Latch - AI Process Capture and Workflow Automation for Finance Teams

Latch captures how your finance team actually works and turns those processes into reliable automations - without anyone writing prompts or code. Instead of building workflows from scratch, Latch learns directly from the people who do the work every day.

A user starts a recording in the Latch desktop app (available for macOS and Windows), performs a task the way they normally would, and talks through their process. Latch captures three streams simultaneously:

• Narration - the user's spoken explanation of what they are doing and why
• Screen activity - screenshots on every event plus a full screen recording
• Accessibility data - clicks, keystrokes, and copy-paste actions

Latch's models process these streams into a detailed process model and a knowledge graph of what was done and why - including exceptions, edge cases, and judgment calls that never make it into written SOPs. Latch then asks follow-up questions about decision criteria and edge cases it observed, ensuring nothing is missed.

From a captured session, users can build a sidekick - a scoped automation that runs the process the same way every time. Sidekicks are deterministic code, not unpredictable AI agents. AI is used only where judgment is genuinely needed.

When a sidekick encounters something that does not match what it learned - for example, a customer name that differs between two systems - it stops, explains the issue in plain language, and asks the user how to proceed. This approach keeps humans in control while eliminating repetitive manual effort.

Transparency and control at every step:

• Before the first run, Latch lists every step in the sidekick's scope and the reason for each permission it requests
• Users approve only the apps the sidekick needs and can test before it touches real data
• Each sidekick runs in its own sandbox on the user's machine
• Every run produces a step-by-step log for full auditability

The captured knowledge graph can also be made available to your organization's own AI tools over MCP, extending the value of documented processes across your technology stack.

Latch is initially focused on accounting and finance teams, whose processes are long, detailed, and exception-heavy:

• Month-end close
• Bank reconciliations
• Invoicing
• Moving actuals into financial models
• Accounts payable and reporting workflows

Real-world example: A controller's monthly task of pulling actuals from Xero into a Google Sheets financial model took about 15 minutes and more than 200 clicks. After one recorded session, a Latch sidekick runs it on a schedule with no manual time required.

Latch is designed for mid-market and larger organizations, typically several hundred employees or more, in industries such as manufacturing, distribution, real estate, and services.

Latch does not replace your current tools - it runs inside the applications your team already uses, including:

• Excel and Google Sheets
• QuickBooks, Xero, NetSuite, and Sage Intacct
• Other finance and productivity applications

Latch is not a general ledger or practice management system. It enhances the systems you already rely on by automating the manual work that connects them.

Latch is SOC 2 compliant. All data is encrypted in transit (TLS 1.2) and at rest (AES-256). Sidekicks run in isolated sandboxes on the user's machine, and every action is logged for complete transparency.

• Eliminate manual process documentation - Latch captures workflows as they happen, creating accurate documentation automatically
• Reduce repetitive manual work - Turn captured processes into sidekicks that run reliably on schedule
• Maintain control and auditability - Every automation is transparent, sandboxed, and logged step by step
• Preserve institutional knowledge - Capture the judgment calls and edge cases that live only in your team's heads
• Deploy without disruption - No system replacements, no code, and no prompt engineering required