B2.1.13 Modelling

The model must be made at the correct scale and with accurate dimensions for the aspects being tested.

If you are testing whether a handle fits comfortably within the grip range of your target user group, a handle that is the wrong diameter will generate completely misleading performance data — regardless of how beautifully it is made.

The model must be made from materials that sufficiently represent the properties of the intended final materials for the aspects being tested.

If you are testing grip security in wet conditions, a model with a smooth 3D-printed surface will not generate representative data about a final product with a textured rubber surface. The material must be realistic enough that the test reflects real-world performance.

If the test involves operational interaction — using the product to perform a task — the model must function realistically enough that user interaction data reflects genuine use.

A model of a can opener that looks correct but whose cutting mechanism does not actually cut cannot generate meaningful data about operational force, cutting quality, or user confidence during real use.

The model must be tested in a context that is sufficiently representative of the real use environment to generate valid performance data.

A kitchen tool tested in a design studio — on a clean, stable workbench, by a relaxed participant with no time pressure — may generate very different performance data than the same tool tested in an actual kitchen — on a wet, cluttered surface, by a user who is simultaneously managing cooking tasks.

What is it?

Low-fidelity models are quick, rough, and inexpensive representations that capture the basic idea of a design without precise dimensions, realistic materials, or refined form. They are made rapidly from whatever materials are most accessible — cardboard, foam, clay, paper, tape.

Characteristics:

• Made in minutes or hours rather than days

• Materials are substitute — foam represents rubber, cardboard represents structural components

• Dimensions are approximate rather than precise

• Surface finish is crude — no attempt at realistic appearance

• Often single-use — expected to be modified or discarded after testing

What low-fidelity models are used to test:

• Does the basic concept make physical sense?

• What is the approximate size and proportion of the design?

• Which of several competing concepts shows the most promise?

• What is the basic interaction pattern — how will the user hold, operate, or navigate the design?

What low-fidelity models cannot reliably test:

• Precise ergonomic fit against specification dimensions

• Material performance — grip, durability, weight

• User confidence and satisfaction

• Detailed mechanism function

Real-World Example:

When designers at IDEO were developing a new patient room layout for a major hospital, their early low-fidelity models were made from cardboard boxes, masking tape, and paper labels — assembled directly on the floor of a hospital conference room to represent furniture, equipment, and circulation routes at full scale.

These crude cardboard room models allowed clinical staff and patients to walk through the layout — physically experiencing spatial relationships and identifying circulation problems — in less than half a day. Design issues that would have taken weeks to identify through drawing review alone were discovered and resolved in a single afternoon session.

The low fidelity of the models was not a limitation — it was the point. Their speed and disposability allowed rapid exploration of multiple layout alternatives that would have been impossible with high-fidelity construction.

What is it?

Medium-fidelity models represent a significant development step from low fidelity — incorporating accurate dimensions, more realistic materials, and sufficient functional detail to support more rigorous user testing and specification comparison.

They are not as quick to make as low-fidelity models, but they generate significantly more reliable and specific performance data.

Characteristics:

• Made over hours or days

• Dimensionally accurate in the aspects being tested

• Materials chosen to represent the performance properties of final materials

• Mechanisms functional to the degree required for testing

• Surface finish approximate — not polished but recognisable

• Designed for multiple use — robust enough for repeated testing sessions

What medium-fidelity models are used to test:

• Ergonomic fit against specification dimensions — grip diameter, reach distance, handle length

• Basic mechanism function — does it operate as intended?

• Operational force requirements against specification thresholds

• Initial user interaction quality — ease of use, error frequency, user confidence

• Material performance in simulated use conditions

What medium-fidelity models cannot reliably test:

• User aesthetic response — appearance is too approximate

• Long-term durability and material degradation

• Manufacturing feasibility

Real-World Example:

When OXO was developing new handles for their Good Grips kitchen tool range, their medium-fidelity models were carved from dense polyurethane foam at precisely specified dimensions — matching the handle diameter, length, and cross-sectional profile of the intended design.

These foam models were dimensionally accurate enough to generate reliable ergonomic test data — grip force measurements, hand pressure distribution maps, and reach assessments — across a diverse user group including participants with arthritis and limited dexterity.

The foam material was not the intended final Santoprene rubber — but its compliance and surface properties were sufficiently similar to generate valid comparative data about grip performance across different handle geometries.

At this stage, OXO was not testing whether the handle looked right — they were testing whether it felt right and performed correctly. The fidelity of the model was precisely matched to the questions being asked.

What is it?

High-fidelity models are detailed, accurate, and realistic representations of the intended final design — made from final or near-final materials, at precise dimensions, with fully functional mechanisms and a resolved surface finish.

They are the most resource-intensive models to produce but generate the most reliable and comprehensive performance data — particularly for aesthetic evaluation, user confidence and satisfaction testing, and pre-production validation.

Characteristics:

• Made over days or weeks

• Final or near-final materials and components

• Precisely dimensioned throughout

• Fully functional — operates as the final product will operate

• Surface finish resolved — appearance closely represents the intended final product

• Robust enough for extended use testing and multiple test sessions

What high-fidelity models are used to test:

• Complete user experience — functional performance AND aesthetic and emotional response

• User satisfaction, confidence, and preference

• Durability and material performance under realistic use conditions

• Manufacturing feasibility — can this actually be made as designed?

• Compliance testing against standards and regulations

• Pre-production validation — does the final design perform as specified before manufacturing investment?

What high-fidelity models cannot efficiently test:

• Rapid iteration — making changes is slow and expensive

• Multiple competing concepts — cost prohibits testing many alternatives simultaneously

Real-World Example:

Before committing to full production of their Xbox Adaptive Controller, Microsoft produced high-fidelity pre-production prototypes made from final-specification materials — the same ABS polymer body, the same button mechanisms, the same external port hardware — that were tested by a diverse group of disabled gamers across extended gaming sessions.

These high-fidelity models were indistinguishable from the production controller in terms of function, feel, and appearance. Testing with real users at this fidelity level generated two categories of critical performance data:

1. Functional validation — confirming that every essential specification criterion was met under real gaming conditions

2. User experience validation — confirming that the emotional experience of using the controller — the confidence, the independence, the sense of equal participation — matched the human goals that had driven the entire design process

The question being asked determines the level of detail required to answer it reliably.

Higher fidelity requires more time, more materials, and more fabrication skill. Matching fidelity to available resources prevents the common mistake of spending excessive time on high-fidelity models early in the development process — when low-fidelity models would generate equally useful data at a fraction of the cost.

If testing is likely to reveal significant changes are needed — as is typically the case in early iterations — low fidelity is appropriate. Making significant changes to a high-fidelity model is expensive and time-consuming.

If the design is well-developed and changes are expected to be minor refinements — as in later iterations — higher fidelity is justified.

Quantitative performance data consists of numerical measurements that can be directly compared against specification criteria:

Recording quantitative data:

Quantitative data should be recorded in structured data tables that clearly show:

• The specific measurement taken

• The result for each participant

• The mean result across all participants

• The specification criterion being tested

• Whether the result passes or fails the criterion

This data table immediately reveals that while the mean operational force meets the specification, one participant (P5 — severe arthritis) could not complete the task — an important finding that requires design investigation.

Qualitative performance data consists of descriptions, observations, and user responses that capture the experiential dimension of product performance.

Recording qualitative data:

Qualitative data should be recorded as direct quotes, structured observation notes, and thematic summaries that connect user responses to specific design features:

These qualitative records reveal dimensions of performance — user confidence, emotional response, coping strategies, dignity concerns — that quantitative measurements cannot capture.

Comparative performance data measures performance between iterations — demonstrating that each refinement cycle has produced genuine improvement:

Comparative data tells the most compelling story of iterative development — showing measurable, progressive improvement toward specification criteria and user need satisfaction across iterations.

Before each testing session, define exactly what performance data the session needs to generate — derived from the gaps and questions identified in the previous iteration.

Testing participants must match the characteristics of the persona — their age, physical capabilities, relevant conditions, and use context. Testing with users who do not represent the target group generates performance data that does not reflect how the design will perform for its intended users.

For universal design projects, participants should represent the full range of the target user group — not just the most typical or most accessible participants. Testing only with the least physically constrained users within a target group will consistently overestimate the design's performance for more constrained users.

A test protocol is a structured plan for how the testing session will be conducted — ensuring consistency across participants and ensuring all required performance data is generated.

The test protocol specifies:

• The task participants will perform — described in neutral language that does not suggest how to perform it

• The testing environment — where testing takes place and what conditions apply

• What will be measured — the specific quantitative data to be recorded

• What will be observed — the specific qualitative data to be recorded

• The questionnaire to be completed after testing

• The interview questions to be asked after testing

Follow the protocol consistently across all participants:

• Use the same instructions for all participants

• Measure the same metrics for all participants

• Do not assist participants during testing — assistance invalidates the independence data

• Record everything — do not rely on memory

After testing, systematically analyse all data collected:

• Calculate means and ranges for quantitative measurements

• Identify themes in qualitative responses

• Compare results against specification criteria

• Identify the most significant findings — what succeeded, what failed, what was unexpected

• Determine refinement priorities for the next iteration

When designers at Radius Design were developing a universally accessible toothbrush handle — designed for users with limited grip strength, children, and elderly users simultaneously — their modelling progression demonstrated precise fidelity matching:

Low-fidelity stage:Twelve different handle cross-sections were carved from modelling clay in a single afternoon — varying diameter, cross-sectional shape, and surface texture pattern. All twelve were tested simultaneously with a group of ten participants who held each one and ranked their preference. This session generated comparative preference data across twelve alternatives in under two hours — a rate of model production and testing that no higher fidelity approach could have matched.

Medium-fidelity stage:The three highest-rated cross-sections were 3D-printed at precise specification dimensions from a flexible polymer approximating the final elastomer material. These medium-fidelity models were tested with 20 participants — including elderly users and users with arthritis — measuring grip force, task completion, and satisfaction scores.

High-fidelity stage:The single selected design was produced as a fully resolved prototype using final materials — a co-moulded rigid nylon core with Santoprene overmould — and tested with 35 participants across the full target user range. This high-fidelity testing generated the comprehensive performance dataset submitted to the manufacturer as evidence of specification compliance.

When Transport for London was developing their standardised tactile paving system — the raised dot and bar patterns on pedestrian surfaces that guide visually impaired pedestrians — they tested models at multiple fidelity levels with end-users:

Low-fidelity testing:Cardboard sheets with raised foam dots and bars in multiple geometric configurations were placed on the floor and walked over by visually impaired participants using white canes. These crude models generated immediate comparative data about which configurations were detectable and navigable — eliminating non-viable configurations rapidly and cheaply.

Medium-fidelity testing:Rubber mats moulded with precise raised pattern geometries at accurate specification dimensions were tested in controlled environments — generating quantitative data about detection reliability, walking speed, and navigation accuracy across participants with varying levels of visual impairment.

High-fidelity testing:Full paving sections were installed in test environments that replicated real street conditions — uneven sub-bases, wet surfaces, varying lighting — and tested with diverse end-users including elderly visually impaired participants and guide dog users. This high-fidelity testing generated the performance data used to establish the national standard.

When the NHS was developing a universally accessible digital patient portal — designed for use by elderly patients, patients with visual impairments, and patients with cognitive conditions — their modelling approach demonstrated expert fidelity matching in digital design:

Low-fidelity stage (Paper prototyping):Screens were drawn on paper — individual interface elements cut out and rearranged by participants during testing. This zero-technology approach generated rich qualitative data about navigation logic and information architecture from elderly and cognitively impaired users who would not have been able to interact with even a basic digital prototype.

Medium-fidelity stage (Clickable wireframes):Digital wireframes — grey-scale, unlabelled, with basic click-through navigation — were tested with screen-reader software to generate quantitative data about accessibility compliance. User testing generated task completion rate data and qualitative observations about navigation confusion without the visual design elements that might have biased responses.

High-fidelity stage (Functional prototype):A fully designed, functional prototype — complete with real content, colour, typography, and interaction animations — was tested with a diverse user group including participants using screen readers, participants with cognitive impairments, and elderly participants using the portal on smartphones, tablets, and desktop computers simultaneously.