Site icon The Invincible India

Startup Deep Dive : Ema — the AI employee pitch investors didn’t get is now valued near $800 million in talks

The Invincible India
The Invincible India Startup Deep Dive featured graphic for Ema.

In early 2023, before “agentic AI” was a phrase anyone pitched with a straight face, Surojit Chatterjee left the C-suite of a company he had just taken public and started writing code again. He says roughly 80% of the investors he first pitched on “AI employees” did not understand what he meant.

Three years on, that same company, Ema, is reportedly in talks to raise fresh capital at a valuation near $800 million (~₹76.8 thousand crore, at $1 ≈ ₹96.0 as of 18 September 2026, Trading Economics) — on an estimated $15.2 million of annual revenue. The gap between the number nobody understood and the number now being discussed is the story.

Quick facts

Company Ema (Ema Unlimited, Inc.)
Founded Incorporated 27 February 2023; headquartered in Mountain View, California (Tracxn, accessed September 2026)
Founders Surojit Chatterjee (CEO), Souvik Sen (Head of Engineering)
Businesses Enterprise agentic AI — “Universal AI Employees” across HR, IT, finance, customer support and sales
Latest FY revenue Estimated $15.2 million annualised revenue as of July 2025 (GetLatka estimate; not company-disclosed)
Latest FY profit/loss Not disclosed — private, venture-funded
Listed Private; no IPO announced
Market value / last valuation Reportedly in talks for ~$800 million as of July 2026 (unconfirmed); prior rounds closed without a disclosed valuation
Key shareholders Accel, Section 32, Prosus Ventures, KPMG LLP, Hitachi Ventures, Wipro Ventures; angel backers include Sheryl Sandberg, Dustin Moskovitz and Jerry Yang

What they do

Ema sells what it calls a “Universal AI Employee” — a configurable, no-code AI agent that enterprises deploy against a specific role rather than a single task. The same underlying platform can be pointed at HR (onboarding, benefits, headcount planning), IT (access requests, ticketing, asset management), finance (payroll, expenses, timesheets), customer support, or sales enablement, using role-specific “hubs” the company introduced in September 2026. Its buyers are large, systems-heavy organisations — Ema names Hitachi, Wipro, PwC, ADP, Moneyview, TrueLayer and Envoy Global among its customers — that are drowning in fragmented software and want a layer that can act across it, not just search it.

The origin

Chatterjee’s case for Ema came out of his own frustration running large product teams at Google, Flipkart and Coinbase: talented people, he found, spent roughly half their time on the “human glue” work of keeping systems talking to each other rather than on anything that moved the business. He spent 14 years at Google, including a stint building its mobile ads business past $100 billion, then ran product at Flipkart between 2015 and 2017, before Coinbase hired him as chief product officer in February 2020 to help take the exchange public. He left in 2023 to start over as an engineer.

Ema’s co-founder, Souvik Sen, brought the machine-learning depth: he had been VP of engineering at Okta overseeing data, ML and devices, after leading data and ML engineering at Google, and holds 37 patents. Between them, the founding bet was specific — that large language models, freshly capable of planning, using tools and holding memory after ChatGPT’s late-2022 debut, could be assembled into something that behaved less like software and more like a hire.

The struggle years

The idea did not arrive fully formed. The team’s first serious attempt, built in 2023, was an analytics product meant to help enterprises tame what Chatterjee called a “spreadsheet jungle.” Within a couple of months they concluded it did not have a real market and pivoted toward HR operations instead, reasoning that something as routine as employee onboarding alone touched twenty to thirty separate internal systems — a more tractable and more painful wedge.

The pitch itself was a harder sell than the product. Chatterjee has said that when he described “AI employees” to early investors, around 80% did not understand what he was proposing, because the term “agentic AI” was not yet in circulation. Rather than chase a market that did not yet have the vocabulary to buy from him, Ema spent roughly 18 months in stealth with no public website, using the time to build compliance and infrastructure most seed-stage startups defer: SOC 2 Type I and Type II, ISO 42001, GDPR and HIPAA coverage, a multi-cloud containerised architecture, and an air-gapped deployment option for customers who would not accept data leaving their own environment. It was an unusually expensive way to spend a company’s first year and a half without revenue to show for it.

The turning point

The proof point that changed the sales conversation, according to Ema and its own published case study, was a production deployment inside Hitachi. Ema’s AI employees were used to unify five separate HR systems across more than 40,000 of Hitachi’s roughly 400,000 employees, going from idea to live deployment in under eight weeks. The results Ema and Hitachi Ventures have publicised: a 70% increase in HR operational efficiency, a 30% month-on-month fall in HR ticket volume, a resolution accuracy above 90%, and onboarding tasks that used to take up to 15 days cut to minutes.

Before Hitachi, Ema’s evidence of fit was a handful of smaller pilots. After it, the company had a large, name-brand conglomerate willing to publicly attach numbers to a production rollout — not a proof of concept — and Hitachi Ventures went on to become an investor. That combination of a public reference customer and a strategic investor from the same company is the leverage Ema has used since to open doors with Wipro, PwC, ADP and others.

The money behind it

No official valuation was disclosed at either the seed or the Series A close, which is common for private rounds this size; the $800 million figure is a reported target for a round still in progress as of the most recent coverage, not a confirmed post-money number.

How it makes money

The numbers

Ema is private and has not disclosed audited multi-year revenue or profit-and-loss figures, so a conventional three-to-four-year revenue table is not something the public record supports yet. What is verifiable is a mix of estimated revenue, disclosed capital raised, and headcount growth, which together sketch the trajectory:

Period Disclosed capital raised (cumulative) Headcount Revenue
December 2022 Not yet funded ~1 (pre-incorporation team) None
March 2024 (stealth exit) $25 million Not disclosed Not disclosed
July 2024 (Series A close) $61 million Not disclosed Not disclosed
October 2024 $61 million + undisclosed KPMG stake 86 (Humanloop, October 2024) Not disclosed
July 2025 Unchanged (no new round disclosed) Not directly reported ~$15.2 million estimated ARR (GetLatka estimate)
March–July 2026 Reportedly in talks for +$80 million (unconfirmed) 221–246 across trackers (Revelio Labs, March 2026; Tracxn, May 2026; Crustdata, July 2026) Not disclosed

The revenue figure is worth a caveat rather than a rewrite: GetLatka’s own page describes the $15.2 million as a modelled estimate, not a management-reported number, and the company has not published an official ARR figure. It is included, clearly labelled, because it is the only revenue estimate this research could find with a stated methodology and date; no profit or loss figure of any kind — audited or estimated — could be located, so none is presented here.

Where the money comes from

The risks

The takeaway

The most transferable part of Ema’s story is not the funding or the Hitachi numbers — it is the decision to spend 18 months building compliance certifications and infrastructure nobody was asking for, while 80% of investors could not follow the pitch. Most founders read investor confusion as a signal to simplify the story. Chatterjee read it as a signal that the story would only make sense once there was a production deployment large enough to make the argument unnecessary. Being early enough that the market lacks the words for what you do is not, by itself, an advantage — it only becomes one if you use the silence to build the proof that will do the talking once the words exist.

Frequently asked questions

What does Ema actually sell?

Ema sells configurable, no-code AI agents it calls “Universal AI Employees,” deployed against specific enterprise functions — HR, IT, finance, customer support and sales — through packaged “hubs” it launched in September 2026.

Who founded Ema and when?

Surojit Chatterjee, formerly chief product officer at Coinbase and a VP of product at Google, and Souvik Sen, formerly VP of engineering at Okta, founded Ema in early 2023; the company was incorporated on 27 February 2023 and emerged from stealth on 5 March 2024.

How much funding has Ema raised?

Ema has disclosed $61 million across a $25 million seed (March 2024) and a $50 million total Series A (July 2024), plus an undisclosed minority stake taken by KPMG LLP in October 2024. As of July 2026, it was reportedly in talks for a further $80 million, unconfirmed at the time of writing.

What is Ema’s valuation?

No official valuation has been disclosed for the seed or Series A rounds. Moneycontrol and Whalesbook reported in July 2026 that Ema was in talks to raise at roughly an $800 million valuation, led by Creaegis, but described the round as still under negotiation, not closed.

Who are Ema’s customers?

Publicly named customers include Hitachi, Wipro, PwC, ADP, Moneyview, TrueLayer, Envoy Global and Hospital for Special Surgery, spanning financial services, conglomerates, professional services and healthcare.

Sources

Figures are as of September 2026. Currency converted at $1 ≈ ₹96.0 as of 18 September 2026 (Trading Economics).

Found an error? Write to us and we’ll correct it in the open, dated, on the piece.

Exit mobile version