Nivoda, a B2B diamond marketplace, just cut its monthly close time from eight days to under 24 hours. Journal postings that used to take days now finish in minutes. 95% of reconciliations run on autopilot. If finance automation could deliver results like that, why did it arrive after legal AI and customer support bots—despite being the most labor-intensive function in enterprise operations?
Because CFOs would rather pay for 100,000 hours of manual labor than trust an AI agent with decisions regulators audit.
That calculus just shifted. Stacks closed a $23M Series A on February 19-20, 2026—one week ago—led by Lightspeed, proving the pain finally exceeds the paranoia. But only for companies desperate enough to rip out their entire data infrastructure first.
Finance teams are burning 100,000 hours annually because CFOs won’t admit the real problem
The labor crisis isn’t a secret. It’s a design choice. Stacks’ customers are saving 100,000+ hours annually across 30+ enterprises in under 12 months—which means those hours were being spent before. On what? Reconciliations. Journal entries. Month-end close cycles that stretch eight days because every number gets checked twice by humans who don’t trust the ERP output.
Admitting that means admitting the entire close process is broken by design.
Nivola’s transformation proves the technology works. But only 30 enterprises adopted it in a year. Why? Because CFOs sign off on $20M+ month-end closes, and betting that on an AI agent—even one backed by Lightspeed’s $23M—is career-defining risk. These are the same high-skill jobs facing AI pressure across other functions, except their decisions get audited by regulators. And unlike shadow AI adoption patterns in marketing or sales, finance teams can’t quietly experiment. Every reconciliation leaves a trail.
The $100B market incumbents built their moats on customer misery
BlackLine, HighRadius, and OneStream dominate the Office of the CFO software market—a $100B+ space where total labor spend “far exceeds” the software revenue. They’re entrenched not because they’re good, but because they have contracts and switching costs. Stacks positions them as “costly to deploy and poorly rated by customers,” which is corporate-speak for: everyone hates them, but no one gets fired for buying BlackLine.
The real insight: Stacks isn’t competing on AI capability. It’s competing on implementation pain.
Finance automation arrived last—after legal AI, after customer support bots, after the automation waves hitting different functions—not because the tech was hard. It arrived last because ERP systems, spreadsheets, and legacy data lakes are fragmented by design. Stacks built a “dedicated financial data layer” before deploying agents. Translation: the AI is the easy part. The hard part is reconnecting 15 years of Excel hell.
The “agentic” marketing hides a deterministic reality—and that’s actually the smart move
Stacks positions itself at the forefront of the agentic AI trend, but the architecture tells a different story. The company emphasizes “deterministic machine learning“—not pure LLM agents—to ensure reliability at enterprise scale. That’s not a criticism. It’s the only honest way to sell this to CFOs who’ve seen ChatGPT hallucinate.
The catch: this limits flexibility. The magic is constrained, rule-based, and less “agentic” than the pitch suggests. But it increases trust, which is necessary for finance.
Most enterprises will face another catch: months-long data integration before they see ROI. Stacks targets greenfield deals and dissatisfied customers, not the installed base. Because ripping out BlackLine mid-audit season is how CFOs get fired. The honest trade-off: this works, but only if you’re desperate enough to rebuild your data infrastructure first.
Finance teams now have proof the technology works. Nivola’s eight-day reduction is real. But the question isn’t whether AI can automate the close—it’s whether your CFO trusts an AI agent more than they fear the audit committee. Stacks raised $23M betting the answer is finally yes. The 100,000 hours say it should’ve been yes five years ago.
