Ema Series B Funding Raises the Stakes for Enterprise Software and IT Services

Ema has raised $77 million in Series B funding to move its AI agents deeper into corporate HR, IT, and finance operations. The round gives the startup more capital to pursue a direct challenge to enterprise software vendors and IT services firms.

The Ema Series B funding was led by Creaegis. Existing investors Accel, Section 32, and Prosus increased their stakes, bringing Ema's total funding to $140 million. The company says its valuation has more than quadrupled since its previous round in 2024, although it declined to disclose the new figure.

The important story is not the size of the round alone. Ema wants its agents to complete work across existing applications, reducing the need for employees to navigate each system manually. If that model holds up, software applications risk becoming databases beneath an AI-controlled workflow layer.

That creates pressure on two established business models. Software vendors sell access to applications through licenses and seats. IT services firms earn revenue by implementing, integrating, and operating those applications. Ema is trying to absorb parts of both businesses into a system priced around completed tasks and business outcomes.

Ema Series B Funding Turns a Product Bet Into a Sales Push

The new capital moves Ema from proving its product toward selling an alternative operating model for enterprise work.

The financing announced on September 23 consists entirely of primary equity, according to the company's account of the Ema funding round. That means the proceeds go into the business, rather than paying early shareholders or refinancing debt.

Ema plans to direct much of the money toward sales, marketing, and international expansion. The Mountain View company has concentrated primarily on customers in the United States and Europe. It now plans to pursue organizations across Asia-Pacific, South America, and parts of the Middle East.

That shift matters because Ema spent its early years developing the platform and establishing deployments. It now employs nearly 200 people, with offices in Bengaluru, London, and Vancouver as well as its California headquarters. A wider commercial organization gives the company more capacity to compete for large, slow-moving enterprise contracts.

Ema was founded in 2023 by CEO Surojit Chatterjee and engineering leader Souvik Sen. Chatterjee previously held senior roles at Google and Coinbase, while Sen worked at Google and Okta. The company emerged from stealth in 2024 with an ambition to build a universal AI employee.

Its original product combined a workflow engine with EmaFusion, a system designed to select among different language models. The strategy was visible in Ema's 2024 launch, when the company positioned model choice as one component of a broader enterprise automation platform.

The latest round advances that thesis. Ema says its platform can now draw from more than 150 frontier and open-source models. It combines them with integrations, domain knowledge, permission controls, and agent coordination.

The company calls the resulting systems "AI employees." In practical terms, Ema AI agents are software processes that plan and execute multistep work across several business applications. They can retrieve records, update systems, check results, request approvals, and route unusual cases to people.

This differs from a chatbot that answers one question at a time. An employee asking about parental leave might need information from a policy repository, eligibility data from an HR system, and an approval workflow. Ema's pitch is that one agent experience can coordinate those steps without making the employee visit each application.

The company has also launched purpose-built products for HR, IT, and finance. That packaging narrows the distance between a general agent platform and a deployable business application. It gives buyers predefined workflows while retaining a common orchestration layer across departments.

Ema says it will continue investing in that platform alongside its commercial expansion. The balance matters. Hiring salespeople can increase contract volume, but enterprise customers will judge the system through reliability, security, governance, and measurable operational results.

The Ema Series B funding therefore marks a transition rather than a victory. Investors are backing the company to repeat its early deployments across more organizations and regions. The next stage requires converting reported traction into durable production use.

Why AI Agents Are Starting to Pressure Software Seats

Ema's central bet is that enterprises will pay for completed work instead of paying primarily for access to separate applications.

Traditional enterprise software organizes work around systems of record. A company buys one application for human resources, another for service management, another for finance, and additional products for collaboration, analytics, and approvals.

Employees must learn each interface and understand how information moves between them. Companies then pay consultants, systems integrators, and internal teams to connect the applications. The arrangement creates recurring revenue for software vendors and continuing implementation work for services firms.

Ema enterprise software does not immediately remove those underlying systems. It first sits above them, using integrations to read information and perform authorized actions. The agent becomes the interface, while established applications continue storing official records.

Chatterjee argues that this arrangement can eventually reduce dependence on large software suites. He told TechCrunch that some customers are moving toward replacing certain SaaS applications because those products are becoming "like a database."

That claim describes the industry's potential reversal. Enterprise applications once controlled both the data and the user experience. AI agents can separate those layers, allowing workers to request an outcome while the agent decides which systems to use.

The consequences extend beyond interface design. Seat-based pricing assumes that users derive value by logging into an application. When an agent performs the work for many employees, the connection between seats, usage, and value becomes weaker.

Ema says it does not charge based on software seats or consumed AI tokens. Its pricing is tied to task completion and business outcomes. That model attempts to align spending with work performed, although customers still need clear definitions for successful outcomes, exceptions, and disputed results.

The shift also changes how buyers compare vendors. A conventional procurement process might evaluate HR software against other HR suites. An agent platform asks whether a company needs another full application, or whether an orchestration layer can complete the required workflow using existing data.

This is why Ema's challenge reaches beyond one software category. The platform spans employee support, IT service management, finance operations, and other internal processes. Success in one department can provide a route into adjacent workflows.

Ema says more than 90% of its customers have expanded beyond their initial use case. Some reportedly use the platform across dozens of workflows. The company also reports net dollar retention near 180%, meaning existing customers increased their spending substantially over time.

Those figures remain company-reported, and Ema has not disclosed its annualized revenue run rate. Still, the expansion pattern matters more than a single demonstration. Enterprise agents become strategically significant when a buyer reuses the same platform across departments.

The timing also reflects improvements outside Ema. Frontier models have become better at reasoning, tool use, and maintaining context across multistep assignments. Cloud providers and software vendors have expanded connectors and permission frameworks around those models.

At the same time, major AI developers are moving closer to enterprise workflows. Anthropic has introduced agents and integrations for tasks such as financial research, compliance, and month-end reporting. Its finance agent templates show how model providers can package domain-specific work themselves.

OpenAI has gone further into deployment services. Its deployment company embeds engineers in customer organizations and brings implementation expertise alongside its models. That approach competes for spending that previously flowed mainly to consulting firms and systems integrators.

Chatterjee portrays these model companies as suppliers rather than direct competitors. Because Ema can route work across many models, improvements at AI laboratories can increase its platform's capabilities. However, those laboratories can also build their own agent products, integrations, and services.

The competitive boundary is therefore fluid. Ema benefits when models improve, but it must keep enough value in orchestration, governance, domain knowledge, and workflow execution. Otherwise, model providers or established software companies can absorb the same functions.

The Main Contest Is Agents Against the Services-Heavy Software Stack

Ema is competing with a combined software and services model, not simply with another AI-agent startup.

Large enterprises rarely buy software as a standalone product. They also pay for configuration, data migration, custom integrations, compliance work, employee training, and ongoing support. Those services can continue long after the original application contract is signed.

Ema's proposition compresses parts of that stack. Its agents are designed to work across existing systems, learn from deployments, and complete recurring processes with less human intervention. The company argues that this can reduce both application dependence and services-heavy implementation work.

The clearest example is IT support. A conventional service desk might receive a ticket, classify it, search documentation, check the user's permissions, update a system, and escalate the case. An agent that handles those steps can reduce the number of routine tickets reaching people.

Ema says one large global systems integrator uses its employee assistant for more than 240,000 associates across 65 countries. According to the company's deployment figures, the system automates over 100 workflows and handles about 2.9 million queries annually.

The company says that deployment shortened response times, improved employee satisfaction, avoided about 60% of tickets, and allowed the people operations team to operate with fewer staff. These are Ema's measurements, not independently audited performance figures.

Another reported deployment handles more than 1 million IT service management tickets annually for a large business-process outsourcing provider. Ema says the associated voice system processes more than 1 million calls across over 15 languages, with fewer than 10% escalated to people.

A third customer, described as a global conglomerate, reportedly moved from concept to production within four weeks. Ema says the deployment connected with over 20 systems of record and serves more than 40,000 employees.

These examples illustrate why services providers face a complicated choice. They can treat agents as competitors that automate billable labor. They can also use agent platforms to serve customers more efficiently, protect accounts, and create new implementation offerings.

Chatterjee says many services companies are already working with Ema while changing their own business models. That cooperation makes sense because enterprises still need process design, governance, integration, and organizational change. AI does not eliminate those requirements.

The pressure falls most heavily on repeatable work billed through large teams. If an agent can resolve common requests, reconcile routine records, or perform standard checks, customers will question contracts based primarily on labor volume.

Yet the services relationship may persist in a different form. Agents require access controls, evaluation systems, escalation rules, and monitoring. Regulated organizations also need evidence showing which data an agent accessed and why it took a particular action.

Some model companies explicitly recognize that reality. Anthropic, for example, has partnered with a major IT provider to embed Claude inside systems used by regulated industries. The DXC alliance includes plans to train forward-deployed engineers rather than remove services work entirely.

Ema must therefore prove that its product-led approach reduces dependence on custom services without merely shifting those services elsewhere. A platform requiring extensive manual configuration for every customer would weaken its economic argument.

The reported gross margin offers one early signal. Ema says its margin is close to 80%, despite completing work commonly associated with labor-intensive services. Chatterjee attributes that level partly to decreasing human support as the system learns from deployments.

Investors appear to believe the platform can retain software-like economics while addressing services-like workloads. Creaegis led the round, while every major existing investor participated with a larger commitment.

That does not settle the contest. Enterprise contracts can hide meaningful deployment costs, customer-specific engineering, and cloud inference expenses. A high stated gross margin must remain stable as the company enters more regions, industries, and complex workflows.

Ema also faces well-funded agent companies. Sierra focuses heavily on customer-facing work and says its agents handle billions of interactions. A reported Sierra funding round valued that company above $15 billion.

Other competitors specialize in customer support, voice systems, departmental workflows, or agent infrastructure. Established vendors such as Microsoft, Salesforce, ServiceNow, Workday, and SAP can place agents inside products customers already trust.

Ema's advantage must come from working across those boundaries. If each incumbent's agent remains strongest inside its own suite, buyers may prefer an independent orchestration layer. If incumbents cooperate effectively across systems, Ema's position becomes harder to defend.

Ema AI Agents Depend on Orchestration, Not One Model

Ema's technical thesis is that dependable enterprise automation requires coordination around language models, not allegiance to a single model provider.

A language model can interpret an employee's request, but enterprise execution requires additional components. The system must identify the user, retrieve authorized data, choose tools, preserve context, verify results, and recognize when a human must intervene.

EmaFusion serves as the routing layer in Ema's architecture. The company says it can select among more than 150 frontier and open-source models. Different models can be assigned according to task requirements, performance, data policies, or operating costs.

This model-flexible design reduces dependence on one supplier. It also lets Ema benefit when a provider releases a stronger model. Chatterjee has described progress at frontier laboratories as beneficial to the company for that reason.

Model access alone offers limited protection, however. Competitors can often license the same foundation models. Ema's defensible value must sit in its integrations, workflow logic, enterprise controls, evaluation data, and knowledge of how processes behave in production.

Consider an HR request involving a leave policy. The language model might understand the question, but accurate execution depends on current policy documents, location-specific rules, employment records, and approval authority. Each source can change independently.

A reliable agent must distinguish between answering a question and modifying an official record. It must know when consent is required, when a manager must approve an action, and when conflicting information makes automation unsafe.

The same principle applies to IT. Resetting access, provisioning software, or changing a security group can affect sensitive systems. The agent needs constrained permissions and a clear audit trail. A fluent response cannot substitute for correct execution.

Finance workflows raise the stakes further. Reconciliations, purchase approvals, and account changes must follow formal controls. An agent can prepare or execute steps, but companies need separation of duties and human oversight for critical decisions.

Ema says its systems can check their own work and route approvals when required. Those controls are central to the product's credibility. They are also difficult to evaluate from customer counts or aggregate query volume.

The platform's broader promise resembles an intelligent control plane. Instead of replacing every system immediately, it coordinates the existing environment through a common interface. Over time, rarely used applications may lose strategic importance if the agent owns the interaction.

That mechanism explains why Ema enterprise software threatens incumbents even before customers cancel applications. The agent can weaken the direct relationship between users and individual products. Vendors become infrastructure suppliers underneath another company's interface.

The same mechanism supports cross-department expansion. A common agent layer can combine information from IT, HR, and finance when one process crosses organizational boundaries. Employee onboarding, for example, can involve identity creation, equipment requests, payroll records, policy acknowledgments, and manager approvals.

This is also where personal and organizational knowledge become important. An agent needs more than transaction records. It may require policies, project history, documents, and context explaining how a team actually works.

For individual knowledge workers, a personal knowledge base addresses a related problem at a smaller scale. It makes scattered information easier to retrieve and apply without replacing the systems where that information originated.

Ema's challenge is applying a comparable context layer across enterprise operations while preserving permissions and accountability. The more systems an agent touches, the more valuable coordination becomes. The potential impact of a mistake also increases.

That tradeoff keeps humans inside the system. Ema's architecture includes human oversight for critical decisions, according to the company. The strongest deployments will likely automate routine paths while directing ambiguous, sensitive, or high-impact cases to qualified employees.

What Ema's Growth Numbers Still Do Not Prove

Ema has reported unusually strong expansion metrics, but the company has not disclosed enough financial detail to verify the durability of its growth.

Ema says it has more than 50 active enterprise deals and over 1 million active enterprise users. The platform has reportedly handled more than 5 million actions and queries across customers including Hitachi, ADP, PwC, Google, KPMG, Wipro, Microsoft, and NTT DATA.

The company also says revenue grew 50-fold during the past two years. Revenue bookings have exceeded $150 million, according to Chatterjee, while net dollar retention is around 180%.

Each number requires context. Fiftyfold growth can begin from a small base. Ema did not disclose the starting revenue or its current annualized revenue run rate, making the absolute scale difficult to assess.

Bookings are not the same as recognized revenue. Ema's figure includes the total value of multiyear agreements, including contracts lasting two or three years. Some of that value will be recognized later and may depend on contract terms or successful delivery.

Net dollar retention near 180% would indicate major expansion among existing customers. However, readers do not know the size of the measured cohort, the calculation period, or how outcome-based pricing affects the result.

Active users and completed actions also measure different things. A company can make an agent available to a large workforce while only a subset uses it regularly. Query volume does not reveal task complexity, accuracy, or how many requests required human correction.

Ema's named customers provide meaningful validation, but individual deployment scope varies. A limited departmental workflow at a major corporation differs from an enterprise-wide replacement of established software.

The language of replacement therefore deserves caution. Ema has evidence that customers are expanding deployments and automating meaningful workloads. It has not shown that large numbers of organizations have removed their core SaaS applications.

The near-term outcome may be addition rather than substitution. Enterprises might continue paying existing software vendors while adding an agent platform above them. That can improve the employee experience but increase total technology spending.

Outcome-based pricing also creates measurement questions. The customer and vendor must agree on what counts as completed work. A resolved support ticket is not valuable if the resolution is wrong, generates a later incident, or frustrates the employee.

Agents can produce hidden costs through monitoring, exception handling, model usage, security reviews, and integration maintenance. These costs may remain manageable, but buyers need complete operational accounting rather than a narrow automation metric.

Security presents another uncertainty. An agent capable of acting across HR, IT, and finance receives broad access to sensitive information and business systems. Compromised credentials, faulty instructions, or manipulated input can spread across connected workflows.

Governance becomes more difficult when the platform uses many underlying models. Enterprises need to understand where data travels, which model handled a request, what rules governed the choice, and how the result was evaluated.

International expansion adds regulatory and localization requirements. Employment rules, privacy obligations, language performance, and data residency vary across Ema's target markets. A workflow that operates well in one jurisdiction may require substantial changes elsewhere.

Competitive pressure will also rise. Model providers can offer agent templates and deployment teams. Software incumbents can bundle agents into existing contracts. Services firms can develop their own orchestration products or become implementation partners for multiple platforms.

Ema's model-neutral architecture helps, but neutrality alone is not a durable advantage. Buyers will expect deeper integrations, stronger evaluations, and domain-specific controls. They will also compare Ema's outcomes with agents embedded in software they already own.

The Ema Series B funding gives the company time and resources to answer these questions. It does not remove them. The most important evidence will come from repeated production deployments, audited outcomes, renewal behavior, and stable economics.

Three Signals Will Show Whether Ema Can Reshape Enterprise Spending

The next test is whether Ema can turn customer expansion into measurable reductions in software and services dependence.

The first signal is application retirement. Ema says some customers are moving toward replacing large SaaS products, but it has not identified a broad set of completed replacements. Specific examples of canceled applications or reduced license commitments would strengthen its central claim.

Buyers should distinguish between fewer employee logins and lower software spending. An agent can hide an application while the company continues paying for the underlying system. True displacement requires a contract change, a smaller deployment, or a retired product.

The second signal is expansion quality. Ema reports that more than 90% of customers have moved beyond their first use case. The stronger evidence will be renewals that preserve high net retention without large custom engineering teams or declining margins.

Watch whether gross margin remains near the level Ema reports as deployments become more complex. Stable margins would support the claim that software can absorb services work efficiently. Falling margins would suggest that enterprise automation still requires substantial human delivery.

The third signal is the response from incumbents and partners. Software vendors can expose more functions to outside agents, restrict platform access, improve their own assistants, or bundle automation into existing agreements. Services firms can resist labor displacement or use agents to redesign their offerings.

Model developers also matter. OpenAI and Anthropic are building enterprise products, deployment organizations, and partnerships with consultants. Their actions can improve the technology Ema uses while simultaneously narrowing the space available to independent orchestration platforms.

Over the next several months, customer disclosures will be more informative than another funding announcement. Named deployments should explain which workflows run autonomously, what humans still handle, how success is measured, and whether older software or services contracts changed.

Enterprise buyers should examine the same evidence before treating AI employees as replacements. They need task-level accuracy, escalation rates, security controls, total operating costs, and contract accountability. A polished interface is not enough when software can modify financial, personnel, or access records.

Knowledge workers should also pay attention because the interface to corporate software is changing. Employees may spend less time navigating applications and more time defining outcomes, reviewing exceptions, and supervising automated work.

That shift can reduce repetitive administration, but it also moves influence toward whichever platform controls context and execution. Organizations must decide who governs that layer, which records remain authoritative, and how employees can challenge an agent's decision.

The Ema Series B funding makes this contest harder to dismiss as a product experiment. The company now has $77 million in new capital to sell a direct alternative to services-heavy enterprise operations.

The question for buyers is concrete: can Ema AI agents remove applications, implementation labor, and operating costs without weakening control? Track actual retirements, renewal economics, and verified production outcomes. Those signals will show whether Ema is changing enterprise spending or simply adding another layer to it.