Triangulation Not Confirmation

Definition of triangulation as discrepancy-led evidence use

Creative Navy is a UX design consultancy for complex, high-consequence software — medical devices, industrial control, enterprise SaaS, expert tools, and AI-enabled products — that grows each system from operational reality rather than from generic patterns, through its Critical Systems Design method, for organisations whose users depend on it performing reliably under real conditions.

Triangulation in Creative Navy's evidence work means using multiple evidence sources against each other. The point is not to run several research methods in parallel until they confirm a shared pattern. The point is to make each source interrogate the others.

Agreement between sources can be useful, but it is not the main signal. Discrepancy is usually more informative because it shows where one source can see something another source cannot. A triangulated finding is therefore not simply a confirmed fact. It is a richer understanding of the product, the market, the competition, user value, sales implications, and technical build requirements.

How triangulation differs from multi-method confirmation

Conventional multi-method research often uses different methods to increase confidence in a finding by confirming it from several angles. Triangulation as Creative Navy practises it uses different methods to challenge each other.

A result that every source confirms may be less informative than a result where sources diverge. Divergence points to something the individual sources cannot see alone. The discrepancy becomes the finding.

This distinction changes the role of research activity. Stakeholder interviews, user interviews, contextual inquiry, field study, competitive analysis, heuristic evaluation, analytics audit, technical constraints discussions, operator environment audit, domain expert knowledge mapping, and legacy system analysis are not valuable only as separate methods. Their value depends on how they are used together.

Feature X example: the difference between time saving and error resolution time

A simple feature example shows why triangulation is not confirmation. Stakeholders may say that feature X is important because users want to save time. Users may also say that feature X is important and that they are under time pressure because each mistake takes a long time to resolve.

Those statements sound like the same finding. In triangulation, they are not treated as identical. The hidden nuance is that the problem may be error resolution time, not general time saving.

That nuance changes the next evidence questions. The issue can be taken back to stakeholders: what does this change about how the feature should be understood? It can be taken into competitive analysis: how do competing products prevent errors or make correction faster? It can be taken into technical constraints discussions: what is technically possible to do differently?

Each source then shines a different light on the nuance. The nuance becomes visible because the evidence sources are used against each other, not because they are made to confirm one another.

User observation reveals behaviour that interview accounts do not contain

User observation is a major source in Creative Navy triangulation because a tool is always embedded in context. The context of use is where many forms of product potential become visible.

User observation can include having users teach the team to use the system and having team members attempt to do the user's job. These activities reveal silent patterns, how users consciously think about the process, how users believe the process should work, how they learned to use the system, how challenges changed as they became more experienced, and the emotional weighting attached to those experiences.

Observation also reveals what users would or would not be capable of if the system were different. This matters because interviews can describe current practice, but observation can expose the compensating behaviour, hesitation, checking, and workarounds that users may no longer recognise as problems.

Users are often more open with a third party than with their own provider. In Creative Navy's evidence work, access to real users through Creative Navy can therefore produce different information from the information the client organisation already has.

Sandbox Experiments extend triangulation from what exists to what could exist

Sandbox Experiments matter to Creative Navy triangulation because discovery is not limited to observing current reality. Prototypes create a new reality that can be investigated alongside the existing one.

This matters because existing behaviour and possible behaviour illuminate each other. A prototype can show what users become capable of when the system changes. Existing observation can show which prototype behaviour is plausible under real conditions. The two forms of evidence interrogate each other rather than sitting in separate phases.

The same activity can produce different evidence depending on how it is pursued. Persisting to find the right participants, returning for a second session, asking a daring question, gaining access to software that is not publicly available for benchmarking, or asking developers to reconsider technical constraints can change what the evidence reveals. The point is not that the design team knows better than technical specialists. The point is that a changed perspective can sometimes generate value even for the developers themselves.

Triopsis workforce management SaaS: five sources revealed structural tensions

The Triopsis workforce management SaaS example used five distinct evidence sources against each other rather than to confirm a single direction. Stakeholder interviews with 5 participants surfaced competing organisational priorities: throughput, stability, safety, and demo performance. These priorities revealed structural tensions rather than a unified product direction.

User interviews included 43 sessions with 21 participants. The user evidence showed that the same interface failed three roles differently. Schedulers described speed problems. Operations managers described exception visibility problems. Technicians described clarity and confirmation problems. These were not one problem described from different angles; they were different operational failures in the same system.

In-situ observation included 3 sessions. Observation revealed behaviour under real-time pressure that interviews did not surface, including hesitation patterns, repeated checking, incorrect sequencing, and duplicate assignments. Behavioural science and ergonomics research was then used to interpret the mechanisms behind the observed patterns, not to generate solutions directly.

Benchmarking provided competitive context and identified specific gaps the client had not recognised as problems. In the Triopsis example, the discrepancies were the findings: stakeholder priorities conflicted with each other, user self-report conflicted with observed behaviour, and benchmarking revealed gaps no internal perspective had named.

Torqeedo maritime HMI: different evidence sources had different roles

The Torqeedo maritime HMI example used five evidence sources with different roles, not merely different perspectives on the same question. Legacy system analysis revealed compensation patterns: how captains had learned to work around fragmented information. This was visible through analysis of the existing system, not through interviews or observation alone.

Sea trial observation involved 15 captains across 12 sessions. It revealed scanning patterns and physical interaction failures that appeared only under real maritime conditions, including vibration, glare from cold water, and night operations. These conditions could not be reproduced in a laboratory and could not have been predicted from interviews.

A controlled experiment with 24 subjects quantified what observation had suggested: captains identified key energy states 50% faster with the new interface. Eye tracking in sea trials with 7 subjects directly counted glance reductions during manoeuvres, producing a measured finding from real operational conditions. Captain satisfaction feedback from 15 captains was unanimous and confirmed the direction of the qualitative findings with structured feedback.

Each Torqeedo source did something the others could not. The controlled experiment quantified what observation had suggested. Eye tracking measured what the controlled experiment had established in principle. Legacy analysis explained why the patterns appeared in the sea trials.

Gexcon CFD simulation software: six sources structured an expert software problem

The Gexcon CFD simulation software example used six evidence sources against each other during the Sandbox Experiments phase. Domain learning included studying manuals, tutorials, and internal training materials; running controlled tests inside the application; and two intensive four-hour stakeholder sessions to reverse-engineer the workflow sequence. This produced an understanding of the system's scientific logic that could not have come from interviews or observation alone.

User interviews included 24 sessions. They showed how senior CFD engineers, safety analysts, process engineers, newer engineers, and risk managers had distinct operational relationships with the same interface. These were not variations on a common theme.

In-situ observations included 23 sessions. Observation revealed behaviour under real working conditions that self-report could not capture, including non-linear navigation between configuration, verification, and interpretation, and the cognitive load of managing dense state within a single environment rather than across distributed screens.

Stakeholder interviews included 9 sessions. They produced organisational and product priorities that did not align cleanly with user research, a discrepancy that pointed at the blanks phenomenon inside the client organisation. Competitor benchmarking across 12 systems identified structural patterns and gaps the client had not recognised as problems. Feature and task analysis documented 102 tasks with goals, frequency, difficulty, and actions, providing the empirical backbone for a specific requirement structure.

The Gexcon discrepancies were the findings. Stakeholder priorities conflicted with observed user behaviour. Internal product knowledge did not account for the full range of user-type differences. Competitor benchmarking revealed gaps no internal perspective had named.

Swiss petrol station operator: transaction coding exposed normalised compensating behaviour

The Swiss petrol station operator example combined qualitative observation and quantitative transaction coding so that each source interrogated the other. Forty hours of structured observation and 24 interviews produced a qualitative picture of cashier friction. A coded dataset of 532 transactions then produced a quantitative picture of how often each friction point arose and under what load conditions.

The coded transactions were categorised by type and complexity, including fuel-only, combined fuel-and-shop, voucher redemption, loyalty, and exception handling. Interview self-report and observation suggested that combined transactions were the primary source of cashier stress. Transaction coding showed a different frequency and complexity distribution: certain transaction types that cashiers rarely mentioned were disproportionately represented in error recovery sequences.

The discrepancy was the finding. Cashiers had normalised personal shortcuts around field-ordering inconsistencies to the point where they no longer registered them as problems in self-report or even in observed-behaviour description. Coding made those patterns visible because it counted them without relying on cashier recognition.

This is structurally different from triangulation based on multiple research methods or multiple user populations. In the Swiss petrol station operator example, a quantitative operational corpus interrogated qualitative observation of the same events.

Akrivia Health: institutional contexts were used against each other

The Akrivia Health clinical research platform example used triangulation across three institutionally distinct user groups: NHS analysts, academic research teams, and pharmaceutical research staff. The groups used the same platform but operated under different institutional constraints.

NHS analysts maintained strict governance boundaries between research and operational use. Academic teams worked through lengthy ethics and data access approvals before touching real patient records. Pharmaceutical teams had more exploratory freedom early in studies but faced strict audit and reporting obligations later.

The groups were not aggregated into a single researcher persona. Their discrepancies were treated as the primary findings. The NHS group's governance boundaries revealed requirements for clear workflow state. Academic approval timelines made the same need visible from a different angle. Pharmaceutical audit obligations revealed requirements for query history and decision traceability that the academic group's ethics constraints had implied but not fully specified.

The critical finding was that confusion concentrated at handover points between governance stages and between team members. No single group's account was complete. Triangulation across institutional positions produced the fuller picture.

World Customs Organization IPM: geographic distribution was the triangulation axis

The World Customs Organization / IPM example used triangulation across geographically and institutionally dispersed user groups in a live enforcement platform. The Sandbox Experiments phase involved frontline inspection officers, intelligence analysts, and rights holder brand protection and legal teams.

These groups had different operational relationships to the same platform. Frontline inspection officers checked shipments against alerts under time pressure in field conditions. Intelligence analysts required structured access to case histories and seizure patterns. Rights holder brand protection and legal teams needed clear paths to file information and monitor enforcement activity.

Research combined interviews and workflow mapping with WCO teams and remote observation across selected member administrations. Validation included 47 usability testing participants from Italy, Romania, Uzbekistan, Algeria, and Spain. This geographic span was not used as a statistical sampling strategy. It was used to interrogate whether design decisions that worked in one administrative and linguistic context would hold in genuinely different ones.

The discrepancies became findings. A clear status indicator in one context was ambiguous in another. Terminology normalised in European customs practice required explicit labelling in Central Asian contexts. Progressive disclosure assumptions that held for experienced officers failed for newer users in administrations with less institutional IPR enforcement history.

The result was not a design validated across five countries. It was a design stress-tested against five different operational contexts and revised where discrepancies showed fragility.

Squaremind dermatology scanning device: five sequential sources tested the evidence chain

The Squaremind dermatology scanning device example used five evidence sources sequentially, from pre-engagement context through post-redesign testing to commercial confirmation. Each source interrogated the preceding findings from a position the earlier source could not occupy.

The first source was the client's pre-engagement test: 14 users and 2 completions. This was client-reported background data, not Creative Navy-generated evidence. It established the scale of the failure and the age-stratified failure pattern, with users aged 45–65 failing faster than users aged 20–35. It did not explain which steps caused patients to get stuck or why.

The second source was 4 Creative Navy field observation sessions in France. These sessions were deliberately not structured as measurement because the existing system was too poor to measure usefully. Field observation produced situated understanding: the physical relationship between patient and screen, the spatial context of the scan, and the observable pattern of confusion as body movement before abandonment.

The third source was structured post-redesign testing in London with 12 users and Paris with 17 users, co-conducted with an independent dermatologist hired by Creative Navy. This was the primary measurement source for binary completion, timed recovery, and failure point catalogue. It interrogated the field observation findings and the pre-engagement failure data because it used the same type of users, the same device, and a different interface.

The fourth source was the independent dermatologist's clinical judgement during testing. This source interrogated the boundary between usability and clinical appropriateness. A patient could complete the scan in the testing data but still position incorrectly during sensitive body segments in the dermatologist's assessment.

The fifth source was Creative Navy-observed clinic demos after the redesign. Creative Navy directly observed 5 of the 9 clinic demos. These were not usability tests. Buyers walked through the patient experience to decide whether real patients could complete the process. All 5 observed demos resulted in purchase.

No single Squaremind source could have produced or confirmed the commercial outcome claim. The claim is credible because five sequential sources, used against each other, produced consistent findings from different positions.

Boundaries of triangulation evidence

Triangulation does not turn every finding into a confirmed fact. In Creative Navy evidence work, triangulation produces richer understanding by exposing discrepancies, mechanisms, role differences, institutional constraints, and contextual conditions.

Some triangulated examples include measured elements, such as the Torqeedo controlled experiment with 24 subjects and the Squaremind post-redesign testing in London and Paris. Other elements are Creative Navy-observed, client-reported, structured feedback, qualitative observation, benchmarking, or analysis. The evidence type matters because each source supports a different kind of claim.

Triangulation also does not mean that every method must be used in every engagement. The value is not in the number of methods but in whether the evidence sources interrogate each other productively.