Behind the Term Sheet: Skan AI’s $63M Series C

How Skan is becoming the blueprint for agentic AI in the Enterprise.

Investors: Simon Wu & Jonathan Healy; Edits: Jaclyn Hartnett

Read on Medium

We first met Skan in 2020, when enterprise automation looked very different. At the time, organizations struggled to understand how work happened across fragmented systems, making automation difficult and expensive. We believed Skan's computer vision approach to process intelligence gave enterprises something they lacked: a reliable, dynamic picture of how work was performed. That conviction led Cathay Innovation to lead Skan's $14M Series A, along with backing its 2022 $40M Series B.

Today, we're excited to deepen that partnership further. Cathay Innovation and Dell Technologies led Skan's $63M Series C, our third investment in the company and Dell's second. VentureBeat has the scoop here.

While AI has transformed the enterprise software landscape, our core thesis has only strengthened.

In the RPA era, understanding work was the end product. In the AI era, it's become the starting point for building digital labor. That's where Skan's evolution has become especially compelling.

A Quick Recap:

-> Over the past several years, the company has quietly built one of the richest operational understandings of work inside the enterprise, earning the trust of 7 of the 10 largest U.S. banks, 2 of the 5 largest healthcare organizations, and 1 in 4 Fortune 50 companies.

-> Today, that same foundation powers what Skan calls its Context Graph of Work, helping organizations identify where AI can create value, transform observed workflows into agent-ready operating procedures, and increasingly build and deploy AI agents that execute those workflows. Customers typically realize 30-40% efficiency improvements from their initial deployments.

-> The business is beginning to reflect that evolution. Pilot cycles have been compressed from months into weeks. Some customers are skipping pilots altogether and moving directly into production.

-> Existing deployments continue to expand, while the platform itself has evolved into three integrated layers, Blueprint, Intelligence, and Agents, closing the loop from discovery to execution.

-> The latest funding follows a breakout year: Skan AI grew more than 300% YoY; average net dollar retention reached 150% as customers expanded across the enterprise; and the company recently surpassed 25B work signals

Rather than hear it from us, we sat down with Avinash Misra (Co-Founder & CEO) and Manish Garg (Co-Founder & CPO / COO) to discuss what changed, why enterprise AI adoption is accelerating, and where they believe the next generation of enterprise software is headed.

Founder Q&A

Skan Co-Founders Avinash Misra & Manish Garg

Q: If you could go back and talk to yourselves at the Series A, what would surprise you most about where Skan is today?

Avinash: What would shock us most is which enterprises run on Skan today: a quarter of the Fortune 50 companies with our largest customer running Skan across 35,000+ desktops in a hybrid human-AI workforce. We wouldn't have believed the scale curve. In 2020 we launched with the simple idea that if we could observe and capture how all work was done at a company, there would be no end to how we could help them. Today, AI has vastly accelerated that vision along with the ability for companies to transform.

Manish: What we didn't anticipate is that within 5 years customers would ask us to deploy AI agents trained on that observed work at enterprise scale, that observation would become the foundation for digital labor, not just dashboards. We knew that the gap between how we think work gets done and how it actually happens is the root cause of failed transformation, and that computer-vision observation with zero backend integrations was the way to close it. But we underestimated how fast generative AI would commoditize reasoning and push demand toward action, and how quickly the bottleneck to true transformation would move from the model to the context.

Q: What actually changed inside your customers over the last two years?

Avinash: Two years ago the conversations we were having with customers were pretty diagnostic: help us understand this work, help us find where to automate. Today those same conversations are about how to make the transformation happen: how do we put agents into this process safely and at scale? We've gone from observe-and-advise to observe, distill, and deploy. We're having so many more discussions about deploying than ever before.

When we say deploying, that means identifying which parts AI can reliably execute, where humans need to stay in the loop, and then putting that new operating model into production. We're not talking about AI that just generates answers. We're moving agents out of pilots and into workflows, actually doing the work alongside people. It's the realization of agentic AI.

Q: Five years ago, workflow intelligence helped enterprises understand work. Today you're helping them build digital labor. What's the biggest thing the industry still misunderstands about that transition?

Manish: The biggest misunderstanding is that you can take a capable LLM, hand it some policy manuals, and get a reliable worker. In our experience most AI agents fail because they don't understand how humans accomplish their tasks. When you train a new employee, you don't just hand them a manual, of course. Someone has to show them how to do the job. And an important part of that is showing them the exceptions and shortcuts - basically, the difference between what's in the manual and what the best employees do in practice. With AI, we've seen that companies have been doing exactly what we'd never do with humans - just uploading SOPs and manuals and hoping for the best.

Q: A lot of people assume the moat in enterprise AI is the model. You seem to believe it's context. Why?

Avinash: Models are becoming a commodity. Every enterprise can connect to the same frontier models through the same APIs. What's scarce, and what compounds, is an accurate, continuously updated picture of how your company's work is being done. That's why context is now a competitive advantage. As companies have learned over the years, acquiring all of that context is very difficult. Structuring it into something agents can use is also very difficult.

Manish: I think of the context graph of work as an evolving, machine-readable blueprint of how work is done across people, systems, and decisions. Think of it as a missing plane in the enterprise stack, sitting between systems of record and systems of intelligence, where agents can operate on observed reality rather than assumed processes. This context graph of work is what makes enterprise AI both capable of correctly doing the tasks we want while having the governance and security required by our clients.

Q: Many companies can identify opportunities for AI. Skan is increasingly helping customers build and deploy the agents themselves. What gives you the right to move from observing work to actually changing it?

Avinash: Our right to move from observation to action is the observation itself. We know that you can't reliably change what you can't see, and after years of capturing how tens of thousands of workers do their jobs at the world's largest banks, insurers, and healthcare providers, we've built a uniquely rich, continuously observed ground truth for the workflows we support. Most companies deploying agents start from documents and guesses. We start from telemetry.

Manish: We close the loop through what we call Observation-to-Agent. That is, we observe, distill, contextualize, then automate. Blueprint captures the footprint of how work is being done across an organization, and it scores the opportunities it uncovers so you know where you can make the most impact. Intelligence distills that into process maps, metrics, variant analysis, and what we are calling Agent Operating Procedures. That is, machine-usable representations of real work that replace static SOPs. Agents then execute on the real applications teams already use, grounded in that context and bound by guardrails, with humans in the loop. Execution is the ultimate test of enterprise AI.

Q: Financial services has been core to Skan since day one. Healthcare and technology are emerging areas that are rapidly catching up. As you push further into agentic deployments, why does that specialization matter even more now?

Avinash: Financial services is key because that's where the problem is hardest and the stakes are highest. When you look at the work done in that industry - underwriting, KYC/AML, onboarding, servicing, claims - these are all multi-system processes with heavy regulation and near-zero tolerance for error. That's why 7 of the top 10 US banks scale AI on Skan, and we've seen significant savings across banking, insurance, and healthcare. Putting an AI agent inside a bank or insurer is something we have to earn. Trust is the through-line.

Manish: Deploying successful AI agents requires context specific to those industries. For example, a claims adjudication agent has to understand case IDs, entitlements, provider contracts, multi-week lifecycles, and the regulatory boundary around every decision. And sure, a generic agent builder can produce a demo. But we've seen time and again that it can't reliably process an insurance claim.

Q: What's still much harder than people realize about deploying AI agents inside a Fortune 500 company?

Manish: The thing people most underestimate is that most enterprises don't know how their own work is being done. The maps they have are interviews and SOPs. Maybe they've hired a consulting firm to come in and talk to people. The logs they have are partial and system-specific. Nothing they've done captures the real picture of what their employees are doing every day.

Avinash: You can't govern, trust, or change-manage an AI agent if you can't ground it in reality first. Redesigning work so humans and agents share it fairly is as difficult as any technical challenge. The hard part is the combination of context, governance, and change management. We frame the requirement as two loops: a deep, continuously updated Context Loop for how work is being done, and a Control Loop on top of it for reliable, safe, governed action. Many teams are solving pieces of this but underinvesting in those loops, and that's why their pilots don't scale.

Q: Looking five years ahead, what role do you hope Skan plays inside the enterprise?

Avinash: Five years out, if agents become as ordinary as CRM or ERP, the scarce, valuable layer won't be the agent or the model. Intelligence will be ambient, it'll even out across every vendor. What won't even out is context: the thing that makes those agents reliable, governed, and useful. Our belief is that Skan will become that layer, the operating context for the enterprise. In the software era, enterprises needed systems of record. In the AI era, they'll need systems of context.

Manish: We believe the future of work is a mosaic of humans and AI agents working together, and that collaboration is powered by context. If that vision lands, Skan is the Context Graph of Work underneath it: the constantly updating system of record that lets an enterprise know its own operations well enough to trust AI to act inside them. In a world where every enterprise has agents, the winners will be the ones who can capture, govern, and operationalize their context. We intend for Skan to be how they do it.

That's it for now, stay tuned for more from Skan.

Want to join an all-star team? Follow the link for open positions @ Skan