a feature that allows users to interact through a chatbot-like interface using natural language queries. This tool is designed to guide users, especially those dealing with specific business challenges, towards employing the appropriate predictive framework that aligns with their business needs.
The second component, the Predictive Notebook, is where generative AI truly shines. It automates the creation of a data science notebook, which serves as the foundation for building predictive models. Bronfman elaborated that this proprietary SQL-based notebook contains auto-generated cells. These cells play a crucial role in transforming a company’s raw data into an AI-ready dataset, suitable for predictive modeling. The process involves various stages like querying, structuring, and joining data. Pecan AI’s backend executes these cells transparently, although users can delve deeper and modify the cells using SQL if desired. Ultimately, the notebook generates queries that transform the native data into a format ready for use in Pecan AI’s predictive modeling library.
Beyond the user interface, Pecan AI is also pioneering in the domain of data preparation and feature engineering. The company is innovating automated solutions to tackle issues like data leakage, which can significantly impair model accuracy. Data leakage refers to the inadvertent use of information during training that wouldn’t normally be available during prediction. Identifying and addressing data leakage is a complex task, often requiring the expertise of professional data scientists. Pecan AI is developing automated methods to detect and mitigate this issue, enhancing the overall reliability and accuracy of predictive models.
Despite the potential of vector databases and embeddings in supporting basic predictive functions, they fall short for comprehensive predictive AI modeling, as Bronfman pointed out. These technologies, while useful, can only handle a limited set of features. Either the models would be overly simplistic, capturing only basic patterns, or they would still require complex feature engineering by a data scientist to prepare the data for more sophisticated predictive models.
In conclusion, Pecan AI’s Predictive GenAI represents a significant step forward in the field of artificial intelligence. By merging the intuitive and adaptable nature of generative AI with the foresight and accuracy of predictive machine learning, Pecan AI is not only enhancing the capabilities of AI in business applications but also making these advanced technologies more accessible to a broader range of users. This development could pave the way for more innovative uses of AI in various industries, making predictive analytics a more integral and user-friendly part of business decision-making processes. As Pecan AI continues to develop and refine its Predictive GenAI tool, it stands at the forefront of a new era in AI, where the power of prediction is coupled with the creative potential of generative models, leading to more sophisticated, efficient, and accessible AI solutions for businesses worldwide.