Ema, a startup that coordinates teams of AI agents to run HR, IT, and finance workflows, closed a $77 million Series B led by Bengaluru-based Creaegis. Accel, Section 32, and Prosus, all prior backers, added to their existing stakes in the round. That capital pushes Ema’s total funding past $140 million, a sum that more than quadruples the valuation the company set during its last round in 2024, though the new figure stays undisclosed. Every dollar arrived as primary equity: no debt, no secondary sales, the startup told TechCrunch.
The raise lands as AI vendors, established software companies, and enterprise labs all compete for a pool of spending that used to belong almost entirely to SaaS subscriptions and IT services contracts. Ema wants a bigger share of that pool by selling what it calls “AI employees,” agent bundles that carry out multi-step workflows spanning a company’s existing systems, instead of tackling one task at a time.
Cofounder Surojit Chatterjee, a former Google and Coinbase executive, described the strategy as a wedge. Ema’s agents first wrap around a customer’s existing applications, he told TechCrunch, then customers gradually reduce their dependence on those tools and, in some cases, drop them entirely. “Many of our customers are already on the way to replace [large SaaS applications] completely, removing dependency on them, because they are mostly becoming like a database,” Chatterjee said.
That is a direct challenge to the SaaS incumbents whose seat-based pricing built enterprise software into one of tech’s largest markets. Ema’s own pricing rejects that model too: it charges by completed task and business outcome rather than by seat or token consumption, a bet that agent labor should be priced like output, not access.
Frontier labs are closing in on the same territory. Anthropic has pushed Claude into corporate finance and legal work, and OpenAI has staffed forward-deployed engineering teams that embed with customers to get its models into production. Chatterjee said he does not see those labs as competitors: Ema’s software draws on more than 150 underlying models, he told TechCrunch, so better frontier models make Ema’s product better rather than obsolete. That framing is convenient for a company whose survival depends on model progress continuing, and it sidesteps the question of what happens if a lab decides to build Ema’s orchestration layer itself.
The company’s growth numbers, as Chatterjee described them to TechCrunch, are steep. Ema counts upward of 50 active enterprise engagements and beyond 1 million active users, spanning clients such as Google, Microsoft, and KPMG, plus NTT DATA, Hitachi, ADP, PwC, and Wipro. Revenue has grown 50 fold over two years, and cumulative revenue bookings, counting the full value of multiyear contracts rather than annual recurring revenue, have topped $150 million. Chatterjee declined to give an annualized run rate, which is the figure investors would actually use to judge whether that bookings number reflects durable revenue or a handful of large multiyear commitments pulled forward.
Two retention figures stand out because they describe expansion rather than just signing. Chatterjee said that upward of 90 percent of Ema’s customer base has moved beyond its original use case, and net dollar retention sits around 180 percent, meaning existing accounts are spending well beyond their starting contract value. He also said gross margins remain near 80 percent even as Ema absorbs implementation and integration work once billed by IT services firms, crediting that margin to agents that need less human oversight as they accumulate deployment experience.
Chatterjee said most of the new capital will go toward sales and marketing, after several years spent mainly on product work. Headquartered in Mountain View, the company has grown to roughly 200 employees, with offices spread across Bengaluru, London, and Vancouver, and plans to expand beyond its current U.S. and European base into Asia-Pacific, South America, and the Middle East over the next year.
The bookings-versus-run-rate distinction is the number worth watching here: a startup selling multiyear enterprise deals can post an impressive cumulative figure while its actual annual revenue is far smaller, and Ema has not said which is true. Any operator evaluating an “AI employee” vendor over the next quarter should ask for the annualized figure directly rather than accepting bookings totals at face value.
Reported by Jagmeet Singh for TechCrunch, September 23, 2026.
