Modelling the work before designing for it: task analysis, scenarios and storyboards

Design work begins before anything is drawn. A team changing how a clinic books appointments must first say what the work consists of now and afterwards. Three instruments do that job: task analysis, scenarios and storyboards. Each turns a claim about future use into something arguable before it is built, and each is a description made for a purpose.

The three answer different questions. A task analysis asks what the work is made of, and where a design changes it. A scenario asks what it would be like for one named person to do that work. A storyboard asks what can be shown, so that someone understands the proposal and disagrees early. The other two keep the claim open. This is the third of ten units; it takes a constructed brief and hands unit four a claim about use to draw.

The work is the first thing to model

The engineering model of interaction begins with the task. Card, Moran and Newell (1983) set out the model human processor and the GOMS framework, in which a designer describes and times routine skilled behaviour from a small set of operators, and claimed that a person's work can be written down precisely enough to predict how long it takes.

Annett (2003) gives the canonical method. Hierarchical task analysis decomposes a task's goals into sub-goals and operations, each operation specified by its goal, the conditions that trigger it, the actions it takes and the feedback showing the goal was met, with plans governing their order. His chapter sets it as an exercise on a small domestic task, such as making a cup of tea. The result is a description a designer can check against observation, and the decomposition makes a design's assumptions explicit: anyone can see which step the design changes and which it leaves alone.

Diaper and Stanton (2004) compare the task analysis approaches for human-computer interaction by their assumptions and outputs, and argue that a task model is a description made for a purpose. Kirwan and Ainsworth (1992) give the procedural version.

A task model can be right about the future

John and Kieras (1996) compare the GOMS family and show how each member predicts aspects of performance from goals, operators, methods and selection rules, giving useful quantitative predictions at low cost for routinised tasks.

Gray, John and Atwood (1993) tested it. In Project Ernestine they built a CPM-GOMS model of telephone operators, compared a current workstation with one sold as faster and more ergonomic, and predicted the new one would be slower. A four-month field trial of 72,450 calls confirmed it, the model predicting 0.63 seconds per call slower against 0.65 in the trial, almost $2 million a year in cost, because the redesign disrupted the overlapped parallel activities experienced operators had built up. This is the unit's best evidence that a model of the work can make a claim about future use testable before anyone builds the system.

What a task analysis cannot see

Stanton (2006) reviews more than thirty years of hierarchical task analysis. He finds it durable, because it describes work at a level that supports design and analysis. He finds it weaker for novel or creative tasks, where no stable sub-goal hierarchy exists to decompose.

Bainbridge (1983) shows why that matters beyond novel tasks. Automating a process removes the operator's routine experience. It leaves them responsible for the abnormal case, which they are least practised to handle. The human's role becomes harder rather than easier. Modelling the work is how a design's removals become visible.

Vicente (1999) offers the alternative. Cognitive work analysis begins with the constraints of the work domain, not with the task. It argues that design should support the worker's control of those constraints, not a fixed sequence of actions.

The instruments assume a reader: Medhi and colleagues (2011) found low-literacy users needed graphical flows.

Scenarios keep the claim arguable

If a task model is an argument, it needs a form people can argue with. Carroll (2000) argues that scenarios are the appropriate unit of design reasoning: a concrete story of use that people can discuss before it hardens into a requirement. A scenario names a person, a situation and a motive, walks them through the work, and its distance from design detail delays commitment.

Rosson and Carroll (2002) turn the device into a method, expressing requirements analysis as problem scenarios and design as activity and interaction scenarios.

Storyboards make the claim visible

The storyboard is the scenario drawn. Truong, Hayes and Abowd (2006) analysed storyboards by novices and professional designers, testing five elements against readers: text, people, level of detail, panels and time. Text was decisive: about 84 per cent of readings identified the story when a panel carried words, against 66 per cent when it did not, and short captions beat long passages. Panel counts had an optimum of three to five, with comprehension falling below three and both comprehension and enjoyment above five, while extra detail did not help. The guidelines were then used in an undergraduate class, so the form of a storyboard changes whether its claim is understood and it is not a neutral container.

Van der Lelie (2006) gives the practitioner's case. Storyboards let a reader experience an interaction by empathising, and reflect on it from their own expertise. They give a team a common ground for surfacing disagreement before building.

Contested question: is the point description or communication?

The literature divides on what these instruments are for. One position holds that a model's value lies in what it can predict: Card, Moran and Newell (1983), John and Kieras (1996) and Gray, John and Atwood (1993) treat descriptive accuracy as the goal, and the 0.63 against 0.65 second prediction is their standard of success.

The other holds that a model's value lies in what it lets people do together. Carroll (2000), Rosson and Carroll (2002) and van der Lelie (2006) treat the claim about use as something to be kept open and shared. Diaper and Stanton (2004) put both in one frame: if a task model is a description made for a purpose, prediction is one purpose among several, and a model that predicts well but that nobody can argue with has failed at what the studio asks.

The positions diverge once they must trade off: is more descriptive completeness what makes a decision better, or what makes the model unusable by those who must argue with it? A task hierarchy can be complete and unreadable, a storyboard readable and silent about half the work, and neither tradition has a study measuring both halves, since Gray, John and Atwood (1993) measure prediction and never use. A study would bear on the dispute if it held the work constant, produced two models of different completeness, and measured both the decision and whether the team used the model.

A second disagreement sits underneath. Vicente (1999) and Stanton (2006) agree that a task analysis struggles when the work cannot be prescribed, but Vicente would move the unit of analysis to the constraints of the work domain while Stanton would keep the hierarchy and accept its limits. Settling this would require showing, on the same complex system, which representation led designers to the better decision about what the human should control.

The predictive side is a family of models and one field trial of 72,450 calls, strong for a routinised task and silent about any other; the communication side is method books and a practitioner's case, weaker by kind. The strongest objection is that neither tradition shows its instrument improving the decision it is meant to inform. While the dispute is open, a designer lets the question choose the instrument, and records which question a model was built to answer before trusting its output.

Where the argument is heading

Bainbridge (1983) and Vicente (1999) both describe a world in which machines increasingly do the routine work that task analysis models, leaving the human the abnormal case that no stable sub-goal hierarchy can describe. Diaper and Stanton (2004) give the reason to expect the instruments to change rather than disappear: if a task model is a description made for a purpose, then as the purpose changes the model must change with it. Carroll and colleagues (2026) reviewed what became of the scenario half of this unit, and across 125 SIGCHI papers found it mature and cumulative, combined with other methods in nearly half of the cases.

The work this unit points to and does not have is a way of making a claim about future use arguable. That claim must stand when the work cannot be prescribed in advance, and when those who must argue with it did not build the model.

The sources this unit rests on

• Annett, J. (2003). Hierarchical task analysis. The Handbook of Cognitive Task Design, Lawrence Erlbaum. doi:10.1201/9781410607775.ch2 Annett sets out hierarchical task analysis, which decomposes a task's goals into sub-goals and operations with plans governing their order, producing a description that can be checked against observation. The decomposition, he argues, makes a design's assumptions explicit, which is why this unit treats the method as its first instrument.
• Bainbridge, L. (1983). Ironies of automation. Automatica. doi:10.1016/0005-1098(83)90046-8 Bainbridge argues that automating a process removes the operator's routine experience while leaving them responsible for the abnormal case, which they are least practised to handle, so the human's role becomes harder rather than easier. Modelling the work is how a design's removals become visible.
• Card, S. K., Moran, T. P. and Newell, A. (1983). The Psychology of Human-Computer Interaction. Lawrence Erlbaum. https://archive.org/details/psychologyofhuma00stua Card, Moran and Newell set out the model human processor and the GOMS framework, arguing that routine skilled behaviour can be described and timed from a small set of operators. The book founded the engineering model of interaction on the structure of the user's task.
• Carroll, J. M. (2000). Making Use: Scenario-Based Design of Human-Computer Interactions. MIT Press. https://archive.org/details/makingusescenari0000carr Carroll argues that scenarios, concrete stories of use, are the appropriate unit of design reasoning because they can be argued about before they harden into requirements. Scenario-based design is the unit's central instrument for keeping a claim about use open.
• Carroll, J. M., Jo, J., Kim, J., Lin, Y.-F. and Chen, E. (2026). What Happened to Scenario-Based Design in HCI? A Scoping Review. Proceedings of the 2026 CHI Conference on Human Factors in Computing Systems. doi:10.1145/3772318.3790899 A scoping review of 125 SIGCHI papers from 2015 to 2025 that practise scenario-based design found scenario-based design mature and cumulative, and extended by speculative design and ethics.
• Diaper, D. and Stanton, N. (2004). The Handbook of Task Analysis for Human-Computer Interaction. Lawrence Erlbaum. doi:10.1201/b12470 Diaper and Stanton compare the major task analysis approaches for human-computer interaction by their assumptions and outputs, arguing that task models are descriptions made for a purpose rather than neutral pictures of work. It gives the unit's claim that a task analysis is an argument, not a record.
• Gray, W. D., John, B. E. and Atwood, M. E. (1993). Project Ernestine: validating a GOMS model through the analysis of a real-world task. Human-Computer Interaction. doi:10.1207/s15327051hci0803_3 Gray, John and Atwood built a CPM-GOMS model of telephone operators and predicted that a proposed workstation would be slower than the current one. A four-month field trial of 72,450 calls confirmed it, the weighted model predicting 0.63 seconds and the trial finding 0.65, almost $2 million a year in cost.
• John, B. E. and Kieras, D. E. (1996). The GOMS family of user interface analysis techniques: comparison and contrast. ACM Transactions on Computer-Human Interaction. doi:10.1145/235833.236054 John and Kieras compare the GOMS family and show how each model predicts aspects of performance from goals, operators, methods and selection rules, arguing that the family gives useful quantitative predictions at low cost for routinised tasks. Its own qualifier bounds the unit's claim.
• Kirwan, B. and Ainsworth, L. K. (1992). A Guide to Task Analysis. Taylor and Francis. doi:10.1201/b16826 Kirwan and Ainsworth assemble task analysis methods for the design of work systems and set out how to choose among them according to purpose, from allocation of function to interface specification. Selecting a method is an explicit design decision.
• Medhi, I., Patnaik, S., Brunskill, E., Gautama, S. N. N., Thies, W. and Toyama, K. (2011). Designing mobile interfaces for novice and low-literacy users. ACM Transactions on Computer-Human Interaction, 18(1), 1-28. doi:10.1145/1959022.1959024 They found low-literacy users needed step-by-step graphical flows instead of text menus.
• Rosson, M. B. and Carroll, J. M. (2002). Usability Engineering: Scenario-Based Development of Human-Computer Interaction. Morgan Kaufmann. https://archive.org/details/usabilityenginee0000ross Rosson and Carroll present usability engineering as scenario-based development, in which requirements analysis is expressed as problem scenarios and design as activity and interaction scenarios. It turns a scenario into a development artefact.
• Stanton, N. A. (2006). Hierarchical task analysis: developments, applications, and extensions. Applied Ergonomics. doi:10.1016/j.apergo.2005.06.003 Stanton reviews more than thirty years of hierarchical task analysis, arguing it has proved durable because it describes work at a level that supports both design and analysis, while noting its limits for novel or creative tasks. It is the unit's honest bound on the method.
• Truong, K. N., Hayes, G. R. and Abowd, G. D. (2006). Storyboarding: an empirical determination of best practices and effective guidelines. Proceedings of the 6th ACM Conference on Designing Interactive Systems (DIS). doi:10.1145/1142405.1142410 Truong, Hayes and Abowd ran two formative studies to identify a storyboard's elements, including text, people, level of detail, panels and the representation of time, and an empirical study of their effects on understanding and enjoyment. It is the empirical half of the unit's storyboard claim.
• Vicente, K. J. (1999). Cognitive Work Analysis: Toward Safe, Productive, and Healthy Computer-Based Work. Lawrence Erlbaum. doi:10.1201/b12457 Vicente sets out cognitive work analysis, which begins with the constraints of the work domain rather than the task, and argues that design should support the worker's control of those constraints rather than prescribe a sequence of actions. It is the alternative when work cannot be prescribed in advance.
• van der Lelie, C. (2006). The value of storyboards in the product design process. Personal and Ubiquitous Computing. doi:10.1007/s00779-005-0026-7 Van der Lelie argues from practice that storyboards let a reader both experience an interaction by empathising and reflect on it from their own expertise, and give a team a common ground for surfacing disagreement. The account is a practitioner's case rather than a study.