A Startup Just Raised $77 Million to Build AI Helpers for Office Jobs
Ema, a startup that builds AI helpers for office work, has raised $77 million. The Series B round was reported September 23, 2026 and was led by Creaegis, a venture firm based in Bengaluru. Existing investors Accel, Section 32 and Prosus put in more money. Total funding is now $140 million. TechCrunch
The round was all primary equity, meaning new money into the company. There was no debt and no sale of existing shares. Valuation more than quadrupled since the last round in 2024. Ema was started in 2023 by Surojit Chatterjee, who previously worked at Google and Coinbase, and Souvik Sen, who previously worked at Okta.
Ema reports more than 50 business deals, over 1 million active users at those businesses, and more than 5 million tasks and questions handled so far. Named customers include NTT DATA, Hitachi, ADP, PwC, Google, KPMG, Wipro and Microsoft. Revenue grew 50-fold in the past two years. Bookings have passed $150 million, which the company defines as the total value of multiyear contracts including two- and three-year deals. TechCrunch
The broader context here is that bookings are promised contract value, not yearly revenue already earned. Long contracts are normal when setup needs integration work, management review and slow rollout across departments.
Ema is not tied to one AI model. It can use more than 150 models, including top commercial systems and open models. The product is sold as AI Employees, digital helpers for recruiting, onboarding, benefits and performance. They come with 250+ ready-made connections and 1,000+ actions. People reach them through Microsoft Teams, Slack and voice. TechCrunch
Its routing system is called EmaFusion. Think of it as a dispatcher that picks the best model for each small job. According to the company, it mixes 100+ models in real time and can cost up to 20 times less than using one best model alone. On September 1, Ema launched its HR, IT and Finance Hub. Ema also took first place for Best Services-as-Software Transformation at the HFS Services-as-Software Awards 2026.
Looking at the growth pattern, two points stand out. A primary-only round with current investors adding money keeps cash inside the company for hiring and for running and supporting the AI, rather than paying out early holders. Fifty deals with one million users suggests a few very large customers rather than many small ones, through central contracts followed by adding more seats.
In my view, the number of models matters less than the basic idea. Depending on one AI model leaves a business open to price rises, drops in quality and rule changes. Picking from many models trades simplicity for choice, balancing cost and speed with accuracy. Whether the 20-times saving holds in daily use depends on task mix, reuse of past answers and careful testing, which is hard to check from outside.
The practical change for buyers here is what is being sold. Bookings over $150 million with ready links and job-specific helpers point to paying for results and task coverage, not licenses alone. That shifts checks from feature lists to handling errors, keeping audit trails, controlling identity and undoing an AI action across record systems. Ema's Okta and large-platform background is relevant here, though execution will matter more than history.
Looking ahead, what this could enable is testing AI helpers in office work that has resisted simple software. Recruiting, onboarding and benefits work involves many steps across many systems. These helpers can read policy, pull facts from different systems and draft the next step, as long as people keep final approval where it counts.
