Method

A journey map records intent, not action

Journey maps organise what a person set out to do. Task analysis describes the work that determines whether they arrive, and it is the level at which design decisions can be defended.

Two people may occupy the same stage of the same journey map, pursuing the same goal, and only one of them completes it. The map is not wrong about either. It is recording something other than the work.

Two transfers

A customer wishes to move money between accounts. On a journey map this is a single stage, with a goal, a channel and an emotional register attached to it.

Two customers reach that stage. The first selects a recognised payee from a stored list and confirms. The second copies an account number from an email, holds it while switching application, and types it into a field that offers no check against the source. Both intend the same outcome. The second is performing a transcription task under a memory load, against a string with no internal redundancy, with no feedback that would reveal a single transposed digit before the transfer completes.

The journey is identical. The work is not, and the difference in work is the whole of the difference in risk. A model that records the first fact and omits the second will not distinguish the two cases, because the distinction does not exist at the level the model operates on.

Mechanism

What a journey map provides is a narrative account of movement through a service. It describes goals, channels, perceptions and points of friction, and it does so in terms that marketing, service design and product teams can hold in common. That shared legibility is the reason it has become the default description of user behaviour in digital product design.

What it does not provide is a mechanism. A map may show that users abandon a process partway through completion. Several explanations remain consistent with that observation. The task may require information that is unavailable at the point of need. The interface may permit an action to be taken in the wrong mode. Workload may rise sharply at one step. Feedback after a completed action may be weak enough that users repeat it. The sequence may impose a memory demand that the design could have removed.

Each of these suggests a different intervention, and none of them can be distinguished from the map alone. The map records that abandonment occurred. It does not contain the evidence that would explain why, because abandonment is an outcome and design operates on causes.

Proportion

None of this makes journey mapping a weak method. It is frequently the correct tool, and the questions it answers are real ones: where a service begins and ends, which part of an organisation owns which touchpoint, what customers expect at each, where the boundaries between channels fall. Those questions are not answerable from a task model, and a programme that attempts it will produce a task decomposition nobody outside the analysis can use.

Every model is a selection. The journey map selects intent and sets aside the work that intent requires. For a large share of product decisions that omission is tolerable, and the map is sufficient on its own. For decisions involving error, reliability, workload or performance under time pressure, the omitted part is the part that governs the outcome.

The choice is therefore not between the two methods. Human performance occurs at the level of tasks whether or not tasks are modelled, and the work exists either way.

Hierarchical task analysis

The methods that do model work were developed for practical rather than theoretical reasons. Military, industrial and transport systems required a means of anticipating performance before failures occurred, because waiting for accidents to expose design weaknesses was expensive and, in those settings, unacceptable.

Hierarchical Task Analysis, set out by Annett and Duncan in 1967, decomposes a goal into sub-goals and the plans that govern the order in which they are pursued. The decomposition continues until further detail would not change what the analyst decides, a stopping rule the original authors framed in terms of the cost of failure against the probability of inadequate performance. The output is not documentation. It is a structure against which the information required at each step, the decisions that depend on that information, and the actions that follow can be examined individually.

Stanton's review of four decades of application found the method durable across domains well outside the industrial settings it came from, and attributed that durability to its being a framework rather than a fixed procedure. The underlying problem has proved equally durable. A banking application is less physically dangerous than a railway signalling system and remains dependent on human action, and the categories of failure that action makes possible are the ones Norman described in 1981 and Reason set out in 1990: capture errors, mode errors, omissions, actions correctly performed on the wrong object. The consequences differ by orders of magnitude. The error forms do not.

Prediction

The more consequential property of a task model is that it supports prediction, and this is where the two methods diverge most sharply in what they can be used for.

Journey maps are assembled from observation. They organise what has already been seen, after users have interacted with a system that already exists. A task model can be built against a design that has not yet been released, and read for the failures the design permits. Where a task requires visual comparison of two similar values, transcription error is foreseeable. Where a workflow requires an operator to carry information across several screens, a memory failure is foreseeable. Where two controls are functionally equivalent in appearance and divergent in behaviour, mode error is foreseeable. No incident report is required to reach any of these conclusions.

This is also why task-based methods persist in sectors that must defend their reasoning to someone else. ISO 14971 requires manufacturers of medical devices to identify reasonably foreseeable sequences of events leading to hazardous situations, including those arising from use, which is a requirement that cannot be discharged by reporting that users found something confusing. ISO 9241-210 places the tasks users perform within the context of use that human-centred design is required to specify and to evaluate against. In both cases the obligation is to show a chain of reasoning from a property of the design to a failure it makes possible, and a task model is what carries that chain.

That obligation is no longer confined to regulated industries. Digital products now perform functions that affect money, health, mobility and access to public services, and the evidentiary burden is arriving with them.

Design decisions

The practical implication is narrow. Where a journey map identifies a stage at which something goes wrong, the useful next question concerns the work being performed at that point rather than the emotional narrative surrounding it: what actions the stage requires, what information supports them, which errors the design permits, and which features of the interface create or remove those errors.

Those questions move a team from describing an experience to accounting for it. A journey map records where somebody intended to go. A task analysis describes what they had to do to get there, and only one of the two can be used to argue that a design is sound.

Sources

Annett, J. and Duncan, K. D. (1967). Task analysis and training design. Occupational Psychology, 41, 211–221. The account of hierarchical decomposition and of the stopping rule is paraphrased from the authors' original formulation.

Stanton, N. A. (2006). Hierarchical task analysis: developments, applications, and extensions. Applied Ergonomics, 37(1), 55–79. The characterisation of HTA as a durable framework rather than a fixed procedure is paraphrased from Stanton's review.

Norman, D. A. (1981). Categorization of action slips. Psychological Review, 88(1), 1–15. Capture and mode errors are referred to in the sense established here.

Reason, J. (1990). Human Error. Cambridge University Press. The distinction between error forms and their consequences follows Reason's treatment.

ISO 14971:2019, Medical devices — Application of risk management to medical devices, clause 5.4 (identification of hazards and hazardous situations), which requires consideration of reasonably foreseeable sequences of events including those arising from use.

ISO 9241-210:2019, Ergonomics of human-system interaction — Part 210: Human-centred design for interactive systems, clause 7.2 (understanding and specifying the context of use), which includes the tasks users perform.

Gibbons, S. (2018). Journey Mapping 101. Nielsen Norman Group. The description of what a journey map sets out to record follows this account.

Commentary comparing the scope of the two methods is analytical. No empirical performance claim is made beyond the cited literature.

← All writing