AI employee is the category Ema has bet its company on, and on 23 September 2026 investors put another $77 million behind that bet. The Series B was led by Bengaluru-based Creaegis, with existing backers Accel, Section 32 and Prosus all increasing their stakes. It takes total funding to $140 million and more than quadruples the valuation set in 2024, although Ema declined to name the new figure.
What makes the round worth reading closely is not the number. It is the argument attached to it. Ema is not selling a faster way to use enterprise software; it is selling a layer that wraps that software and, in its founder’s telling, eventually replaces much of it. Autonomous AI agents have been pitched that way for two years. Ema is the first AI employee vendor arriving with named logos, a net dollar retention figure and a gross margin. Our AI employees and autonomous AI agents page covers the wider category.
This article sets out the round’s structure, separates the metrics that mean something from the ones that are presented to look like revenue, does the arithmetic on the two named customer deployments, and examines the outcome-based pricing model that makes the whole pitch coherent.
Table of contents
- What an AI Employee Is in Ema’s Definition
- The AI Employee Round, Structured
- The AI Employee Traction Numbers, and What Each One Measures
- What an AI Employee Deployment Looks Like in Practice
- Why an AI Employee Threatens Services Revenue, Not Just Software
- The Pricing Model That Makes the AI Employee Pitch Coherent
- The AI Employee Risks Ema’s Numbers Do Not Cover
- What to Take From the Round If You Are Buying, Not Investing
- Frequently Asked Questions About the Ema AI Employee Round
- References
What an AI Employee Is in Ema's Definition
The phrase is doing specific work here, and it is not a synonym for chatbot or copilot.
Multiple agents, coordinated
Ema describes an AI employee as a system that coordinates multiple agents rather than a single model answering a single request. The coordination is the AI employee product: the value claim rests on completing a multi-step business process end to end, not on producing a good answer to one question.
It works across the applications you already have
Rather than replacing a system of record, an AI employee operates across a company’s existing applications. Chatterjee describes this as wrapping the estate first. That ordering matters commercially, because it means the product can be bought without a migration project in front of it.
Three named components
Ema’s platform is built around Autopilot for autonomous execution, an Agent Library of prebuilt AI employee configurations, and a Generative Workflow Engine that assembles the steps. The library is the part that makes deployment weeks rather than quarters.
The functions it targets
HR, IT and finance are the three departments an Ema AI employee is sold into. All three share a profile: high volume, repetitive, heavily ticketed, and already instrumented, which means an AI employee deployed there has both plenty of work and plenty of measurement.
Model-agnostic by design
Ema’s software can draw on more than 150 models, frontier and open source. Chatterjee says the company focuses on domain knowledge, integrations and orchestration instead, and does not view the frontier labs as direct competitors: “Progress in frontier models is actually very beneficial to us.”
The AI Employee Round, Structured
The financing details behind the AI employee round are unusually clean, and the cleanliness is itself informative.
| Item | Detail |
|---|---|
| Amount | $77 million |
| Stage | Series B |
| Lead investor | Creaegis, Bengaluru |
| Returning investors | Accel, Section 32, Prosus, all increasing stakes |
| Total raised | $140 million |
| Valuation | More than 4x the 2024 round; figure undisclosed |
| Structure | Entirely primary equity, no debt, no secondary |
| Founded | 2023, Mountain View |
| Founders | Surojit Chatterjee (ex-Google, ex-Coinbase), Souvik Sen (ex-Okta) |
| Headcount | Nearly 200; offices in Bengaluru, London, Vancouver |
No secondary means no founder liquidity
An all-primary AI employee round with no secondary component means none of the money went to existing shareholders cashing out. Every dollar is working capital. At this stage that is a signal about intent rather than about valuation.
The lead is Indian, the customers are not yet
Creaegis leading from Bengaluru while Ema sells primarily into the US and Europe fits the company’s stated plan to expand across Asia-Pacific, South America and parts of the Middle East over the next year. Prakash Parthasarathy, Creaegis managing partner and chief investment officer, framed the thesis around scale already achieved rather than potential.
More than half the money arrived this week
The company has raised more in this one round than in its entire prior history, which is what a four-times valuation step usually looks like from the outside. It also means the operating history behind the metrics below was funded by the smaller half.
The AI Employee Traction Numbers, and What Each One Measures
Ema published an unusually full set of AI employee figures. They are not all the same kind of thing, and the differences matter.
| Figure | Value | What it tells you | What it does not |
|---|---|---|---|
| Revenue bookings | Over $150 million | Total contract value signed | Annual recurring revenue |
| Revenue growth | 50x over two years | Rate of change | The base it started from |
| Net dollar retention | Around 180% | Existing customers expanding | New logo acquisition |
| Gross margin | Close to 80% | Software-like unit economics | Operating profitability |
| Active enterprise deals | More than 50 | Customer count | Contract size distribution |
| Active enterprise users | Over 1 million | Reach inside customers | Frequency of use |
| Actions and queries | More than 5 million | Cumulative work done | Success rate |
| Expansion beyond first use case | More than 90% of customers | Land-and-expand working | Revenue per expansion |
Bookings are not revenue, and Ema said so
Chatterjee was explicit that the $150 million bookings figure includes the total value of multiyear contracts, two- and three-year deals among them, rather than representing annual recurring revenue. He declined to disclose the current annualised run rate. Reading bookings as revenue would overstate the business substantially.
50x growth without a base is a rate, not a size
A fiftyfold increase from an undisclosed starting point is compatible with almost any absolute figure. It is a genuine signal about trajectory and a meaningless one about scale, and the pairing of a disclosed multiple with an undisclosed base is a deliberate choice.
180% net dollar retention is the strongest number in the set
Net dollar retention around 180% means that for every $100 an existing cohort spent last year, the same cohort is spending about $180 now — an $80 expansion before a single new customer is counted. Combined with more than 90% of customers going beyond their first use case, it describes an AI employee product that grows inside an account without a new sale.
80% gross margin is the claim under most pressure
Software margins on an AI employee that also absorbs implementation and integration work are hard to hold. Chatterjee’s explanation is that the systems need less human support as they learn from deployments, so margins improve over time. That is a plausible mechanism and an unverified one.
What an AI Employee Deployment Looks Like in Practice
Two AI employee deployments were described with figures attached, which is rarer than it should be in this category.
| Customer | Scale | Reported outcome | Time to production |
|---|---|---|---|
| Wipro | 240,000 employees, 65 countries, 2.9m queries a year | 50% reduction in IT support tickets | Not stated |
| Hitachi | Not stated | 30% fewer support tickets, 70% efficiency gain claimed | Under four weeks |
The Wipro figures divide into something checkable
Two point nine million queries a year across 240,000 employees works out at roughly 12.1 queries per employee per year, or about 7,945 queries a day across the whole estate. Those are modest per-person numbers, which is exactly what a help-desk deflection workload looks like — a lot of people asking occasionally, not a few people asking constantly.
Under four weeks is the number that sells the category
Hitachi going from concept to an AI employee in production in under four weeks is the claim most likely to move a buyer, because it addresses the objection that every previous automation programme took a year. The prebuilt Agent Library is the stated mechanism.
Percentage reductions need a denominator
A 50% cut in IT support tickets and a 70% efficiency increase are both reported without the baselines they are measured against. They are the vendor’s figures, not audited results, and should be read as directional.
The named customer list is the strongest evidence
NTT DATA, Hitachi, ADP, PwC, Google, KPMG, Wipro and Microsoft are all named as customers. Four of those are themselves IT services or professional services firms, which is directly relevant to the claim in the next section.
Why an AI Employee Threatens Services Revenue, Not Just Software
The headline framing — an AI employee eating into enterprise software and services — is a two-part claim, and the services half is the less obvious one.
The SaaS argument, in Chatterjee’s words
He describes customers as “already on the way to replace [large SaaS applications] completely, removing dependency on them, because they are mostly becoming like a database”. The argument is that once an AI employee sits over the top of a system of record, the interface, workflow and reporting layers people actually paid for become redundant.
The services argument is the sharper one
Chatterjee also expects AI to absorb implementation, integration and consulting work that companies have traditionally paid IT services firms to do around that software. “A lot of the services companies are working with us,” he said. “They are also dramatically changing or disrupting their own business models because they understand the human-forward model may not be the best model going forward.”
Wipro and NTT DATA are on both sides of that trade
Two of the named customers sell exactly the implementation services this thesis says will shrink. Deploying an AI employee internally while selling the human-delivered version externally is a hedge, and it is how large services firms have historically navigated every automation wave.
The frontier labs are moving into the same space
Anthropic has expanded efforts to bring Claude into core company operations, including financial and legal work, and OpenAI has built teams of forward-deployed engineers who work alongside customers. Chatterjee argues the labs are suppliers rather than rivals, since Ema draws on more than 150 models — but the forward-deployed engineering model competes for the same budget line. We covered the pricing side of that fight in our piece on the AI price war between Anthropic and OpenAI.
The Pricing Model That Makes the AI Employee Pitch Coherent
An AI employee that claims to replace software and services cannot be priced like software, and Ema does not price it that way.
Not seats, not tokens
Ema charges for an AI employee neither per software seat nor per token consumed. Pricing is tied to the completion of tasks and business outcomes. That single decision is what lets the company argue it is replacing a headcount-shaped cost rather than adding a licence-shaped one.
| Model | Buyer pays for | Vendor incentive | Weakness |
|---|---|---|---|
| Per seat | People with logins | Maximise named users | Breaks when software does the work |
| Per token | Model consumption | Maximise verbosity | Cost unpredictable, value untracked |
| Outcome-based | Completed tasks | Maximise successful completions | Requires agreeing what counts as done |
Outcome pricing exposes the vendor to its own accuracy
If revenue depends on tasks completing, a system that fails silently does not get paid. That aligns incentives better than either alternative and puts real pressure on evaluation, natural language processing quality and error handling rather than on model size.
The catch is definitional
Outcome-based pricing needs both sides to agree what a completed task is. In HR and IT ticketing that is comparatively easy — a ticket resolves or it does not. In finance it is considerably harder, and that is where disputes will land.
Spend per customer is already doubling
The average Ema customer doubled AI employee spending over the last year. Under outcome pricing that is a usage statement rather than a licensing one: customers are not buying more seats, they are routing more work through the system.
The AI Employee Risks Ema's Numbers Do Not Cover
Every figure in the announcement describes something working. Three categories of risk sit outside that frame entirely, and none of them is unique to Ema.
Access concentration
An AI employee spanning HR, IT and finance holds credentials across all three. That is the same structural exposure that made a single stolen token so serious in Meta’s assistant, and it applies with more force in an enterprise, where the accounts involved are shared rather than personal.
Silent failure in multi-step work
A single-shot model that gets something wrong produces a visibly wrong answer. An AI employee executing a nine-step process can complete eight steps correctly and corrupt the ninth, and the output still looks like a finished job. Outcome-based pricing helps here only if the definition of a completed task includes correctness.
Dependency where the SaaS used to be
The pitch is that the underlying applications become commodity databases. If that happens, the strategic dependency does not disappear; it moves to the orchestration layer, and that layer is a private company three years old rather than an incumbent with two decades of escrow arrangements and exit clauses behind it.
Measurement gets harder, not easier
Ticket-reduction percentages are easy to produce and hard to audit, because deflected tickets are tickets nobody filed. A 50% reduction can reflect genuine resolution or users giving up. Any AI employee business case should pair the deflection number with a satisfaction or reopen-rate measure.
Governance has not caught up
There is no established audit standard for an autonomous system acting under a human’s authority across finance systems. Buyers are currently writing those controls themselves, which is a cost that does not appear in the licence and belongs in the evaluation.
What to Take From the Round If You Are Buying, Not Investing
The interesting question for most readers is not whether Ema is worth its valuation but what the round says about the AI employee category.
Named logos have arrived, audited results have not
Eight large enterprises named, two AI employee deployments with figures, and no third-party verification of any of it. That is the current state of evidence for an AI employee at enterprise scale — much better than a year ago, still short of what a procurement team should accept without its own pilot.
Ask for the base, not the multiple
Any vendor quoting 50x growth should be asked what it grew from. Any vendor quoting bookings should be asked for annual recurring revenue. Ema was candid about the distinction, which is a reasonable standard to hold others to.
Time to production is the metric to negotiate on
Under four weeks from concept to production is the claim that most changes a business case. Put it in the contract as a milestone rather than treating it as marketing, and make the first deployment one with a measurable ticket baseline.
Watch what happens to your services spend
If the services thesis is even partly right, the second-order effect of deploying an AI employee is a smaller integration and support bill. That saving belongs in the business case, and it is usually owned by a different budget from the one buying the software.
Frequently Asked Questions About the Ema AI Employee Round
How much did Ema raise and who led it?
$77 million in a Series B led by Creaegis, with existing investors Accel, Section 32 and Prosus increasing their stakes. The round was entirely primary equity with no debt or secondary component.
What is Ema’s total funding and valuation?
Total funding now stands at $140 million. The round more than quadrupled the valuation set in 2024, but Ema declined to disclose the new figure.
What exactly does Ema sell?
An AI employee: a system coordinating multiple AI agents to carry out multi-step business processes across a company’s existing applications, built around Autopilot, an Agent Library and a Generative Workflow Engine, targeting HR, IT and finance.
Is the $150 million figure revenue?
No. It is revenue bookings, including the total value of two- and three-year contracts. Chatterjee explicitly distinguished it from annual recurring revenue and declined to give the current run rate.
Who are Ema’s AI employee customers?
Named customers include NTT DATA, Hitachi, ADP, PwC, Google, KPMG, Wipro and Microsoft, across more than 50 active enterprise deals and over 1 million active enterprise users.
How does Ema charge for an AI employee?
Not per seat and not per token. Pricing is tied to completed tasks and business outcomes, which is what supports the claim that the product substitutes for headcount rather than adding a licence.
Who founded Ema?
Surojit Chatterjee, a former Google and Coinbase executive, and Souvik Sen, formerly of Okta, in 2023. The company is headquartered in Mountain View with offices in Bengaluru, London and Vancouver and nearly 200 staff.
References
Ema raises $77M as AI starts eating into enterprise software and services
Ema raises $77M to deploy AI employees across enterprise HR, IT and finance
Ema Raises $77M Series B to Put AI Employees to Work Across the Enterprise
Ema Raises $77M Series B as Outcome-Based Pricing Challenges the Per-Seat SaaS Model
Ema raises $77 million as enterprise AI pushes into software and IT services
More AI coverage: explore Progressive Robot's AI Models, Tools & Releases hub — hands-on reviews, setup guides and benchmarks in one place.
