Latch.ai: Automate complex accounting work without a single prompt
Latch launched its accounting automation platform on Product Hunt on November 10, 2026. It targets month-end close work that accountants repeat by hand, such as pulling statements, matching ledgers and overriding rates. Instead of writing prompts, a user records a task once on their desktop while talking through it. Latch captures voice, screen and clicks, then asks about exceptions that written procedures often miss, such as a three-day matching window.
From that session Latch builds a sidekick, a scoped automation that repeats the steps in existing tools. When a case falls outside the pattern, it stops and returns the decision to a person. Each sidekick runs in its own sandbox with access limited to approved connections, and users can preview actions before files or data change. Latch's own finance team reports that a month-end Xero-to-Google Sheets workflow fell from more than 200 clicks to about a dozen, saving 15 minutes per run, according to a Business Wire release covered by CPA Practice Advisor. A separate write-up by CEO Stefan Kalb cites five clicks, so the exact figure differs by source.
The launch came alongside Latch's State of AI Automation at Work 2026 survey of 742 U.S. working professionals, which found that 73% say AI output needs a person to step in at least sometimes, rising to 84% in finance and accounting. Customers pay only when a sidekick runs, according to the company's website. A launch write-up on KP's Column said the Product Hunt listing offered 50% off the Solo plan through December 10, 2026.
Why it matters
For founders, Latch's pitch rests on the claim that most AI finance tools skip the knowledge that lives in people's heads. The company frames AI shortfalls as a missing context problem rather than a model problem, and it asks follow-up questions instead of only recording screens. Teams selling to finance buyers should expect those buyers to probe controls, so sandboxing, approval gates and step-level logs are part of the core product rather than add-ons.
For investors, the public evidence is still thin. A Kalb write-up reports that Latch has raised $8M, and the 200-click example comes from the company's own team rather than an outside customer. Per-run pricing ties revenue to usage, while MCP access lets other AI agents draw on the captured process knowledge. That could widen distribution, but it also raises a lock-in question that diligence should test.
Key points
- 1Latch records a user's voice, screen and clicks while they perform a recurring accounting task once.
- 2The company reports a month-end Xero-to-Google Sheets workflow fell from more than 200 clicks to about a dozen.
- 3Each sidekick runs in its own sandbox on the user's machine, with access limited to approved connections.
- 4Latch's survey of 742 U.S. workers found 84% of finance and accounting staff say AI output sometimes needs human review.
- 5