UX research and redesign for a VR collaboration tool used by architects and interior designers across India's AEC industry.
Year2023
RoleUX Researcher & Designer
CollaboratorsTrezi Product & Engineering
TypeUX, Product, Research
IndustryArchitecture, Real Estate
CompanySmartvizX
ToolsFigma, Miro
Trezi Lens (by SmartVizX) is a VR collaboration tool for the Architecture, Engineering, and Construction industry. It lets multiple stakeholders (architects, interior designers, civil engineers) enter the same virtual space using VR headsets, walk through building models, place furniture, and make real-time design decisions together.
SmartVizX was seeing strong initial adoption but poor renewal rates after Year 1. Users saw the potential of the tool but found it too difficult to use independently. The business needed to understand where usability was breaking down, and fix it.
The Platform
Trezi Lens operates across two distinct modes that architects switch between within a single session. In Tabletop Mode, the user stands inside a virtual studio room and views the imported architectural model as a scaled miniature on a circular platform, like a physical scale model brought into virtual space. From here, teams orient themselves, initiate cross-sections, and place furniture before stepping inside. In Immersive Mode, a single teleport gesture drops the user inside the model at 1:1 scale, where they can walk through corridors, check ceiling heights, and experience the space as a future occupant would.
The desktop companion, TreziViewer, lets non-VR participants observe and annotate from a laptop. TreziViewer mirrors the active session in real time, showing the floor plan, each participant's avatar position, and object properties as they change. This allowed clients and senior stakeholders without a VR headset to attend a spatial review and interact with the 3D model from a conventional workstation, making Trezi accessible across the full project team.
Live collaboration is central to the Trezi proposition. Up to six participants (architects, interior designers, engineers, and clients) occupy the same model simultaneously, each represented as a colour-coded avatar with their name visible overhead. Furniture placements, cross-section cuts, and navigation made by one participant are reflected for all users in real time, turning a spatial review into a shared working session rather than a one-way presentation.
Three failure patterns kept surfacing across every session. First, high cognitive load: menus were deeply nested, with too many steps to complete basic tasks, and features discovered only by accident. Second, navigation confusion: users teleported into walls, got stuck in tabletop mode, and lost track of their position inside the model. Third, low confidence: no feedback, no progress indicators, and no contextual guidance left users guessing at every step.
Research Approach
The engagement was structured around three methods, run across 21 participants from architecture, interior design, and BIM roles.
One-to-one interviews covering background, workflow, tool proficiency, and prior relationship with VR, conducted before any observed sessions.
Contextual shadowing: observing participants perform tasks in their own work environments, watching how they actually used the product versus how they described using it.
Think-aloud usability testing where participants narrated decisions, hesitations, and failures in real time while completing specific tasks, with SUS scores captured before and after redesign.
One-to-one interviews covered background, workflow, tool proficiency, and relationship with VR. Shadowing involved observing participants perform tasks in their own environments, watching how they actually used the product versus how they described using it. Contextual inquiry had participants think aloud through specific tasks, with probing on decisions, hesitations, and failures.
Before any sessions, the team mapped the existing design workflow at architecture firms: how projects move from brief to execution, what tools are used at each stage, and where Trezi was being introduced into that flow.
Who We Talked To
Participants ranged from senior architects at large firms with dedicated IT support, to junior architects designated as the "Trezi person" for their entire organisation, to interior designers at small studios working without any VR experience at all.
Three archetypes emerged clearly. The old-school architect: senior, conventional, prefers in-person communication, uses Trezi only when the IT team sets it up for client presentations. The enthusiastic junior: fast learner, comfortable with SketchUp and AutoCAD, designated to handle all VR sessions, wants to use Trezi for design review, not just demos. The interior designer: works in a small firm, communicates via IM and calls, has never explored Trezi independently, relies on a colleague to guide every session.
What the Research Revealed
No one sets up Trezi properly. Almost every participant just plugged in the hardware and launched the app. VR room setup was skipped entirely, causing tracking failures, blurry visuals, and controllers going out of range mid-session. Nobody read the documentation. Nobody knew the setup mattered.
Discovery was near-impossible. Users defaulted to the left-hand UI to look for options even when the action required direct controller interaction. Features like the cross-section tool were attempted a few times, then abandoned. Users had no way to know if a task they wanted to do was even possible.
Tabletop mode felt like a video game. The blank Farnsworth House used as the default environment had no meaning for AEC professionals. Users felt emotionally disengaged and couldn't relate to it as a professional context. Several said it looked like a game and didn't feel like work.
VR is used only for client demos. Trezi was never explored out of curiosity or self-motivation. There was typically one designated person per organisation who handled all VR sessions, making it an event, not a daily workflow tool.
Collaboration was aspirational, not functional. Although participants named real-time multi-user sessions as a key reason they chose Trezi, almost none had successfully run one without external support. Session invites, avatar management, and participant controls were not discoverable. When errors occurred during a join attempt, there was no recovery path. Users simply called the Trezi support team and tried again.
Usability Baseline
Tasks were measured across time taken, number of attempts, and errors made. The average SUS score across the first round of testing was 58.3, sitting in the "marginal" range.
The furniture placement task averaged over 20 attempts before completion, with the insert flow requiring 8+ steps and no placement guidance. The cross-section tool averaged 5+ attempts across participants, with most never completing it independently. Returning to tabletop mode after navigating inside a model was failed by the majority of participants who tried it on their own.
Design Decisions
Three use cases were redesigned end-to-end: setting up a space, inspecting a building model, and collaborating in real time. Each was mapped from first interaction to task completion, with failure points documented and redesigned against specific usability criteria.
Setting Up a Space
The asset library was buried inside Insert > Object and had no scroll indicator, making the list appear static. Object placement required 8+ sequential steps with no visual guidance, no snapping, and no scale reference. Transform tools were hard to locate after inserting an object.
The redesign surfaced the asset library as a persistent primary action accessible from any state. 3D previews and scale indicators were added to all objects. Placement became drag-to-world: point, drag, release. Contextual action buttons (move, rotate, duplicate, delete) appeared on top of the selected object, not hidden in a separate tab.
Inspecting a Building Model
The cross-section tool was undiscoverable: no contextual prompt, no indication it existed at the tabletop. Teleporting placed users inside walls repeatedly. Rendering mode changed accidentally when users tried to interact with other controls. Height adjustment was numeric input only, with no visual feedback.
The redesign added a contextual prompt at tabletop mode pointing to cross-section tools. Teleport was given an undo shortcut (right stick backward). Rendering mode was moved off the left-hand menu to a persistent toggle button. Invalid interactions were blocked with explicit redirects to the correct method.
Collaborating in Real Time
Collaboration was the feature that most users had heard about but never successfully run. Session invites were sent through a separate desktop application outside the VR environment, avatar names were illegible at model scale, and there was no shared pointer or in-session annotation tool. When two participants needed to direct each other's attention, saying "look at this wall" or "come stand here", they did it verbally over a phone call. The interface gave them nothing. Multi-user sessions were abandoned more often than completed.
The redesign addressed three gaps. First, session management moved into the VR interface itself. A persistent Collab panel showed active participants, allowed direct invite from within the model, and surfaced join errors with clear recovery steps. Second, a shared laser pointer was introduced: any participant could activate it to draw another user's eye to a specific surface, object, or spatial position in real time. Third, avatar names were rescaled to be legible at both tabletop and immersive viewing distances, with a colour-coded participant list visible from any location in the model.
Onboarding
A mandatory SteamVR room-setup gate was added before first launch. A spatial first-run tutorial was designed using balloons and rings placed around the environment to teach teleporting, view rotation, object selection, and cross-section in context, one interaction at a time, before the user enters a real model. Controller tooltips were added to all buttons explaining what each one does.
Outcome
The average SUS score after the redesign round was 90.8, an improvement of 32.5 points, moving from marginal to excellent.
SUS Score: 58.3 → 90.8 (marginal to excellent, +32.5 points)
Furniture placement attempts: 20+ reduced to ~3 on average after redesign
Cross-section task completion: from near-zero independent success to consistent first-attempt completion
Multi-user session success: participants able to join and run collaborative sessions without external support for the first time
Beyond the score, the SmartVizX design team went through the full research and design process end-to-end alongside the work, building internal capability to continue iterating. Affinity analysis, SUS scoring, and contextual inquiry were formalised as repeatable methods. The onboarding tutorial was delivered as a full interaction spec ready for engineering.
Reflections
VR usability is physical, not just digital. Users don't just struggle with the interface. They struggle with wearing the headset, staying within tracking limits, and physical disorientation. Designing for VR means designing for the body, not just the screen.
Self-reported usage and actual usage were very different. Users said they used the cross-section tool. Sessions showed they had never tried it independently. Research method choice changed what was visible.
Handing off a report wouldn't have worked. Running the design thinking process with the team, not just for them, made the insights actionable and gave them confidence to keep going.
The biggest unlock was realising tabletop mode was perceived as a completely separate application. Reframing it as the primary entry point, not a secondary inspection tool, changed the onboarding logic entirely.
