What it is
Ema sells prebuilt “AI Employees”: agent systems that run a back-office process end to end inside the applications you already own. An HR AI Employee answers a policy question, files the leave request in Workday, and routes the approval. An IT one resolves a ticket in ServiceNow. Finance and customer-support roles work the same way, and the support personas run over chat, voice, and email. Each AI Employee is a set of cooperating agents built on a no-code builder. A routing layer called EmaFusion picks a model per task from a pool of 100+ public and private models (TechCrunch puts it at 150+). Ema lists 250+ prebuilt integrations and 1,000+ actions, and users reach the agents through Slack, Microsoft Teams, and voice.
The closest category leader is ServiceNow’s employee-service agents. The difference is that Ema sits on top of whatever HRIS, ITSM, and ERP you run instead of asking you to consolidate onto one platform first.
Ema was founded in 2023 by Surojit Chatterjee (formerly at Google and Coinbase’s chief product officer) and Souvik Sen (formerly at Okta), and came out of stealth in March 2024. It is headquartered in Mountain View with offices in Bengaluru, London, and Vancouver, and has close to 200 staff. On 2026-09-23 it announced a $77M Series B led by Creaegis, with Accel, S32, and Prosus adding to their stakes, for $140M raised in total. It did not disclose the valuation.
Why it shows up in back-office ops stacks
The use case is a company whose HR or IT queue is spread across several systems. Ema’s flagship customer story is Hitachi Digital, the shared HR services arm for 40,000+ employees across five business units, each on its own HR system: ServiceNow, SAP, Workday, and ADP. Ema reports that the agent went live in under eight weeks and that HR ticket volume fell 30% month over month. Those figures are vendor-reported. Other named customers include NTT DATA, ADP, PwC, KPMG, and Wipro.
Two things set it apart in procurement. First, it supports on-premises and air-gapped deployment, and lists SOC 2 Type II, ISO 27001, ISO 27017, ISO 27701, ISO 42001, and CSA STAR. Second, it has been on the Microsoft Marketplace since September 2025, so a Microsoft customer can pay with committed Azure spend instead of opening a new vendor budget line.
Pricing reality
Ema publishes no prices. Its pricing page is a demo form, and there is no self-serve tier. The CEO told TechCrunch that Ema does not charge per seat or per token. Contracts are tied to completed tasks and business outcomes.
The one public signal of deal size comes from the same interview: more than 50 active enterprise deals and more than $150M in bookings, counted as total multiyear contract value. That works out to an average of up to about $3M per customer over the life of a contract. Treat Ema as a six- or seven-figure annual commitment, not a team tool. Net dollar retention is around 180%, and more than 90% of customers expand past their first use case. Budget for function two and three now, because the vendor plans on it.
Best for
HR shared-services and IT service-desk leaders at enterprises of roughly 2,000+ employees whose employee requests span several HRIS, ITSM, and ERP systems. Ema fits teams that want a vendor-built agent working across all of them in weeks, and have no in-house team to build the same thing.
Not for: companies under about 250 employees, teams with one queue in one system, or anyone who needs a price before a sales call. A 50-person ops team gets the same “answer and file the request” agent from n8n or Lindy for tens of dollars a month.
Versus the alternatives
- ServiceNow — the market-share leader in employee service. Its EmployeeWorks front door is built from Moveworks, which ServiceNow bought for $2.85B and closed on 2025-12-15. Pick ServiceNow when it is already your system of record for IT and HR cases. Pick Ema when your cases are split across several systems and you will not consolidate them.
- Microsoft Copilot Studio — the build-it-yourself default inside a Microsoft tenant, at $200 a month per 25,000 credits. Pick it when you have Power Platform builders and want to own the agents. Pick Ema when you want to buy finished agents and pay per outcome instead of per step.
- Serval — the fastest-growing entrant. It reached a $1B valuation on a Sequoia-led $75M Series B in December 2025 and has expanded from IT into HR, legal, and finance. Pick it for a cloud-native company of 200 to 2,000 employees that is replacing its helpdesk.
- Relevance AI, Dust, or n8n — pick these when you have ops engineers and want usage-priced agents you control. You give up Ema’s prebuilt HR and IT process logic and its on-premises option.
Watch-outs
- “Outcome” is whatever the contract says it is. Outcome pricing only protects you if a completed task is defined tightly. Guard: in the order form, define a billable outcome (for example, closed with no reopen within 7 days), name your own ticketing system as the source of the count, and set an annual cap.
- Every performance number is vendor-reported. The Hitachi results come from Ema’s own case study, and we found no independent reviews on G2 or Capterra. Guard: measure your baseline resolution time and ticket volume in your own system for 30 days, run a pilot on one queue against it, and ask for a reference customer in the same function and size band.
- Agents that write to Workday or SAP act on live records. A wrong leave balance or access grant spreads at machine speed. Guard: give Ema scoped service accounts with least privilege, keep the platform’s human approval step on every write action until accuracy holds on a sample, and require exportable audit logs in the contract.
- A young vendor sits between your staff and your systems of record. Ema is three years old with about 200 people. Guard: put workflow-definition export and a data-return clause in the master agreement, so you can move the logic if the vendor is acquired or changes direction.