AI IntegrationEnterprise UXInteraction DesignUX Strategy

Help Me Choose,
reimagined with AI

Redesigning Microsoft's product recommendation experience — evolving a fragmented quiz into a unified, Copilot-assisted decision system across Windows and Surface ecosystems.

Company

Microsoft

My Role

Senior Product Designer

Team

PMs · Engineers · Data · Content Strategy

Platform

Windows · Surface · Web

Screens shown at reduced fidelity — reach out to see the full process
Help Me Choose · Copilot
What do you mainly use your device for?
Creative work & design
Business & productivity
Gaming & entertainment
C
Based on your choice, I'll narrow this down to 2-3 options that fit your workflow.

The Problem

Two flows, two models, one fragmented experience

Windows and Surface each had their own "Help Me Choose" flow — separate interaction models, separate recommendation logic, separate teams maintaining them. One used rigid decision trees; the other, filter-based profiling. Neither produced confident users.

The deeper issue wasn't information — users had access to specs and comparisons. The issue was confidence. Users couldn't tell if they were making the right decision, and the experience didn't help them feel like they were.

Duplicated internal effort across Windows and Surface teams

Inconsistent interaction models made the system feel unreliable

Cognitive load front-loaded before users understood the trade-offs

AI guidance felt opaque — users couldn't tell why they got a recommendation

No reassurance layer — confident users got the same experience as confused ones

Impossible to scale to a unified system without a shared foundation

My Approach

Reframe from comparison to confidence

Rather than starting with UI, I focused on understanding how people make technical decisions under uncertainty. The reframe: from "help users compare products" to "help users feel confident moving forward." That shift changed almost every subsequent design decision.

1

Understand the underlying recommendation logic

I worked with the data team to map the taxonomy and logic powering recommendations — activities, preferences, environments, product attributes. Understanding this system was prerequisite to designing a UI that reflected it honestly.

2

Define a shared vocabulary that scales

Rather than designing isolated flows, we built a structured taxonomy that could map user intent consistently across products, surfaces, and future recommendation experiences. The design and data architecture were designed together.

3

Design AI as guidance, not automation

Every interaction was designed so the AI supported user judgment rather than replaced it. Recommendations remained editable, explainable, and transparent throughout.

Key Decisions

Three choices that shaped the system

01

Interaction Model

Conversational and reassurance-first, not form-like

We moved away from rigid step-by-step forms toward a more adaptive, conversational structure. Users needed the experience to feel supportive — not technical. Each question was designed to feel like a natural next step, not a requirement.

The tension: more conversational structure meant less explicit progress indication. We introduced subtle reassurance signals — "Based on this, I'll narrow to 2-3 options" — to compensate.
02

Architecture

A shared taxonomy foundation across products

Instead of designing isolated recommendation flows, we created a scalable taxonomy system that mapped user intent consistently. This meant making design and data architecture decisions jointly — unusual but essential.

The tension: shared systems move slower than point solutions. Getting alignment across Windows and Surface teams on a unified vocabulary was a significant coordination challenge.
03

AI Role

Guidance that stays transparent

We designed AI to surface reasoning, not just results. Every recommendation was explainable — users could see why they were being guided toward an option and change their inputs if it didn't feel right.

The tension: showing AI reasoning increases interface complexity. We kept explanations brief, contextual, and non-technical — one sentence, not a paragraph.

Screen placeholder — reach out to see full design

Adaptive Flow · Copilot IntegrationConversational recommendation flow with Copilot-assisted guidance. Progressive narrowing from intent to recommendation. Screens shown at reduced fidelity.

Outcome

Clearer decisions, more confident users

The experience was still in concept and validation stages, but usability testing and stakeholder reviews showed consistent, directional signal.

Decision Flow

Users moved through decisions more naturally without requiring step-by-step guidance — the conversational structure reduced perceived complexity.

Cognitive Load

The single adaptive scroll reduced cognitive load compared to traditional multi-step flows — users reported feeling less overwhelmed.

AI Trust

Transparent reasoning helped users trust recommendations — they understood why they were being guided, not just where.

Internal Alignment

The shared taxonomy created stronger cross-team alignment and established a scalable foundation for future AI-assisted experiences.

Reflection

What I'd do differently

"Start with the taxonomy before the first screen. The data architecture is the design."

The most valuable work on this project happened before any UI was designed — mapping the underlying recommendation logic and building a shared vocabulary with the data team. In hindsight I'd push even harder for that structural work upfront, earlier in the project.

I'd also advocate for more diverse user testing earlier — specifically users across the full spectrum of technical familiarity, not just the middle. The experience behaved very differently at the extremes.

Up next

Talkspace
Empowering therapists with clarity and better tools.
Mental HealthPlatform Design
Talkspace

Want to see the full picture?

The full process is
worth a conversation

There's a lot more to this project — detailed flows, research synthesis, interaction specs, and design system components. If you'd like to see how it all comes together, I'd love to walk you through it.

See the full process

Usually reply within 24 hours · manalishah19691@gmail.com