How Ema’s AI Employees Are Rewriting the Enterprise Software Buyer’s Job

Ema just raised $77 million to prove that the person who used to buy your company's HR software might soon be managing a team of AI agents instead of a vendor relationship. The Series B, led by Bengaluru-based Creaegis with Accel, Section 32, and Prosus all increasing their stakes, brings Ema's total funding to $140 million and more than quadruples its valuation from 2024, according to TechCrunch. The round was pure primary equity — no debt, no secondary sales — which signals investors betting on growth, not cashing out early believers.

The headline number matters less than what Ema is selling: not another SaaS tool, but a replacement for the operator role itself.

The Pitch: AI Employees, Not AI Features

Founded in 2023 by former Google and Coinbase executive Surojit Chatterjee and ex-Okta executive Souvik Sen, Ema builds what it calls "AI employees" — coordinated teams of agents that execute multi-step business processes across HR, IT, and finance rather than handling one task at a time, per TechCrunch. Chatterjee's stated end goal is blunt: reduce enterprise dependence on the SaaS applications that already dominate corporate budgets. Ema first "wraps" around a customer's existing software stack, then — according to Chatterjee — many customers begin phasing that software out entirely, reducing it to what he called essentially a database layer.

That's a direct challenge to the seat-license economics that have defined enterprise software for two decades.

Meet the Operator This Is Built For

To understand what changes, it helps to picture a specific role: the shared services director who owns HR case management, vendor onboarding, and monthly financial close. This is an illustrative composite, not a named Ema customer, but it reflects the job function Ema is explicitly targeting.

Before Ema, this operator's week looks like a relay race across disconnected systems. An employee files a benefits question in the HR portal; it routes to a ticketing queue; a human agent manually cross-references policy documents, then updates three separate systems — HRIS, payroll, and a compliance tracker — before closing the ticket. Multiply that by finance close reconciliations and IT access requests, and the operator spends most of their time as a human API between systems that don't talk to each other, plus a rotating cast of IT services contractors brought in to patch integrations.

After Ema, the operator's job shifts from executing tasks to supervising outcomes. Agents ingest the same ticket, pull the relevant policy, update HRIS and payroll simultaneously, and flag only the exceptions that need human judgment. The operator's daily work becomes reviewing agent decisions, tuning workflows, and expanding automation into adjacent processes — which tracks with Chatterjee's claim that more than 90% of Ema's customers expand beyond their initial use case, some across dozens of workflows, as reported by TechCrunch.

What the Adoption Numbers Actually Say

Ema reports more than 50 active enterprise deals, over 1 million active enterprise users, and more than 5 million actions and queries handled, with customers including NTT DATA, Hitachi, ADP, PwC, Google, KPMG, Wipro, and Microsoft, according to TechCrunch. Revenue has grown 50-fold over two years, and cumulative revenue bookings — the total value of multiyear contracts, not annual recurring revenue — have surpassed $150 million. Net dollar retention sits around 180%, meaning existing customers are dramatically expanding spend rather than just renewing.

Analysis: the 90%-plus expansion rate is the number operators should scrutinize hardest. It suggests Ema's real product isn't the first automated workflow — it's the platform effect of getting one workflow live and then having internal champions push for the next ten. That's the same land-and-expand playbook SaaS companies used for years, now running on an agent architecture designed to eventually cannibalize the software it wraps.

An Implementation Sequence Worth Studying

Based on how Chatterjee describes deployment, a rough sequence emerges for any operator evaluating this model:

• Wrap, don't replace. Ema connects to existing HRIS, ERP, and IT systems first, leaving legacy software in place while agents handle the workflow layer on top.
• Prove one workflow end-to-end. Early deployments target a single high-friction process — benefits inquiries, vendor onboarding, IT access requests — where multi-step coordination across systems is the bottleneck.
• Let outcomes justify expansion. Because Ema prices on task completion and business outcomes rather than seats or token consumption, per TechCrunch, operators can point to measurable results before requesting budget for the next workflow.
• Reassess the underlying software. Once agents handle enough of the process, the original SaaS license becomes a data store rather than an active tool — the point where, per Chatterjee, some customers begin removing dependency on it altogether.

Success Criteria for Operators Considering This Path

The metrics that matter aren't uptime or ticket volume — they're the same ones Ema uses to sell itself. Track expansion rate (are workflows spreading to new departments without new procurement cycles?), net retention (is the same team spending meaningfully more after twelve months?), and margin impact (is your services spend on IT consultants and integrators shrinking as agents absorb implementation work?). Ema claims gross margins near 80%, attributing improvement over time to systems that need less human support as they learn from deployments — a dynamic operators should demand evidence of internally, not just take on faith.

The Tradeoff Nobody's Advertising

The same forward-deployed engineering push that Anthropic and OpenAI have made into enterprise operations, as noted in TechCrunch's coverage, means Ema isn't competing in a vacuum — it's racing frontier labs and traditional IT services firms simultaneously, some of which are already restructuring their own business models in response. Chatterjee frames model progress as a tailwind because Ema draws on more than 150 models rather than building its own. That's a reasonable technical bet, but it also means Ema's moat is domain integration and orchestration, not model capability — a moat that gets tested every time a frontier lab ships a more capable agent framework of its own. Operators adopting this model should treat vendor lock-in risk as real: replacing SaaS with an agent layer only reduces dependency if that layer itself doesn't become the next unreplaceable system.

Sources

TechCrunch AI — "Ema raises $77M as AI starts eating into enterprise software and services" — https://techcrunch.com/2026/09/23/ema-raises-77m-as-ai-starts-eating-into-enterprise-software-and-services