loading...
Skan AI observes employee desktops to build living models of how enterprise work actually happens – then deploys AI agents trained on those models.
ENTRY ANGLES
Benchmarking product using anonymized cross-enterprise process observation data · Supervised AI cockpit for real-time agent oversight in financial services · Process observation for professional services time capture
VERTICALS
CAPABILITIES
On-device processing and privacy-preserving architecture, Enterprise workflow expertise in regulated industries, Compliance and governance knowledge
The reason enterprise AI pilots fail is not usually the model. The model is fine. The reason is that nobody told the model what work actually looks like in this organization. Process documentation describes what was supposed to happen when the system was designed. The actual work – with seven years of workarounds, tribal knowledge, and routing decisions made by one person and never written down – is something else entirely.
Skan AI was built to close that gap. Founded in 2019 by Avinash Misra and Manish Garg, the company deploys observation software to employee desktops that tracks how work moves across applications in real time: spreadsheets, CRM systems, email clients, decades-old mainframes. Screenshots are processed on-device; only anonymized, abstracted metadata transmits to the analytics platform. The output is a living model of actual process – not the org chart version but the version with the detours and exceptions intact.
That model feeds three products. Skan AI Intelligence identifies process patterns and inefficiencies from observation data. Skan AI Blueprint converts those patterns into structured models. Skan AI Agents uses the models to build autonomous agents trained on thousands of observed real cases, tested against reality before deployment, updated as workflows evolve. The sequencing – map first, automate second, verify before releasing – is the core product thesis. A $63 million Series C co-led by Cathay Innovation and Dell Technologies Capital closed in August 2026, bringing total funding to roughly $120 million across more than three rounds. Seven banks, insurers, and multiple Fortune 500 companies are customers. Citi is simultaneously an investor and an operational customer.
Enterprise AI deployments have produced a consistent underperformance pattern that the industry now discusses openly: the pilot works, the broader rollout doesn't. The gap between them is almost always the gap between the controlled scenario the system was designed for and the messy reality of production work. Agents trained on idealized process flows fail on edge cases that experienced employees handle by instinct. Skan's argument is that the map is the product: an accurate model of how work actually flows is more valuable than a faster AI model built on a false picture of the workflow.
The investor composition signals something about the deployment pattern Skan operates in. Citi Ventures is both a backer and an operational user, which means Citigroup's operations teams work inside the platform while the investment team holds equity – an alignment that creates a reference customer with direct incentive to make the deployment succeed. State Farm Ventures and Wipro Ventures carry similar implications: insurers and IT services firms with their own large back-office workforces, watching Skan work in environments similar to their own before committing further capital.
Gartner has recognized Skan across three separate research publications. That matters less as a quality signal than as an enterprise procurement signal: large financial institutions and regulated businesses rarely evaluate infrastructure software that hasn't appeared in analyst coverage. Gartner placement is a prerequisite for vendor shortlist consideration in those organizations, not a marker of product leadership in itself. Skan has been in this market long enough to have accumulated that recognition; a Series C competitor launching today would not.
The $63 million round arrives at a specific moment in the enterprise AI cycle. Eighteen months ago, the conversation was about which model to deploy. Now it's about why the deployment didn't scale. That timing makes Skan's value proposition more legible – and more urgently relevant – than it would have been in an earlier environment.
The observation data Skan accumulates is a proprietary asset whose applications extend beyond the current product. A benchmarking product that lets an operations team compare their loan origination or claims processing workflow against anonymized industry norms derived from Skan's cross-customer dataset would compete directly with consulting benchmarks that cost hundreds of thousands of dollars and rely on self-reported data. Skan's dataset is granular, current, and based on observed behavior. The buyer is the same operations leader who approved the Skan deployment; the product is a natural extension of the same 'see how work actually happens' framing.
The handoff layer between AI agent decisions and human oversight is more urgent and more defensible. Skan maps process and deploys agents; it does not yet build the interface through which a compliance officer monitors what the agents are deciding in real time, flags cases where the agent's confidence is below threshold, and receives notification before an agent takes an action with material downstream consequences. That interface – the cockpit for supervised AI – is the product that determines whether enterprise agent deployment survives governance review. Regulatory and compliance requirements in financial services specifically require demonstrable human oversight; a product that makes that oversight tractable at scale is a deployment prerequisite, not a feature. The buyer is the enterprise risk function, the sale is a complement to the existing Skan contract, and the product is distinct enough from process observation that it doesn't cannibalize the existing relationship.