Ema raises $77M Series B to scale enterprise AI Employees

Ema said it raised a $77 million Series B led by Creaegis, with Accel, S32, and Prosus increasing stakes — total funding $140M and a more-than-quadrupled valuation — to expand AI Employees across enterprise HR, IT, and finance.

FINANCE desk — the money is chasing AI that replaces workflows, not chat demos: Ema’s raise is a bet that “AI Employees” can eat the SaaS + services stack once they survive production.

What an independent report adds on the same day, and what it does not. TechCrunch’s story by Jagmeet Singh is stamped 5:00 a.m. PDT on September 23, 2026. That is 8:00 a.m. Eastern. It calls Creaegis a Bengaluru venture firm. It writes the third investor as Section 32. The company post and the announcement graphic write S32. This filing treats them as the same named investor, under the spelling on the graphic. TechCrunch says the valuation more than quadruples from the last funding round in 2024, and that Ema declined to disclose the latest valuation. The year 2024 is TechCrunch’s. It is not a date on the company post. The story also says the startup confirmed the round was entirely primary equity, with no debt and no secondary sale. Primary equity means new shares sold by the company, so the money goes into the company. No debt means nobody lent this round. No secondary means existing holders did not sell their shares in it. That confirmation is TechCrunch’s, from the company. The blog does not print it. This desk did not see the share register.

What Ema sells. The company calls the product AI Employees. TechCrunch describes them as teams of AI agents that automate corporate processes across human resources, IT, and finance. An agent, here, is software that takes a sequence of steps, not only a chat reply. Multi-step means the work has more than one action, such as opening a request, checking a policy, and sending the answer, across the apps the company already runs. Human resources, shortened to HR, is hiring, pay, and the employee desk. IT is the technology help desk. Finance is the money work. The company post says a customer can start in one function and expand across HR, IT, and finance on the same platform. Chatterjee told TechCrunch the software first wraps around those existing apps, and that many customers are already on the way to replace large software-as-a-service products, because those products are “mostly becoming like a database.” Software as a service, shortened to SaaS, is software you rent by the login instead of installing it yourself. “Many” and “on the way” are his words. They are not a count of canceled contracts. This desk did not watch a workflow run.

Who is speaking, and the Wipro example. The company post calls Surojit Chatterjee CEO and Founder. He said customers are not running experiments. They are running HR, IT, and finance operations on AI Employees at a scale of millions of interactions a year, and that very few agentic AI companies have reached that. Agentic, here, means the software takes the steps, not only answers a question. A second line on the post says enterprises do not need more software. They need outcomes. The round, he said, will push that change at companies still stuck in pilot purgatory. A pilot is a trial. Purgatory, in that sentence, means the trial looks good in a demo and never becomes the real operation. TechCrunch says he is a former Google and Coinbase executive, and that he founded Ema in 2023 with Souvik Sen, a former Okta executive. The company page this desk read does not print those past jobs or Sen’s name. Do not treat the blog title as a biography. At Wipro, the company says, Ema powers an employee assistant for more than 240,000 associates across 65 countries. It automates more than 100 workflows and handles approximately 2.9 million employee queries a year. A workflow is a repeatable process. A query is one question an employee asks. The company also says response times dropped from days to seconds, and employee satisfaction went up 20 percent. Twenty percent is about one fifth higher, if that measure holds. Those Wipro figures are the company’s. This desk did not audit Wipro.

The scale lines TechCrunch reports, and whose they are. The story says Ema has more than 50 active enterprise deals, more than 1 million active enterprise users, and has handled more than 5 million actions and queries. Customers named there include NTT DATA, Hitachi, ADP, PwC, Google, KPMG, Wipro, and Microsoft. Over the past two years, the story says, Ema said revenue grew 50-fold, while revenue bookings surpassed $150 million. Fifty-fold means fifty times the earlier revenue, if the claim holds. The company post prints the same kind of figure as revenue growing 50 times over the past 24 months. Twenty-four months is two years. Chatterjee told TechCrunch the bookings figure is the total value of multi-year contracts, including two- and three-year deals, not annual recurring revenue. Annual recurring revenue, shortened to ARR, would be one year’s subscription value. Bookings here are the whole contract. He declined to disclose the current annualized revenue run rate. A run rate takes recent sales and stretches them to a year. More than 90 percent of customers have expanded past the first use case, he said, and some use the technology across dozens of workflows. Net dollar retention is around 180 percent. Net dollar retention, shortened to NDR, is how much the same customers spend a year later, counting expansions and losses. Around 180 percent means those customers are spending a bit under twice as much, if the rate holds. The company post says something nearby and not the same: on average, each customer spends twice the initial spend within 12 months of signing. Twice in a year is the blog. Around 180 percent NDR is the interview. Do not merge them. Gross margin is close to 80 percent, he said. A gross margin is what is left after the direct cost of delivering the product, before the rest of the company’s costs. Close to 80 percent is about four fifths. These figures are what TechCrunch attributes to Ema and to Chatterjee, except the 50-times and twice-the-spend lines, which are also on the company post. This desk did not see the books.

How the platform is described, and where the two pages do not match. The company post says Ema combines the output of more than 100 large language models, shortened to LLMs, tuned for common enterprise jobs, and that this is meant to be more accurate and cheaper than using one model such as GPT for every task. An LLM is a model that reads and writes text. TechCrunch quotes Chatterjee saying the software can draw on more than 150 models, including frontier models and open-source models, while Ema focuses on the domain knowledge, the connections, and the orchestration that finish a process. One hundred plus is the blog. More than 150 is the interview. This filing does not merge them into one count. The company also says time to value is under two months, and that it delivers more than 95 percent accuracy at scale. It calls that accuracy industry-leading. Under two months, more than 95 percent, and industry-leading are the company’s words. This desk did not time a rollout or score the answers. Pricing, as Chatterjee told TechCrunch, is tied to finishing a task or a business outcome, not to software seats and not to tokens. A seat is a paid login. A token is a small chunk of text the model reads or writes. The company post does not print that price rule. File it as the interview’s.

Where the money goes, and how big the company says it is. The company post says the new funds will scale the go-to-market organization and expand into new places, including Asia-Pacific and Europe, the Middle East, and Africa, which it shortens to APAC and EMEA. Go-to-market means sales and the work of getting the product into more companies. TechCrunch says much of the capital will go to sales and marketing, after the first years were spent building the product. It says the Mountain View headquarters has grown to nearly 200 employees, with offices in Bengaluru, London, and Vancouver. The company page prints the address 321 Castro St, Mountain View, CA 94041. It does not print a headcount. Nearly 200 is TechCrunch’s. TechCrunch also says Ema has focused so far on customers in the United States and Europe, and plans new markets over the next year, particularly Asia-Pacific, South America, and parts of the Middle East. The blog names APAC and EMEA. The interview adds South America and is more specific about the Middle East. File each list as the page that printed it. This desk did not count the staff.

The investor’s words, and the blanks left open. Prakash Parthasarathy, managing partner and chief investment officer at Creaegis, said many agentic AI companies have compelling demos, and few have proven they can deliver in production at enterprise scale. He said Ema’s platform already supports millions of employee interactions at some of the world’s largest organizations, with the governance those companies require, and that the business is a product platform with strong economics, not a services model that needs a custom project for every customer. Governance, here, means the rules and the record of what the software did. A quote is not a customer count this desk audited. There is no dollar split among Creaegis, Accel, S32, and Prosus. There is no exact valuation. There is no hour on the company post.

Plain English for the rest of the card: Series B = a later private funding round, after earlier seed and Series A checks. $77 million is this round. $140 million is the total the company states. The page does not name the earlier checks. valuation = a price on the whole company. The new price is undisclosed. More than quadruples is the company’s comparison with the previous round. primary equity = new shares, so the cash goes into the company. AI Employee = Ema’s name for software that runs a multi-step HR, IT, or finance process. agent = software that takes steps, not only a chat reply. pilot purgatory = a trial that never becomes the real operation. SaaS = software rented by the login. workflow = a repeatable process. query = one question. 240,000, 65 countries, more than 100 workflows, and about 2.9 million queries a year are the company’s Wipro figures. 50 times / 50-fold = the revenue-growth claim. $150 million bookings = multi-year contract value, not one year’s ARR. NDR = how much the same customers spend later. Around 180 percent is the interview. Twice the initial spend within 12 months is the blog. LLM = a text model. More than 100 is the blog’s model count. More than 150 is the interview’s. seat = a paid login. token = a small chunk of text. nearly 200 = TechCrunch’s headcount. This filing is the 23 Sep announcement. It is not a priced public listing.

PRIMARY here: Ema’s 23 Sep 2026 post, “Ema Raises $77M Series B to Put AI Employees to Work at Enterprises,” at ema.ai — Tier A PRIMARY, the company’s own record. The page did not print an hour. TechCrunch’s same-day story, stamped 5:00 a.m. PDT, is the independent report. The $77 million Series B, the Creaegis lead, Accel, S32, Prosus, the $140 million total, the more-than-quadrupled valuation with no exact price, the Chatterjee title and the millions-of-interactions quote, the pilot-purgatory line, the Wipro figures, the Parthasarathy quote, the more-than-100 LLMs, the 50-times revenue line over 24 months, the under-two-months and more-than-95-percent lines, the twice-the-spend line, the APAC and EMEA plan, and the Mountain View street address are the company’s. The 2024 prior-round year, the primary-equity confirmation, the Section 32 spelling, the Google, Coinbase, and Okta biographies, the 2023 founding with Souvik Sen, the more-than-50 deals, the more-than-1-million users, the more-than-5-million actions, the customer names beyond Wipro, the $150 million bookings as multi-year contract value, the declined run rate, the more-than-90-percent expansion, the around-180-percent NDR, the close-to-80-percent gross margin, the task-and-outcome price, the more-than-150 models, the nearly-200 headcount, the Bengaluru, London, and Vancouver offices, and the South America plan are TechCrunch’s. NOT claimed: an exact valuation, a split of the $77 million, that this desk audited Wipro, the books, the headcount, or the model counts, that “on the way” to replace SaaS is a canceled-contract count, that the blog’s 100-plus models and the interview’s 150-plus models are one number, that twice-the-spend and 180 percent NDR are the same metric, a stock tip, or investment advice. Distinct from the already-filed verda-189m, subconscious-51m, and go-ai-85m.