Ema Raises $77M to Scale Enterprise AI Employees Across HR, IT and Finance
Ema has raised $77 million in a Series B round reported September 23, 2026. The round was led by Bengaluru-based venture firm Creaegis, with existing investors Accel, Section 32 and Prosus increasing their stakes. Total funding now stands at $140 million. TechCrunch
The round was all primary equity. There was no debt and no secondary transactions. Valuation more than quadrupled from the company's last round in 2024.
Ema was founded in 2023 by Surojit Chatterjee and Souvik Sen. Chatterjee is a former Google and Coinbase executive. Sen is a former Okta executive.
On traction, the company reports more than 50 active enterprise deals, over 1 million active enterprise users, and more than 5 million actions and queries handled to date. Named customers include NTT DATA, Hitachi, ADP, PwC, Google, KPMG, Wipro and Microsoft. Revenue grew 50-fold over the past two years. Revenue bookings have surpassed $150 million, a figure the company defines as total value of multiyear contracts including two- and three-year deals. TechCrunch
That bookings distinction matters for enterprise readers. It is contracted value, not recognized annual recurring revenue, and multiyear structures are common when deployment involves integration work, governance review and phased rollout across business units.
Architecturally, Ema is model-agnostic. Its software can draw on more than 150 models, including frontier and open-source models. The product is packaged as AI Employees for functions including recruiting, onboarding, benefits and performance, delivered through 250+ prebuilt integrations and 1,000+ actions, with access via Microsoft Teams, Slack and voice interfaces. TechCrunch
The routing layer is called EmaFusion. According to company materials, it combines 100+ models in real time for each task, with Ema stating it delivers up to 20x lower cost than the best single model. On September 1, the company announced the launch of its HR, IT and Finance Hub, extending that agent framework into shared corporate functions.
Ema also won first place for Best Services-as-Software Transformation at the HFS Services-as-Software Awards 2026.
Looking at what this means for enterprise buyers and builders, the interesting elements are capital structure and deployment footprint. A primary-only Series B at this size, with insiders adding exposure, keeps cash on the balance sheet for hiring and inference and support capacity rather than providing liquidity. Fifty active deals paired with one million active users suggests a small number of large deployments rather than broad long-tail adoption, which is consistent with how HR, IT and finance agents tend to land, through central contracts followed by seat expansion.
In this author's view, the model count itself is less important than the operational premise behind it. Enterprises have learned that pinning workflows to a single foundation model creates exposure to price changes, capability regressions and policy shifts. A router that selects across frontier and open models per task trades simplicity for optionality, with cost and latency as tuning parameters alongside accuracy. Whether that 20x cost claim holds across production workloads will depend on task mix, cache hit rates and evaluation rigor, all of which are difficult to verify from outside.
Worth flagging is the shift in what is being sold. Multiyear bookings over $150 million alongside prebuilt connectors and role-specific agents point to outcomes and automation coverage rather than licenses alone. For CIOs, that changes diligence from feature checklists to exception handling, audit trails, identity controls and reversibility when an agent acts across systems of record. Ema's Okta and large-platform lineage is relevant here, though execution will matter more than pedigree.
What this enables, if the deployment numbers continue to compound, is experimentation with agentic automation inside functions that have historically resisted self-service software. Recruiting, onboarding and benefits administration are process-heavy and integration-bound. They are good candidates for agents that can read policy, pull state from multiple systems and draft the next step, provided humans retain approval authority where it counts.
