Ikigai Labs brings the power of an emerging form of generative AI to crucial enterprise data.
Generative AI tools, particularly large language models (LLMs) such as ChatGPT4, are starting to find many uses in corporations. But LLMs aren't designed to work with the key operational data, such as sales figures, that businesses run on.
Ikigai Labs, an MIT STEX25 startup, takes on this challenge with large graphical models (LGMs), an emerging major form of generative AI.
Most enterprise data are in the form of structured tabular data, such as financial transactions that are tracked over time. LGMs are extremely good at handling exactly this kind of data, says Vinayak Ramesh, co-founder and chief customer officer.
The Ikigai Labs cloud service offers end-to-end LGM applications tailored for business analysts working with corporate tabular data, particularly time-series data. Designed for speed and ease of use, the software can deliver strikingly better analyses. “One major retailer improved its product demand forecast by close to 40 percent,” says Ramesh. “We more than tripled the claims auditing capacity of one major insurance company.”
Making LGMs practical Ikigai Labs is Ramesh's second AI startup. Immediately after graduating from MIT with a degree in computer science, he co-founded Wellframe, a digital health company employing AI to help health insurance companies better manage their patients.
After successfully growing and selling Wellframe, Ramesh returned to MIT in 2016. “I wanted to get back to working on technical problems,” he says. “What better place than MIT to explore with smart people and work on something interesting and new?'
Studying with Devavrat Shah, the Andrew (1956) and Erna Viterbi Professor of Electrical Engineering and Computer Science, Ramesh joined in the quest for more capable LGMs.
Graphical models, a class of probabilistic algorithms, have been around for more than a century and used in physics, communications and many other applications, Ramesh notes. However, making LGMs, graphical models using machine-learning techniques, accessible for enterprises was no easy task.
“Large graphical models are very powerful; they're like neural networks on steroids,” Ramesh says. “But it was very hard to use them and scale them in practice. Also, in most cases, you needed a domain expert to come in and tune the model for that particular use case.” To avoid that problem, he and Shah sought a way to make LGMs work in any enterprise.
After achieving and patenting major advances in LGMs, they began thinking about ways to commercialize their technology. Ramesh remembered his experience at Wellframe, where the nursing care managers who were now overseeing much larger numbers of patients needed to strengthen their data-handling skills. That scenario offered a dramatic example of the general need to close the learning gap between business users and AI tools.
In 2019, Ramesh and Shah launched Ikigai Labs, named for a Japanese word for “purpose”. (It was also the second AI software firm for Shah, who in 2013 founded Celect, which helps retailers optimize inventory and was acquired by Nike in 2019.)
Keeping tabs on enterprises Ikigai Labs debuted its enterprise LGM platform in 2023. The software brings the raw power of generative AI to analyses of tabular data, especially times-series data, for tasks such as demand forecasting and labor forecasting. Unlike many other AI products, the platform is an end-to-end suite that allows users to do everything from connecting the data and building the AI models all the way to integrating the models into the apps that the business runs every day.
The Ikigai software includes three classes of models: aiMatch, which allows users to stitch together multiple disparate data sets; aiCast, a time-series forecasting tool; and aiPlan, which provides outcome-based scenario analysis. (Usually in scenario analysis, you tweak the input and see how the output changes, Ramesh explains. In outcome-based scenarios analysis, you simulate different outputs and see what inputs will get you there.)
Crucially, all these models come with human “expert in the loop” abilities. “With AI, exceptions are the norm, not the exception,” Ramesh says. “So you always have to integrate a human in the middle, to effectively give a thumbs up or thumbs down on what the AI is saying, and then have the AI learn from that.”
Some customers will choose to run their models in completely automated fashion, but analysts can always make necessary contributions and corrections. For instance, if a retailer decides not to run its usual Labor Day sale this year, an analyst can add that information to the model.
In another example, an AI model performing financial auditing may not be able to verify certain activities. “We can flag those activities for a reviewer to go in and manually verify,” Ramesh says. “Only the highest risk or highest priority things must be manually verified.”
Ikigai software works off the shelf but trains itself with each customer's data. “If you're a health insurer, it will learn the model just on your data,” he says. “Because it's domain-agnostic, you don't need a bunch of data scientists tweaking the model.”
The company strongly emphasizes user education through “Ikigai Academy” courses that are overseen by Paul Marca, a former Stanford University associate vice provost with expertise in online education. “Our mission is to bring AI to everyone, but also to educate everyone,” Ramesh says. “We answer questions like, What is AI? What are its foundations? Where can it be used? And then how do you actually use the Ikigai platform?”
Ikigai has found customers in retail, manufacturing, life sciences and financial services. “Supply chain demand forecasting is where we started,” says Ramesh. “But we're quickly gaining traction in workforce planning, like call center staffing or warehouse staffing.” Within banking and insurance firms, the LGM models often help to investigate fraud or audit cases. Customers also are finding new roles for the software that hadn't occurred to Ramesh and his colleagues, “which is exciting,” he says.
Applying LGM apps to company operating data brings several major benefits. “For example, everyone does demand forecasting, but they can do it much better with Ikigai,” says Ramesh. The software not only can provide better forecasts with existing tabular data, it can easily work with external sources such as weather databases. In addition, the LGM models can provide insights that human analysts might miss, such as spotting two stores that are cannibalizing each other's sales.
Based primarily in the San Francisco Bay area and Cambridge's Kendall Square, Ikigai Labs raised $13 million in seed money and $25 million in a follow-up round in August 2023. The company employs 60-plus people and is growing rapidly.
“There's a lot of demand for AI right now; enterprise buyers are taking it as a very serious imperative,” says Ramesh. “Everyone's looking at it, and a lot of these business units have budgets to implement something.”