Case Demonstration
Strategic problem ownership:
identifying a known industry failure mode and proactively re-framing the product's approach before it became a launch barrier
Methodology innovation:
Building a research framework from scratch when standard UX tools were insufficient
Data Driven Results
Quantitative population reasoning driving architecture decisions rather than assumed averages
Safety-aware design:
understanding fit as a structural and liability constraint, not just a comfort preference
Cross-functional leadership through expertise:
coordinating hardware, optics, and manufacturing decisions without formal authority
Manufacturing and business awareness:
connecting design decisions to component costs, production volume, and market adoption risk
The Problem
Our product's core value proposition was a next-generation AR experience for a remote or travelling worker able to use multi-screen workflows, anywhere they go. Unlike conventional displays where resolution specs (4K, etc.) communicate quality, AR display quality, is unknown, and is not easy to capture unless the person wears the product. This is especially true for our product that far exceeds monitor resolution.
As part of word-of-mouth marketing, we wanted to make sure that share-ability was possible. A colleague, a potential customer, a manager in a meeting; anyone needed to be able to pick up the glasses and immediately understand what the product could do.
As part of word-of-mouth marketing, we wanted to make sure that share-ability was possible. A colleague, a potential customer, a manager in a meeting; anyone needed to be able to pick up the glasses and immediately understand what the product could do.
The original assumption was universal fit. I knew this assumption needed serious scrutiny, as at the same time, we needed precise eye position across the population so that the glasses can display correctly.
AI Generated depiction of the AR use-case. On the go, full virtual screens with retina level resolution, that exceeds monitors, while being in a light form factor.
AI was used to quickly evaluate aesthetic style, ergonomic fit, and use-case quickly.
Why Universal Fit Was a Known Industry Failure
North Glasses: One of the more well-funded AR eyewear companies had attempted this problem before us, before folding. A significant factor was that each frame required an in-person fitting appointment to calibrate the optics to the individual wearer. It was a fundamental adoption barrier that their product never recovered from.
Our product couldn't repeat that mistake. At first, we looked at replicating the form factor of safety glasses as they are indeed universal fit. However, safety glasses are lightweight and flexible, allowing them to conform and self-adjust. Our frames were heavier and more rigid as they contain internal components for electronics and optics. The safety glasses benchmark was a false signal.
I presented these findings to the team and proposed we move away from universal fit toward a sized architecture, initially proposing 3 sizes, later consolidated to 2 during cost review, then confirmed back to 3 as the data supported it.
Why are Standard UX Research Methods not appropriate here?
Investigating fit through conventional means hitting immediate walls:
Users couldn't self-report meaningful measurements
Even people wearing glasses daily couldn't describe their IPD (Interpupillary Distance, the length between two eyes), head breadth (The width of the widest part of the head), or where their temple sits in their ear
Critical measurements don't scale proportionately
Someone with a small IPD can have a large head breadth, invalidating any correlated sizing assumption
Publicly available Anthropometric data (Measurements of human anatomy) for specific landmarks
Particularly the otobasion superior, where the temple rests on the ear, was surprisingly scarce
Manual data collection would have required a formal professional study of up to a year
This forced a complete rebuild of the research methodology from scratch.
The Safety Failure That Reframed Everything
Early in physical prototyping, I tested 3D printed frames on my boss. They shattered erupting plastic shards from the weak point, the hinge, near the eye.
That moment fundamentally re-framed the project. A frame too small for a user's head doesn't just feel uncomfortable. It can fail, potentially causing injury to the eye and face. Fit integrity became a safety constraint, not a comfort preference. This raised the stakes of getting the sizing architecture right beyond usability into product liability territory.
Building a New Research Framework
Identifying the measurements that actually matter
Rather than relying on user self-reporting, I identified the bio-mechanical landmarks that govern fit:
IPD: for accurate AR display alignment
Otobasion superior to eye center: vertical eye position
Nosepad position: vertical frame resting point
Head breadth and cheekbone points: frame clearance, critical for Asian fit
Brow ridge geometry: an original finding: the frame resting on the brow ridge, not the nosepad, sets the Z-position (depth). This landmark had no established substitute in published studies and required original measurement definition
Generating synthetic population models with AI
Iterating against real head geometry from photographic references was unreliable as the complexity of head shapes made consistently identifying landmarks very difficult. A larger problem emerged when attempting AI-generated reference views: top-down orthographic head images essentially don't exist. Most people have never seen another person from directly above, let alone photographed it. The rare exception, I suspect are of barbers photographing their work. While close, this introduces close-focal-length distortion that makes them useless as design references. No amount of verbal prompt engineering produced usable results because the model had almost no training data to draw from.
The solution: I rendered a scaled top-down orthographic view from a 3D head model, included a visible scale reference, and used that image as the AI's visual anchor instead of text descriptions. With a concrete reference to work from, the AI generated consistent, usable orthographic views across all three perspectives — front, side, and top.
Using raw anthropometric data from public sources, I generated statistically valid synthetic head profiles at the 1st, 5th, 25th, 50th, 75th, 90th, and 99th percentile for both men and women for a repeatable, scalable reference system for design iteration without requiring physical study participants.
Benchmarking the Market
I 3D scanned safety glasses from multiple manufacturers. Despite coming from different companies, the designs were surprisingly similar. The industry had converged on a rough consensus for universal fit geometry. However, direct replication wasn't viable given our frame constraints.
A useful parallel emerged from consumer eyewear: sunglasses. The closest commercial product to a multi-size solution as typically offer one small and one large per sex. Notably, the large female and small male are often the same physical model. This would later validate the sizing architecture I arrived at independently through the population data.
The Key Insight: Universal Fit Is a False Premise
Testing frame geometry against the full percentile range confirmed that a single universal size couldn't perform well across the population. The typical human factors instinct is to design for the 99th percentile. Usually this would allow for a design that can fit anyone, the way door frames are sized for tall men allows anyone to past through. But this breaks down for eyewear for both extremes:
Too large: Frames slip off small faces. At a design industry event, I informally tested a leading AR product by asking attendees to try it on. On smaller women, the frames fell off instantly. Ignoring even 5% of the population isn't a rounding error — it's a meaningful segment of users who simply cannot use the product, undermining the shareability proposition entirely.
Too small: as the prototype test demonstrated, frames placed on too large of a head cause discomfort and mechanical stress which can break, shattering dangerously near the eye
Why 2 sizes also fails
Two sizes seems like the logical middle ground. But it introduces a critical self-selection problem. Men and women follow different bell curve distributions with different standard deviations. A person at the 40th percentile has no reliable way to know whether to choose the small or the large — and critically, most people have no frame of reference for where their head size falls relative to the population. Designing the sizes to overlap deliberately doesn't solve this; it just relocates the ambiguity onto exactly the people least equipped to resolve it.
The 3-size architecture
The population data revealed a natural structure:
Small: optimized for smaller female fit (approximately 1st–20th percentile female)
Unisex: covering approximately 70% of the population, leveraging the significant overlap between larger women and small-to-mid males
Large: accommodating larger male proportions (approximately 75th–99th percentile male)This structure also solves the self-selection problem behaviourally. It's cognitively easy to identify yourself as clearly small or clearly large. The ambiguous middle — the majority of users — naturally self-selects into the unisex size without requiring precise self-measurement.
Manufacturing and Component Strategy
The 3-size architecture created downstream efficiencies beyond fit:
The unisex size serves ~70% of demand, concentrating manufacturing volume where it drives the most cost reduction
The small and large sizes are produced in lower quantities, present enough that no user is excluded but not overproduced
A further finding: the lens combiner for small and large can use the same component as the unisex size. This is partly driven by fashion. Women's frames conventionally skew larger as oversized lenses create a slimming effect, which means the optical geometry overlaps more than head size alone would suggest. I verified this independently and confirmed it against industry norms.
Only the large size requires extended temples, as consistent with industry standards and confirmed through independent analysis, further reducing component permutations
The result is a 3-size system with fewer unique components than it might initially appear, reducing manufacturing complexity while maximizing population coverage.
Cross-Functional Coordination
In the absence of a formal product manager, I coordinated directly with the electrical and mechanical engineering leads to ensure design decisions remained feasible across the full hardware stack:
With the electrical engineer: directed board component placement, negotiated board shape to fit within frame geometry, and established the boundaries of what the internal layout could and couldn't accommodate — a design-led process rather than receiving constraints passively
With the mechanical engineer: worked through hinge design (particularly complex given internal component routing), wall thickness constraints, and material behaviour at the tolerances required for both fit and structural integrity
These collaborations were iterative and bidirectional as design decisions shaped engineering constraints as much as the reverse.
Outcomes
A 3-size architecture confirmed manufacturer and validated against full anthropocentric population data
A novel AI-assisted orthographic modelling methodology developed from scratch, applicable to any product requiring spatial fit validation without a physical study
A component reduction strategy (shared lens combiner, shared temple across two sizes) lowering manufacturing complexity
A self-sorting size system that solves the user self-selection problem behaviourally, removing the need for measurement or fitting appointments, directly addressing the failure mode that contributed to North Glasses' collapse