← Back to ObviousWins.com

Case · UX Lead · Webroot & OpenText · 2021–2025

Scam detection designed to open a new revenue channel for Webroot

Customers wanted their security app to answer one question: is this real? I led UX for the feature that answers it — mobile first, with a VPN-bundled premium desktop tier and a partner-distributed channel behind it.

Role
UX Lead & AI Product Design
Scope
Research · Conversation Design · Responsible AI · Partner API UX
Team
UX, Engineering, Product, Marketing, Sales
Organization
Webroot & OpenText

01 The Work

“Is this real?” — a verdict in one paste.

One input takes email text, messages, files, images, video, and links. Detection runs on a partner API; the design owns the trust layer around it — what the product asks for, what a verdict is allowed to say, and how the wait and the answer read. Swipe the journey, then the mobile flows, verdict states, and supporting tabs.

Swipe or use arrows

Fig. 1–8 From suspicious content to a plain-language verdict with one recommended action. Tap any figure to view it full size.

Mobile first

I wrote the cross-functional user requirements for the mobile product, then managed and coached offshore design teams to deliver against them — running reviews, resolving requirement questions, and holding the verdict vocabulary and entry-point model consistent with the desktop UX.

Swipe or use arrows

Fig. 9–12 Mobile first: paste, camera, gallery, and QR entry, one decision per screen.

Four verdicts, one vocabulary

Swipe or use arrows

Fig. 13–16 Every state commits to a position and names a next step. No countdowns, no fear copy, no upsell in the verdict.

Supporting tabs

Swipe or use arrows

Fig. 17–19 The other two tabs carry the handoff to mobile and the record of what’s been checked.

FigmaAI Scam Protection (desktop UX) · animated interaction prototype, opens in a new tab

02 Process

The detection was accurate. The assessment wasn’t readable.

The partner’s engine found fraud well. What it returned was written for analysts: a confidence score, matched signals, vendor terms — no verdict, no action. In testing, people couldn’t tell whether they’d been told yes or no.

Redefine the assessment itself

Plain-language verdict first, ranked evidence second, one recommended action third. The spec changed what Webroot rendered and what the partner reported — they adopted the revised model on their own side, after joint reviews on real output and annotated redlines both engineering teams built against.

Four principles, set before any wireframes

Generative research across three personas found the same thing: people felt protected without knowing what protection meant, and abandoned tools that felt alarmist.

Principle 01

No decision burden

Users shouldn’t have to know something is suspicious before submitting it. Paste anything; the product classifies it.

Principle 02

Verdict first, never narrative

Answer, then reasoning, then action — the inverse of how most AI assistants build to a conclusion.

Principle 03

No fear copy, no dark patterns

Every verdict went through bias/risk review. Catastrophizing language and upsell prompts were cut, and the rules went into a copy guide for content and engineering.

Principle 04

Privacy as a design constraint

Data minimization, consent, and retention were settled in design review — before the architecture was final, and before the law required it.

Why it mattered

Security products earn trust by answering plainly, not by alarming. A verdict a person can act on — and a copy guide that forbids fear — is what makes the feature worth paying for and worth distributing.

Demo scenario content is fictional and AI-generated for illustration.