Skan AI raised $63M in Series C funding to expand a platform that observes how employees actually use enterprise software, converting that behavior into a live model for automating workflows.
The strategic thesis here is sharper than the funding number. The industry has spent two years assuming enterprise AI stalls because models aren't good enough. Skan's argument inverts that: the models are fine, but they're dropped into businesses blind, fed sanitized process documentation that describes how work is supposed to happen rather than how it actually does. That gap—the undocumented exceptions, the workarounds, the tribal handoffs between a CRM and a 40-year-old mainframe—is exactly where agents break. If that diagnosis is right, the winning layer of the AI stack isn't a better model or a better prompt. It's ground-truth observation. That reframes the competitive map away from foundation-model labs and toward whoever owns the accurate picture of how a claims desk or compliance team truly operates.
Globally, this collides with process-mining incumbents like Celonis, which reconstruct workflows from backend logs. Logs only show completed transactions; they miss the messy human middle. Desktop-level observation captures that middle but detonates a governance question executives can't wave away: continuously watching 1,500 employees' screens is surveillance-grade telemetry. Expect works councils, privacy regulators, and labor pushback to shape adoption as much as ROI does.
For Japan, this lands directly on a sore spot. Japanese enterprises poured enormous capital into RPA—WinActor, UiPath, and armies of SIer-built bots—precisely because so much operational knowledge lives in tacit, undocumented routines held by veteran staff. Those bots are brittle for the same reason Skan describes: they encode the documented process, not the real one, and they shatter the moment a screen layout or an exception appears. Skan's approach is, in effect, the intelligence layer Japan's RPA wave was missing.
That's both threat and opportunity for SIers like NTT Data, Fujitsu, and Nomura Research Institute. Their labor-intensive model—flying in consultants to map processes by hand—is exactly the cost Skan wants to automate away. But Japanese SIers also own the deepest client relationships and the compliance credibility to deploy screen-observation responsibly under strict labor and privacy norms. The smart move is to reposition from manual process-mappers to orchestrators of observation-grounded agents, capturing tacit senior-employee knowledge before demographic attrition erases it. The firms that treat this as a productivity threat will lose margin; those that treat it as a new service line could turn Japan's automation debt into an advantage.
