A critical assessment of Ema’s $77M Series B (led by Creaegis) to put autonomous AI “employees” to work across enterprise HR, IT, and finance — a story built almost entirely on numbers only Ema can see, sitting on a stack Ema doesn’t own.
Ema claims ~80% gross margins while routing every task across 100+ frontier models it buys from OpenAI, Anthropic, Google, and Azure. Those suppliers set its cost floor — and are building the same “AI employee” natively. The dependency is never disclosed in the raise.
The pitch is that AI is “eating into enterprise software.” But Ema only works by executing inside Workday, ServiceNow, SAP, and ADP. It is an overlay that needs the SaaS stack to keep existing and exposing APIs — the opposite of replacing it.
1M+ users, 180% NDR, $150M bookings, 50x growth — every headline figure traces to one press release. No audited financials, no third-party estimate, and several marquee “customers” (Hitachi, Wipro, Prosus) are also investors.
Key Finding: Ema has a genuinely strong team riding one of the largest waves in software, and real commercial engagement clearly exists. But the investment case rests on unverifiable company metrics and a structurally thin moat — an orchestration layer wedged between the model vendors above it and the systems-of-record below it, in a niche where its three closest peers (Moveworks, Sana, Aisera) were all acquired rather than allowed to win.
Ema’s value is real, but it is a coordination value, not a model or a system of record. Every layer it depends on is owned by someone larger.
OpenAI, Anthropic, Google, Azure. Ema buys inference here. Its suppliers set its COGS — and ship competing agents.
Routes each task across 100+ models. The genuine IP — but a routing layer, replicable in principle.
Coordinates multi-step “AI employees.” No-code builder plus governance and audit trails.
Workday, ServiceNow, SAP, ADP. Ema acts inside these — and depends on their APIs staying open.
Ema’s differentiation is real but it is orchestration and governance, not deep technology. The founder frames model progress as pure tailwind — “Progress in frontier models is actually very beneficial to us.” That is also the tell: a business whose product improves only because its suppliers improve has handed the value creation to those suppliers.
Ema claims roughly 80% gross margins while every action it takes runs on models it neither owns nor prices. If frontier-model API pricing rises, or a supplier throttles a competitor, the margin claim compresses from the bottom up. Ema discloses none of this dependency in its funding materials, and never addresses what happens to unit economics when the vendors above it change terms.
CEO — ex-Coinbase CPO, ex-Google product leader. The strongest asset in the story; deep enterprise credibility.
Bengaluru growth fund leading the round. CIO Prakash Parthasarathy calls Ema “the default execution backbone” — a talking point, not independent diligence.
Multi-model router across 100+ LLMs. Independently described — confirms Ema is an orchestration layer, not a model builder.
Hitachi, Wipro, and Prosus are named as both backers and marquee customers — a conflict that inflates logo credibility.
$150M+ is multi-year bookings over 2–3 year contracts. Implied run-rate ARR is far smaller — roughly $50–75M (derived).
SOC 2, ISO 42001, EU AI Act compliance are marketed heavily — but no third-party agent accuracy or error-rate data is published.
Ema’s exact niche — autonomous employee support across IT, HR, and finance — is the one where the nearest comparables were absorbed by incumbents in 2025, and where the leaders are capitalized an order of magnitude beyond Ema.
Sierra (Bret Taylor) raised $950M at a $15B valuation on ~$200M ARR; Cognition sits near $26B; Glean at $7.2B; Decagon at $4.3B. Ema’s $77M / ~4x round is modest in a field where the leaders raise more in a single tranche than Ema has raised in its life.
Moveworks — Ema’s closest direct comp — was bought by ServiceNow for ~$2.4B. Sana went to Workday; Aisera was acquired; Adept was acqui-hired by Amazon. The base-rate outcome for the “AI employee” niche has been a roll-up into a platform incumbent, not standalone escape velocity.
The disintermediation problem is structural: ServiceNow, Workday, Microsoft, Google, and OpenAI can build the “AI employee” natively inside the systems where the data already lives. Ema’s suppliers and its integration targets are its most dangerous competitors — and both have already shown they will build or buy this capability rather than route it through a third party.
Seven structural risks the $77M Series B does not resolve.
Ema’s ~80% margin and its entire product rest on third-party frontier models it doesn’t own or price. Its suppliers are also its competitors, and the dependency is undisclosed in the raise.
ServiceNow (bought Moveworks), Workday (bought Sana), Microsoft, Google, and OpenAI can ship native “AI employees” where the enterprise data and system-of-record already sit.
Every headline figure — 1M users, 180% NDR, $150M bookings, 80% margin, 50x growth — is company-sourced with zero audited or third-party corroboration.
Google, Microsoft, PwC, and KPMG are named only by Ema; several “customers” (Hitachi, Wipro, Prosus) are also investors, and deployment depth is undisclosed.
Autonomous action-taking in HR, finance, and IT — payroll, hiring, access provisioning, financial controls — with no published error-rate data turns a hallucinated action into a liability event.
Disclosed prior rounds (~$25M + a Series A grown to $50M) plus $77M imply ~$152M, yet Ema states ~$140M total. The split is reported inconsistently and the numbers don’t cleanly add up.
A ~180% net dollar retention is Snowflake-tier — but Ema is only ~2 years old. Early expansion off a small denominator produces flattering NDR that regresses as the base grows; this is more likely a peak than a durable cohort metric.
Ema is a strong team, on the right wave, telling a story only it can verify. The $77M Series B is a reasonable bet on enterprise agentic automation and a credentialed founder. But the two questions that decide the outcome — can it hold ~80% margins on inference it rents from its own competitors, and are the headline metrics anything an outside party can confirm — are exactly the two the raise leaves unanswered.
Based entirely on publicly available information, including the TechCrunch announcement of September 23, 2026 and Ema’s own press release. Every operating metric in this report is company-sourced unless otherwise labeled; none could be independently audited.
