Helping AI models to meet the real world

2026-07-14 · Source: MIT News - Artificial intelligence · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics, Operations & Process Management · Depth: Intermediate, short

Summary

Professor Devavrat Shah, a principal investigator at MIT's Laboratory for Information and Decision Systems, has developed methods for real-time decision-making using limited computational resources. His research led to a patented foundation model for tabular, time series data, which became the core technology for Ikigai Labs, a company he co-founded in 2019. Unlike most AI models trained on text and images, this system processes structured tabular data, like spreadsheets, to provide large-scale, real-time planning. It continuously learns by validating predictions against actual outcomes, enabling forecasting and decision-making for large businesses, such as consumer goods and pharmaceutical companies, by optimizing interdependent processes like supply chain, pricing, and promotions. Ikigai Labs was recently acquired by Celonis, where Shah now serves as chief scientist, aiming to integrate this cost-effective AI model into Celonis's platform to build "enterprise process world models" for over 1,400 companies.

Key takeaway

For Directors of AI/ML or Business Operations evaluating solutions for enterprise planning, you should prioritize AI models specifically designed for structured, tabular, and time series data. This specialized approach, exemplified by Ikigai's foundation model, offers a cost-effective and precise method for real-time forecasting and decision-making across complex business processes. Integrating such focused technology can significantly enhance your organization's ability to optimize operations, predict outcomes, and simulate strategies at scale.

Key insights

A foundation model for tabular, time series data enables real-time, large-scale business forecasting and decision-making.

Principles

• AI for structured data offers cost-effectiveness.
• Continuous learning improves prediction accuracy.
• Narrower AI focus yields sharper technology.

Method

The system takes tabular data, learns continuously by testing predictions against real outcomes, and provides real-time planning and optimization for enterprise processes.

In practice

• Optimize supply chain and pricing strategies.
• Forecast demand for new product versions.
• Simulate business process options.

Topics

• Foundation Models
• Tabular Data
• Time Series Data
• Enterprise AI
• Business Process Optimization
• Real-time Decision Making

Best for: CTO, VP of Engineering/Data, Executive, AI Scientist, Data Scientist, Director of AI/ML

Related on AIssential

• Relational Foundation Models for Enterprise Data with Jure Leskovec - #768 — Relational Foundation Models enable accurate, no-training predictions on multi-table enterprise data by learning directly from graph-structured relationships.
• The context problem: Why enterprise AI needs more than foundation models — Enterprise AI requires deep institutional context to move beyond demos and deliver real production value.
• The brewing GenAI data science revolution — Foundation models are extending beyond text/image to transform tabular and time-series data analysis.
• Interview with AAAI Fellow Yan Liu: machine learning for time series — Physics-informed foundation models enable accurate, adaptable time series analysis, especially in data-scarce scientific and real-world applications.
• SAP agrees to acquire German AI startup Prior Labs — Tabular foundation models are poised to revolutionize structured business data analysis, mirroring LLMs' impact on unstructured data.
• [Webinar] TabH2O #2 Webinar Recording — Tabular Foundation Models simplify predictive analytics for structured business data by learning patterns across diverse tables.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by MIT News - Artificial intelligence.