Structuring the product: information architecture and the cost of finding things

A product can look finished and still be unusable, and the reason is rarely the styling: a patient cannot work out which heading hides the appointment history, a customer abandons a checkout because the next step is not where they expected it, and the interface behaves as specified while still costing the person work. A team sets that cost largely before it draws any screen, by how it organises and names the content. This unit argues that information architecture decides how much work a person does to find what they need, and the evidence on depth, breadth, information scent and category agreement decides how large that cost is.

Rosenfeld, Morville and Arango (2015) describe an information space's organisation, labelling, navigation and search systems, and argue each must match how people look for information rather than how the organisation sees itself. Wurman (1996) gave the role its name earlier. The object here is not a page but a structure that a person moves through. This is the ninth of ten units; it takes that structure and leaves unit ten to judge the product built on it.

Retrieval cost is a quantity, not a preference

The cost of finding something can be modelled. Pirolli and Card (1999) set out information foraging. In that model a person follows cues of relevance, called information scent, along available paths. The person gives up when the expected value of continuing falls below its cost. A link carries a scent that the reader estimates before clicking, and the structure decides whether a plausible path exists at all.

Two studies turn the model into an instrument. Chi, Pirolli, Chen and Pitkow (2001) built two computational methods on scent, one inferring the need behind a pattern of browsing and one predicting where a user will go, both effective in their evaluation. Fu and Pirolli (2007) took the idea into SNIF-ACT, in which users judge links in sequence and decide to click or go back, and showed it predicted their choices from the relevance of link texts.

A shallower hierarchy costs the searcher less, on one narrow task

If structure sets cost, the first decision is the shape of the hierarchy. Kiger (1984) ran the trade-off over five arrangements of a sixty-four item menu tree, and found time and errors rising with depth, with the shallowest arrangements fastest and most preferred, concluding that designers should keep the hierarchy shallow, offering eight or nine selections at a level, a figure he tied to limits on human information processing. Snowberry, Parkinson and Sisson (1983) ran the earlier version, on four menu hierarchies of sixty-four items, and found search speed and accuracy improving as breadth increased and depth decreased. Both point the same way, and both use one narrow paradigm: a named target, a screen of words and a single search task. The searcher's age changes the cost a structure imposes.

Miller (1956) argued that the span of immediate memory is about seven items, expandable by chunking. The paper concerns memory, not interfaces: it contains no experiment on menu structure, yet is cited for menu breadth more often than it can support.

The label carries the scent

Larson and Czerwinski (1998) built three versions of the same web content and set participants search tasks in each: sixteen choices leading to thirty-two, thirty-two leading to sixteen, and eight leading to eight leading to eight. The three-level version was slowest by a wide margin, at fifty-eight seconds on average, against forty-six for the thirty-two by sixteen version and thirty-six for the sixteen by thirty-two. The moderate structure won, which the authors trace to its labels, since the sixteen top-level categories named what lay behind them more distinctly than the eight of the deep structure. A hierarchy can be well shaped and still fail at its words.

Users' categories differ from the team's, and agreement is no proof

An architecture is a claim about how a population groups things, and that claim can be tested. Faiks and Hyland (2000) report a card-sort study at Cornell University Library in which users sorted concepts into their own groups that differed from the structure the design team had assumed, since card sorting replaces the designer's mental model with the users' actual one. It does not settle which grouping is best, since users can agree on a poor structure as readily as a good one. Rosenfeld, Morville and Arango (2015) leave that judgement to the designer.

Structure shapes comprehension, not only search

Architecture also reaches inside the document. Hornbaek and Frokjaer (2003) had twenty subjects read and answer questions about scientific documents in three interfaces: overview-plus-detail, a fisheye and a linear one. They found overview-plus-detail supported comprehension best, and the fisheye sped up reading at the cost of noticing less of the surrounding text, so structure changes not only how fast an answer is found but how well the material around it is understood. The result rests on a small laboratory study, so it is a mechanism rather than a general law.

Contested question: does the structure decide, or do the task and the label decide?

One position holds that structure determines retrieval cost, and that the relationship can be predicted. The foraging tradition states it most clearly. Pirolli and Card (1999) make the cost of a path a function of scent against expected value. Chi and colleagues (2001) show that scent-based models predict where users go. Fu and Pirolli (2007) show that a model can predict link choices from link text. On this reading, a well-shaped hierarchy is reliably easier to search.

The other position holds that the effect is conditional, and reads the same literature more carefully. Kiger (1984) measured the depth penalty only on a search for a known target, and the shape that suits a directed search need not suit browsing. Larson and Czerwinski (1998) traced a depth penalty to weaker labels at the top level, so a structure cannot be judged without its wording, and Faiks and Hyland (2000) found that users' categories differed from the designers'. On this reading, structure over styling is true but incomplete, and the seven-item rule often invoked in its support is a misreading of Miller (1956).

What would settle the question is a study that crossed the shape of a hierarchy with the quality of its labels in a factorial design, using realistic tasks, pre-registered measures of time to find an item and of errors, domain experts and novices, and a working product rather than an artificial menu. Kuric, Demcak and Krajcovic (2025) ran part of it: across 180 participants and 1800 task completions they compared three tree-testing variants with three high-fidelity prototype menus over the same architecture, and the variant itself changed the result, with none reproducing prototype navigation exactly, so a structure scored in the abstract is partly a score of the instrument. The label-versus-shape crossing is still missing. Until then, structure is a strong and predictable influence on cost, and its size depends on the task and the wording.

The predictive side rests on models fitted to behaviour, which forecast navigation well but were never tested against a redesigned product, while the conditional side rests on menu experiments with artificial targets and one three-condition search study, smaller but timing real retrieval. Miller's paper is weaker still as a warrant for menu breadth, since it studies memory and ran no interface experiment, yet is attached to the shallow rule more often than it can support. The strongest objection is that the unit's claim may be circular: if a structure scores well only once its labels are good, the labels rather than the shape carry the effect, and no study here separates the two. While the dispute is open, a designer tests the words on the paths that matter, and reports structure and wording as one decision.

Where the argument is heading

The first moves information architecture from a catalogue towards an interaction. Toms (2002) argues that information interaction, the way people seek, use and organise information, should frame the discipline, and offers a model joining user, content and system. The second widens the boundary of the design, since Morville (2005) argues that findability now extends beyond any single site, depending partly on search engines and links the designer does not own. The third is the maturity of the practitioner's craft, where Rosenfeld, Morville and Arango (2015) remain the standard reference, and Nielsen (1993) supplies the usability vocabulary. What none of this supplies is a test on a working product.

The sources this unit rests on

• Chi, E. H., Pirolli, P., Chen, K. and Pitkow, J. (2001). Using information scent to model user information needs and actions and the web. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI). doi:10.1145/365024.365325 They built two computational methods on information scent, to infer the need behind a browsing pattern and to predict where a user would go, and both were effective in their evaluation. It is evidence that scent-based structure predicts web navigation.
• Faiks, A. and Hyland, N. (2000). Gaining user insight: a case study illustrating the card sort technique. College and Research Libraries. doi:10.5860/crl.61.4.349 They used card sorting to gather users' own groupings of concepts for a library help system before it was designed, and found it effective at surfacing a structure the designers had not assumed. It grounds the claim that users' categories can differ from the designer's.
• Fu, W. T. and Pirolli, P. (2007). SNIF-ACT: a cognitive model of user navigation on the World Wide Web. Human-Computer Interaction. doi:10.1080/07370020701638806 They modelled users who evaluate links in sequence and decide to click or go back by a satisficing rule based on the scent of link texts, and it predicted their choices across navigation tasks. It makes structure and wording an object of prediction.
• Hornbaek, K. and Frokjaer, E. (2003). Reading patterns and usability in visualizations of electronic documents. ACM Transactions on Computer-Human Interaction. doi:10.1145/772047.772050 They had twenty subjects read scientific documents in overview-plus-detail, fisheye and linear interfaces, and found overview-plus-detail supported comprehension best while the fisheye speeded reading at the cost of noticing less. It shows that structure changes how well content is understood, not only how fast it is found.
• Kiger, J. I. (1984). The depth/breadth trade-off in the design of menu-driven user interfaces. International Journal of Man-Machine Studies. doi:10.1016/S0020-7373(84)80018-8 He ran an experiment on depth and breadth in menu trees and concluded that designers should minimise depth by offering menus of eight or nine selections, a limit tied to human information-processing capacity. It is the classic evidence that the shape of a hierarchy changes search time.
• Kuric, E., Demcak, P. and Krajcovic, M. (2025). Validation of information architecture: cross-methodological comparison of tree testing variants and prototype user testing. Information and Software Technology, 183, 107740. doi:10.1016/j.infsof.2025.107740 Across 180 participants and 1800 task completions, three tree-testing variants were compared with three high-fidelity prototype menus over the same architecture. The variant itself changed the result, and none reproduced prototype navigation exactly.
• Larson, K. and Czerwinski, M. (1998). Web page design: implications of memory, structure and scent for information retrieval. Proceedings of the SIGCHI Conference on Human Factors in Computing Systems (CHI). doi:10.1145/274644.274649 They tested how a web page's organisation and its scent affected retrieval and found that both the grouping of headings and the match between link text and the reader's goal mattered, with scent playing a large role. It supplies the qualification that structure and labelling must be judged together.
• Miller, G. A. (1956). The magical number seven, plus or minus two: some limits on our capacity for processing information. Psychological Review. doi:10.1037/h0043158 He argued that the span of immediate memory is about seven items and discussed the chunking that expands effective capacity. It is the origin of the number commonly invoked to cap menu breadth, and it is about memory rather than about interfaces.
• Morville, P. (2005). Ambient Findability. O'Reilly Media. https://archive.org/details/ambientfindabili0000morv_l3w9 He argues that findability, the quality of being locatable, is a central design concern across media, and that internal and external findability must be designed together as information escapes any single site. It is where the unit's claim meets the search engine the designer does not control.
• Nielsen, J. (1993). Usability Engineering. Morgan Kaufmann. https://archive.org/details/usabilityenginee0000niel He sets out usability as an engineering objective, with its components, its evaluation methods and the heuristics that inspection later used. It is the source of much of the module's vocabulary and the bridge to the evaluation methods in unit ten.
• Pirolli, P. and Card, S. (1999). Information foraging. Psychological Review. doi:10.1037/0033-295X.106.4.643 They model information seeking as foraging, in which people follow information scent along paths and give up when the expected value of continuing falls below its cost. It is the theory that turns the cost of finding something into a quantity a designer can reason about.
• Rosenfeld, L., Morville, P. and Arango, J. (2015). Information Architecture: For the Web and Beyond. O'Reilly Media. https://archive.org/details/louis-rosenfeld-peter-morville-jorge-arango-information-architecture-for-the-web They set out information architecture in full, covering organisation, labelling, navigation and search systems, and argue that structure must match how people look for information. It is the unit's standard practitioner text and the source of its four systems.
• Sharit, J., Hernandez, M. A., Nair, S. N., Kuhn, T. and Czaja, S. J. (2011). Health problem solving by older persons using a complex government web site. ACM Transactions on Accessible Computing, 3(3), 1-35. doi:10.1145/1952383.1952386 They found older people's performance limited by the site's structure.
• Snowberry, K., Parkinson, S. R. and Sisson, N. (1983). Computer display menus. Ergonomics. doi:10.1080/00140138308963390 They studied computer display menus and found search speed and accuracy improving as breadth increased and depth decreased, across four hierarchies of sixty-four items. It is the earlier companion to Kiger's result on the shape of a hierarchy.
• Toms, E. G. (2002). Information interaction: providing a framework for information architecture. Journal of the American Society for Information Science and Technology. doi:10.1002/asi.10094 She argues that information interaction, the way people seek, use and organise information, should frame information architecture, and proposes a model joining user, content and system. It connects the practitioner discipline to HCI research.
• Wurman, R. S. (1996). Information Architects. Graphis Press. https://archive.org/details/informationarchi0000rich He profiles information architects and argues that the organisation of information is a design discipline in its own right, concerned with making the complex clear. It is the origin of the role the unit is about.