// CASE 04 · SHARELOCK · AI-NATIVE SECURITY
Sharelock detects identity threats by learning how every user behaves and spotting the anomaly. The machine learning was patented. The interface was the problem: analysts could not finish critical tasks in the first prototype, and managers could not read the risk reports.

fig 01 · Access, compliance and telemetry in one view, redesigned.
fig 01 · Access, compliance and telemetry in one view, redesigned.

fig 02 · Dashboard details: the anomaly, then the action.
Sharelock collects data from any source, logs, sensors, applications, builds a behavioural baseline per user and flags what deviates. That is a lot of data on a screen, for two kinds of people: analysts who need depth and speed, and managers who need a decision. We observed analysts doing their tasks, or failing to, redesigned the critical flows around what they were actually trying to do, and set up a design system so the product could stay coherent across every view without losing usability.

fig 03 · The component library the development team built from.
Qualitative usability testing
Atomic design
Cross-functional design ops
Component library

Clarify complexity, keep the depth
Surface the critical action first
Show what each action does before it does it
Enable expertise, do not dumb it down
Critical workflows became completable end to end by analysts, and managers got reports they could act on. The design system gave Sharelock a way to add features without redesigning the product each time, and the interface became part of the sales pitch. What I cannot show you: task completion rates come from our internal testing, not from a published study, so treat them as our observation. What stays with me: the same principle, organise rather than simplify, is what I apply today when I design interfaces on top of AI systems that produce more signal than a person can read.