Consider the ordinary indignity of ordering inventory. Too much, and cash sits on a shelf wearing last season’s colors. Too little, and the customer leaves. A demand forecast helps, but it does not choose between those disappointments. Someone still has to decide. Ikigai built its business around that gap: the distance between a prediction in a table and a decision in the world.
- AI for business tables: reconcile data, forecast demand, compare plans.
- A spreadsheet-like interface keeps operational experts involved.
- The 2026 Celonis deal puts those capabilities inside a larger operations platform.
The company’s territory is deliberately practical. Its applications cover supply-chain planning, financial reconciliation, workforce requirements, and related jobs where numbers arrive with dates attached. A planner wants to know what to buy. A finance team wants records to agree. An operations manager wants to compare alternatives before committing money. These are useful ambitions, even if they are unlikely to win a poetry prize.
A table has relationships, too
Ikigai began in 2019 with Devavrat Shah, an MIT professor, and Vinayak Ramesh, his former student. Their backgrounds joined mathematical research with company building. Shah had co-founded retail analytics company Celect, acquired by Nike. Ramesh had co-founded digital-health business Wellframe. Both had encountered organizations whose decisions depended on imperfect information.
Their technical choice was Large Graphical Models, or LGMs. Think of a table as a set of relationships: one product substitutes for another; one location behaves differently from its neighbor; sales change over time. Modeling those relationships can help recover missing information and estimate what comes next. Ikigai built tools around structured and time-series data, the material in spreadsheets, databases, and transaction systems.
That focus matters when judging the product. A language model’s fluency tells you little about whether an inventory forecast is useful. Ikigai’s proposition concerns the underlying numbers and their dependencies. The fair test is operational: does a forecast improve a decision when evaluated on the buyer’s own data?
Three verbs, one decision
Its core tools form a sensible sequence. aiMatch reconciles data from different sources and helps address gaps and inconsistent records. aiCast produces forecasts. aiPlan explores possible decisions and their consequences. The point of putting them together is continuity: the information used to predict demand can carry into planning without another round of spreadsheet exports.
↳ Expert review runs through the workflow ↲
A forecasting demonstration shows users comparing models and inspecting results at different levels, including products and stores. A planning demonstration then tackles the familiar inventory tradeoff. Overordering ties up cash and creates carrying costs. Underordering risks lost sales. The forecast estimates demand; the plan asks what action makes sense given those costs.
Ikigai describes this as expert-in-the-loop. Someone who knows the operation can examine exceptions and contribute judgment. That is an important design choice. A database may record an unusual sale; a planner may know why it happened. The interface aims to make mathematical tools usable by people who have that context without requiring them to write the machinery themselves.
A sales-and-operations walkthrough makes that collaboration tangible. Demand planners inspect forecasts, inventory teams examine stock and capacity, and executives review financial implications and stockout risks. The template includes checks that show which stages have been reviewed. An elegant algorithm can be stranded by an unfinished handoff; here, the mundane business of knowing who has checked what receives its own space. For a team working across departments, that detail may matter as much as the curve on the forecasting chart.
“Supply chain demand forecasting is where we started.”
Vinayak Ramesh, speaking to MIT Startup Exchange
Public accounts place customers across retail, manufacturing, life sciences, and financial services. OpenCrowd, a partner with financial-services expertise, reported building a diamond-index forecasting application on the platform in four weeks. That is a concrete example of application development speed; it does not, by itself, establish a return on investment.
An algorithm still needs an invitation
Ikigai sold enterprise software as a service. Its Azure offering also provided a marketplace purchasing route and connections to Microsoft’s data tools. It occupied ground between planning systems such as Anaplan, broader enterprise AI platforms such as Dataiku, and custom software assembled by internal teams. Buyers were choosing a way to prepare data and operate models as much as a forecasting technique.
The company announced a $13 million seed round in December 2021 and a $25 million Series A in August 2023. The latter was led by Premji Invest, with Foundation Capital and e& capital participating. Financing bought room to develop the business. It should not be confused with what a customer pays to use the software.
Two disclosed rounds. Bar lengths are proportional to round size.
Education was another part of its approach. Ikigai Academy offered instruction in AI and forecasting, including practical exercises. The company reported more than 6,000 learners across over 90 countries by August 2023. Those learners were an audience for education, a different measure from enterprise customers.
The commercial difficulty emerges in investor Foundation Capital’s account of the Celonis acquisition. Enterprise adoption requires industry knowledge, integration, and trust. An impressive demonstration does not automatically become an installed system. The investor argues that Celonis supplies established relationships and an operational data platform that can accelerate deployment.
On May 12, 2026, Celonis announced an agreement to acquire Ikigai. Foundation Capital described it as acquired the following day. The combination brings Ikigai’s forecasting and simulation into the Celonis Context Model, which represents how a business operates. Shah announced he would join as Chief Scientist, Enterprise AI; Ramesh as Field CTO. The founders’ stated logic was scale and a shared view of reliable decisions.
Start with the decision you can afford to get wrong
There is a useful method here for anyone evaluating enterprise AI. Pick a specific recurring decision. Identify its inputs. Compare predictions against held-out historical periods. Then measure the cost of the resulting action, including excess stock or missed demand. Give an accountable expert a place to review exceptions. This is our reading of Ikigai’s approach, and a practical way to assess it.
The conditions matter. Sparse data still needs relevant information. A planning model needs realistic costs and constraints. Relationships learned from past behavior deserve scrutiny when conditions change. A team unable to connect its systems or act on results may gain little from a clever forecast. Ikigai’s appeal rests on closing that final distance: helping the person who must place the order make a better call.