An AI agent startup just raised $77 million on the pitch that it replaces buying more software. Here's the back-office test before a GC believes it
Ema raised a $77 million Series B saying its AI agents already handle more than a million IT tickets and calls a year at up to 95% accuracy — but those numbers come from clean digital tickets, not the scanned COIs and PDF submittals a construction back office runs on.
Ema, a startup that builds "AI employees" to automate back-office work across a company's existing software, raised a $77 million Series B this week — funding that matters less for the check size than for what it's betting on: that companies will stop buying a new SaaS tool for every workflow gap and instead pay for an AI layer that works across the systems they already have. For a GC or sub whose software stack has grown by accretion — one more login for every new problem — that's a bet worth understanding before the next renewal comes up.
What is Ema, and why does the funding matter?
The round, led by Creaegis with Accel, Section 32, and Prosus increasing their existing stakes, brings Ema's total funding to $140 million and, the company says, more than quadruples its valuation from its last round (Ema didn't disclose the new number). More telling than the money are the adoption figures Ema is citing: over 50 enterprise deals, more than 1 million active users, roughly 180% net dollar retention, and 50-fold revenue growth in two years. Those are the numbers of a product that customers are expanding, not just piloting.
What does an "AI employee" actually do?
Ema's agents aren't a new app with its own login and its own data silo — they're built to plug into a company's existing HR, IT, and finance systems and carry out multi-step work across them: triaging a ticket, pulling data from three systems, resolving the request, and routing anything it can't handle to a person. The company says its agents already process more than 1 million IT service tickets a year and handle more than 1 million calls in 15-plus languages, with up to 95% accuracy and under 10% of interactions escalated to a human.
Point solution or orchestration layer — what's the actual choice?
| Buy another point solution | Deploy an orchestration-layer agent | |
|---|---|---|
| Fixes one workflow gap | Yes, well | Yes, if scoped right |
| Requires a new login/system | Yes | No — works inside existing systems |
| Integration work | One-off per tool | Ongoing, across many systems |
| Vendor stack over time | Grows with each new need | Stays flat in theory |
| Accuracy on your actual documents | Known once live | Unproven until piloted |
Do Ema's numbers translate to a construction back office?
Not automatically, and this is the honest caveat. Ema's 95%-accuracy, sub-10%-escalation figures come from IT tickets and support calls — inputs that arrive as structured, digital, well-formatted requests. A construction back office runs on messier material: a scanned certificate of insurance with a signature over the expiration date, a subcontractor invoice that doesn't match the line items on the schedule of values, an RFI status that lives half in Procore and half in an email thread. None of that is a knock on Ema specifically — it's the same gap every horizontal AI tool has to close before it earns trust on construction paperwork. The accuracy number that matters is the one you measure on your own documents, not the one in the press release.
What should a GC or sub's ops team actually do?
Before your next software purchase closes a workflow gap — chasing expiring COIs, reconciling vendor invoices, following up on stalled submittals — ask whether an agent layer wired into what you already run (your ERP, your project management platform, your email) could close it instead of adding one more tool and one more login. Then don't take anyone's marketing accuracy rate on faith: run a two-week pilot against your own real, messy paperwork and count the escalations yourself. If it holds up on your documents, it's a real alternative to the next SaaS renewal. If it doesn't, you've lost two weeks, not a contract.
This is the same procurement question we raised when xAI's own AI agent found $100,000 in vendor waste inside its internal buying process — an agent doing the cross-system work a person or a point tool used to do. That one was about procurement; this one is about whether "buy an agent instead of another app" holds up once the inputs get as messy as construction paperwork actually is.
Forward this to whoever on your team just requested budget for another back-office software seat.
Friday one chart. Every week, one piece of data that should change a decision on your project. Subscribe at constructionaibrief.com.
- What did Ema just announce?
- On September 23, 2026, Ema announced a $77 million Series B led by Creaegis, with existing investors Accel, Section 32, and Prosus increasing their stakes. That brings Ema's total funding to $140 million, and the company says its valuation more than quadrupled from its last round (it declined to give the new figure). Ema reports 50+ enterprise deals, more than 1 million active users, roughly 180% net dollar retention, and 50-fold revenue growth over two years.
- What does an 'AI employee' do differently from ordinary software?
- Instead of being a single-purpose app a company logs into, Ema's AI agents are built to plug into a company's existing software systems and carry out multi-step work across them — routing approvals, triaging tickets, resolving requests end to end. The company says its agents handle more than 1 million IT service tickets a year and more than 1 million calls across 15-plus languages, with up to 95% accuracy and fewer than 10% of interactions escalated to a human.
- Should a contractor expect that same accuracy rate on its own back-office work?
- Not without testing it first. Those figures come from IT-ticket and call-center environments with structured, digital inputs — not from scanned certificates of insurance, PDF submittals, or a handwritten packing slip, which is what most construction back-office paperwork actually looks like. A pilot run on a contractor's own documents is the only way to know if that accuracy holds up.
- Does this mean AI agents will replace tools like Procore or Sage?
- Not directly. Ema's pitch is to sit alongside and orchestrate a company's existing systems of record, not rip them out. But the funding reflects a broader bet that companies will pay for an agent layer that automates work across tools they already own instead of buying a new point solution for every workflow gap — worth watching if your firm's software vendor list only ever grows.
- What should a GC or sub's back office actually do with this news?
- Before signing for another point-solution SaaS tool to close a workflow gap — subcontractor insurance-certificate tracking, invoice-to-schedule-of-values reconciliation, chasing RFI or submittal status across email and a project management platform — ask whether an orchestration-layer agent wired into the systems already in place could do it, and demand a pilot on real, messy documents before trusting any vendor's headline accuracy number.