Bridging AI Models and the Real World
Ikigai Labs, an MIT spinout, was acquired by Celonis to embed tabular and time-series AI forecasting in enterprise operations.
Key Takeaways
- Celonis has acquired MIT spinout Ikigai Labs to integrate AI forecasting and decision-making for tabular and time-series enterprise data into its business-operations platform.
- Ikigai’s model differs from text- and image-based AI by learning directly from structured business records such as sales, inventory, shipments, pricing, and sensor data.
- The technology extends graphical-model research to support scalable, real-time predictions and planning for complex, interdependent operational decisions.
- The acquisition signals a shift in enterprise AI toward systems built on organizations’ own digitized process data, making forecasts and decisions more actionable and cost-effective.
- With Celonis’s base of more than 1,400 large companies, Ikigai’s tools can be layered onto existing operations to simulate options, predict outcomes, and optimize strategies at scale.
Meta description: MIT spinout Ikigai Labs, founded in 2019, was acquired by Celonis to bring tabular-data AI forecasting and decision-making to enterprise operations.
Tags: enterprise AI, tabular data, time series forecasting, Celonis, Ikigai Labs
Featured image: abstract illustration of a digital factory floor with flowing spreadsheet rows, timelines, and forecast curves; friendly, modern, no text.
Celonis, an international firm that specializes in digitizing and automating business operations, has recently acquired Ikigai Labs, a spinout founded in 2019 by Devavrat Shah, a principal investigator at MIT’s Laboratory for Information and Decision Systems (LIDS), faculty member in the department of Electrical Engineering and Computer Science (EECS), and member of the Institute for Data, Systems, and Society (IDSS). The move places tabular-data AI decision-making at the center of Celonis’s effort to help large organizations forecast, plan, and act on their own process data. For enterprise data teams, operations leaders, and vendors of AI tools, the news signals a shift from generic text-based models toward systems built around the structured, time-based records that already sit inside business software.
Why business AI often misses the point
Business AI tools for forecasting, planning, and decision-making have spread widely, but many omit the organization-specific detail that makes advice useful. A model may sound confident, yet it may not know a company’s actual product mix, supplier network, service obligations, or pricing history. Shah has worked on a related constraint: how to design methods that can handle second-by-second decision-making using limited computational resources. He frames the problem as doing a lot of heavy lifting with a small amount of resource. His research interest is extracting information from data at scale as effectively as possible.
Shah is the Andrew (1956) and Erna Viterbi Professor and has taught at MIT since 2005. In 2019, he co-founded Ikigai Labs. The company built a foundation model for tabular and time-series data, based on years of research in Shah’s lab. The model was patented and licensed by MIT to the company. It can take input from enterprise data from varied sources, continuously and at scale, and learn as it goes by testing its predictions against real outcomes.
What makes the Ikigai model different
Most large AI models are trained using text and images. Ikigai’s model takes tabular data as its input: structured data in the row-and-column format familiar from spreadsheets. It also handles time-series data, where values are recorded over time, such as daily sales, inventory levels, shipment delays, or sensor readings. A foundation model can be thought of as a broad AI base that can support many related tasks rather than one narrow application.
Shah explains that the system is an extension of graphical models. Graphical models are statistical structures that represent relationships among variables. They are used, for example, by GPS devices to convert sparse satellite data into an accurate model of a position on Earth’s surface. They are also used by communication systems, such as those in a digital watch, that send data at high speed in an energy-efficient manner. Shah’s question was how to design such graphical models for generic tabular data. The result provides forecasting and real-time planning on a vastly larger scale.
How it could change everyday operations
The idea for Ikigai was to provide forecasting and decision-making technology for large businesses, such as consumer goods manufacturers and pharmaceutical companies. Shah gives the example of a consumer electronics company that makes headphones and various other products. Each product contains many small pieces sourced from different parts of the world. Once the device is sold, it needs to be supported and maintained. The company must develop new versions, market them, and price them.
The questions become operational and time-dependent: If I sell these products next quarter or next year, how many will be sold in different places? What would happen to demand if I change the price? What would happen if I introduce a promotion? Shah adds that all of these processes are interdependent. At every stage, decisions have implications over time. His view is that digitizing these processes, making predictions, and constantly optimizing them leads ultimately to better business operations.
What Celonis adds to the picture
Ikigai was recently acquired by Celonis. Shah is now chief scientist at Celonis in addition to his roles at MIT. He hopes the model developed for Ikigai will help Celonis deliver tools that integrate with a company’s own data and business processes to provide real-world analyses that help make forecasts, plans, and decisions.
Celonis has built a base of more than 1,400 large companies worldwide by digitizing and automating their operations. Those systems are now fully digitized, Shah says, creating a platform for Ikigai’s software to take the next step. It can read data from digitized systems and provide detailed models that allow simulation of different options, predict optimum strategies, and forecast the results of a given set of decisions. In his framing, once the digital layer of these processes exists and the information layer exists, the Ikigai stack can be put on top of it to enable decision-making at a much larger scale than otherwise.
Why this matters and what to watch
The acquisition matters because it points to a practical gap in enterprise AI. Text and image models are powerful, but many operational decisions depend on structured, time-domain data that already exists inside companies. Shah says the work focuses on a part of the domain that the rest of the world is not paying attention to. By starting from tabular and time-series data, he argues, it provides a very cost-effective version of AI. A narrower focus can come with sharper technology, he says, while still being broad enough to be valuable.
He also connects the approach to the term “world model,” a phrase that has become common in AI discussions. In this setting, the goal is to build an enterprise process world model: a computational representation of how a company’s products, customers, supply chains, prices, promotions, and support processes behave. That representation can let teams test decisions before they affect real operations.
For readers, the takeaway is that the next useful wave of business AI may not be another general chatbot. It may be a system that reads a company’s own tables and timelines, learns from outcomes, and helps operations leaders answer questions about demand, pricing, inventory, and maintenance. What to watch next is how Celonis turns the acquired technology into tools for its installed base of more than 1,400 large companies.
Comments (0)