Ema’s $77M Series B Is a Warning Shot for $100B SaaS and IT Services
Most enterprise AI is a flashy demo with a consultant hiding behind the curtain. Ema just raised $77 million to prove the software can do the work — and that is bad news for anyone selling seats and billable hours.
Most enterprise AI is a flashy demo with a consultant hiding behind the curtain. Ema just raised US$77 million to prove the software can do the work — and that is bad news for anyone selling seats and billable hours.
On September 23, Ema announced a US$77 million Series B led by Creaegis, with Accel, Section 32 and Prosus increasing their investments. That brings the company’s total funding to US$140 million and, according to Ema, more than quadruples its valuation from its 2024 round.
The valuation is undisclosed. Fine. I care far more about what the company is selling, who is buying it and how it gets paid.
Ema is selling what it calls AI Employees: coordinated AI agents that work across HR, IT and finance systems. Not a chatbot that drafts a polite email and leaves the messy bit for Karen in operations. The pitch is that the agents complete multistep work across existing enterprise applications: handle a request, collect information, trigger approvals, update the systems and close the loop.
That is the real fight in AI now. It is not over who can generate the prettiest PowerPoint slide. It is over who gets paid when work actually gets done.
Ema is attacking the fattest part of enterprise spending
For two decades, enterprise software has had a lovely business model. Sell companies a stack of subscriptions. Charge per seat. Add implementation partners. Add consultants. Make every extra system and workflow feel unavoidable.
The customer gets more tools. The workers get more tabs. The consultants get more billable hours. Everyone tells themselves this is digital transformation.
Ema is making a different bet: that enterprises do not want another application to learn. They want the existing mess to produce an outcome.
The company was founded in 2023 by Surojit Chatterjee, previously at Google and Coinbase, and Souvik Sen, formerly of Okta. Its product coordinates multiple agents across a customer’s existing software rather than asking the customer to rip everything out on day one.
That detail matters. Incumbent enterprise systems are sticky because replacing them is painful, risky and politically exhausting. A startup that says, “Throw out your core HR, IT and finance stack,” will get a nice meeting and then a security questionnaire the length of War and Peace.
Ema’s wedge is smarter. Sit on top. Connect the systems. Prove value on a workflow. Expand from there.
The company says it has more than 50 active enterprise deals and more than 1 million active enterprise users. It names customers including NTT DATA, Hitachi, ADP, PwC, Google, KPMG, Wipro and Microsoft. It says revenue has grown 50-fold over two years and bookings have exceeded US$150 million.
Read that last number properly, though. Bookings are not annual recurring revenue. Ema told TechCrunch its figure includes the full value of multiyear contracts, including two- and three-year agreements. Plenty of founders blur those numbers because a big bookings figure looks fantastic on a slide. Do not let them.
Still, the commercial signals are serious. Ema says more than 90% of customers have expanded from their initial use case, while net dollar retention sits around 180%.
That is the metric I’d be staring at if I were an investor or a competitor. It suggests customers are not merely running a pilot for the board deck; they are giving Ema more work after the first deployment.
The interesting bit is not the chatbot. It is the pricing.
Ema says it charges for completed tasks and business outcomes, not seats or AI tokens.
Good. That is where this market has to go.
Seat-based pricing made sense when software was a tool used by a person. If I give 100 staff members access to a payroll platform, 100 seats is a reasonable proxy for value. It is blunt, but it works.
An AI agent breaks that logic. One agent can touch ten systems, perform thousands of repetitive actions and support workers across departments. Charging by seat becomes a leftover from the old world. Charging by token is even worse for most buyers: it turns the vendor’s underlying computing cost into the customer’s problem.
Outcome pricing sounds wonderful until you have to define an outcome. Was the IT ticket actually resolved? Was it resolved correctly? Did the agent save time, or did it create an error that someone cleaned up later? What happens when a workflow is half automated and a human does the difficult 20%?
That is where serious operators separate themselves from tourists.
If Ema can reliably price against completed, auditable work while maintaining the close-to-80% gross margin it reported, it will have built a far more dangerous model than a generic AI copilot. It means the company can take a slice of a budget currently spent on software subscriptions, integration projects and outsourced service work.
And unlike the old SaaS model, revenue can expand as the customer gives the system more jobs to do.
Why IT services firms should pay attention
The lazy take is that AI kills SaaS. I think that is too neat.
Large enterprise systems are not disappearing next Tuesday. They hold records, permissions, compliance history and decades of operational logic. In many cases they will remain systems of record for a long time.
The squeeze lands elsewhere first: the expensive layer of humans paid to move information between those systems.
Ema’s founder told TechCrunch that customers may eventually reduce dependence on some large SaaS products, which become more like databases underneath an agent layer. That is plausible. But the nearer-term threat is to the labour wrapped around the software: implementation, integration, support, routine administration and the endless manual handoffs between teams.
Ema says Wipro uses its employee assistant for more than 240,000 associates across 65 countries, automating more than 100 workflows and handling about 2.9 million employee queries annually. The company says response times fell from days to seconds and employee satisfaction rose 20%.
Those are company-reported numbers, so keep your head screwed on. But the direction is obvious. If an agent can resolve a meaningful share of millions of internal requests with limited human escalation, the economics of outsourced support and enterprise implementation change quickly.
The service firms are not necessarily dead. The good ones will become implementation partners, governance specialists and owners of difficult industry workflows. The bad ones will keep selling armies of junior people to do predictable coordination work and call it transformation.
That is not a strategy. That is a melting ice cube.
The contrarian take: Ema’s biggest threat may be success
Everyone assumes the danger to Ema is OpenAI, Anthropic, Microsoft or some other giant deciding to enter the same category.
Of course that risk is real. Ema operates in the most crowded, well-funded patch of technology on earth. The frontier model companies are pushing deeper into enterprise deployments, while incumbent software vendors have distribution, data access and long-standing procurement relationships.
But Ema’s more immediate problem may be that it proves the category works.
Once a large company sees agents reliably handling HR requests, IT tickets and finance workflows, every existing vendor will repackage its own product around outcomes. Every major consulting firm will claim it has an agentic transformation practice. Every procurement team will demand lower pricing because they now have a benchmark.
That is why the US$77 million matters. This round is not just fuel for engineering. Ema says much of it will go into go-to-market expansion. That is sensible. In enterprise software, being technically right is not enough. You need security approvals, integrations, references, salespeople who can navigate a six-month buying process and customer-success teams that make sure the pilot turns into a rollout.
Founders love product. Markets often reward distribution more.
What this means for you
If you run a business, stop asking your team where AI can “help.” That question produces a graveyard of chatbots and workshop notes.
Instead, pick one recurring workflow with four features: it happens at volume, it crosses more than one system, it has a measurable end state and it currently consumes real human hours. Employee onboarding, supplier setup, IT access requests, expense exceptions and customer-service triage are better starting points than vague ambitions to “use AI in operations.”
Then measure the before and after like an adult. Track completion time, error rate, human escalations, customer or employee satisfaction and fully loaded cost per completed task. If you cannot measure the outcome, do not sign an outcome-priced contract.
If you are a founder, do not build another AI wrapper with a clever prompt and a subscription page. Find a workflow where the buyer already spends money on people, software and delays. Sell the result, not the model.
And if you invest, be wary of businesses claiming massive AI revenue without explaining whether it is recurring software revenue, services revenue or multiyear bookings dressed up for the cameras.
Ema’s US$77 million round is a useful signal because it points to where the money is heading: away from software that merely records work, and toward systems that can complete it.
That will create winners. It will also expose a lot of businesses whose margins depended on work being slower, more fragmented and more manual than it ever needed to be.
Good. That is called progress.