Dashboards in organisations: instruments, governance and the cost of a wrong panel

A chart on a wall is read and forgotten. A dashboard is read on a schedule, by people who can act. The two fail differently. A chart fails when a reader misreads it. A dashboard fails when an organisation acts on it and the action is wrong, or when the watching changes what the numbers report. This ninth of ten units argues that the second failure is the one the design literature is least equipped to catch. It is usually filed under the wrong heading. A panel nobody trusts, a target met by moving the work rather than improving it, a tile that reports a well-defined number for a purpose nobody stated: these are design outcomes. They are governed, however, before they are drawn.

Every other unit here keeps the reader as the unit of analysis. This one changes it to the organisation and asks what a panel does to the people measured by it. The visualisation literature is thin exactly here: design studies say how to make a tile legible, not what happens when the tile becomes a monthly target. For that question the useful literature is performance measurement, which has spent seventy years documenting what happens when a number is installed in an organisation. The people a number governs are not the ones who built the panel, and an older nurse may meet it on a shared screen in a ward office.

The promise rests on firms, not on displays

The case for dashboards is usually made with corporate evidence, so it is worth stating what that evidence measures. Brynjolfsson, Hitt and Kim (2011) surveyed the practices and information technology investments of 179 large publicly traded firms and compared adopters of data-driven decision making with the rest. Adopters had output and productivity five to six per cent higher than expected given their other investments. The pattern also appeared in asset utilisation, return on equity and market value. Instrumental variables argued against reverse causality. That is a firm-level association between an analytics posture and performance, not a result about any panel.

Provost and Fawcett (2013) supply the caution: data science has been hard to define because it is intertwined with big data and data-driven decision making.

The field defined the dashboard late, and the design knowledge is recent

Sarikaya, Correll, Bartram, Tory and Fisher (2019) systematically reviewed the literature on dashboard use. They constructed a design space and identified major dashboard types, characterised by design goals, levels of interaction and the practices around them, and found dashboards among the most common uses of visualisation and unlike exploratory tools in context. Bach and colleagues (2022) reviewed 144 dashboards to derive eight groups of design patterns, which combine into narrative, analytical and embedded genres. Wexler, Shaffer and Cotgreave (2017) approach the same problem from practice, offering dozens of worked scenarios across industries and surfaces. Srinivasan and colleagues (2025) then widened the sample. They model a dashboard as a graph of content blocks and the relations between them, then apply it to 25,620 dashboards from Tableau Public, released as a corpus.

The clinical evidence is thin and it favours availability

A systematic review of clinical dashboards in health care is candid about its own limits. Dowding and colleagues (2015) searched seven databases for 1996 to 2012, retrieved 122 full-text papers and included eleven. They found considerable heterogeneity in setting, users and indicators. Where dashboards were easily accessible to clinicians, for instance as a screen saver, their use was associated with improved care processes and patient outcomes. Their conclusion was that higher quality studies were needed to establish efficacy. A unit on dashboards cannot claim a strong efficacy result, because the review that looked hardest for one did not find it.

A measure changes the thing it measures

Ridgway (1956) set out the founding claim of performance measurement. Quantitative measures, whatever their form, have undesirable consequences for overall performance, and even measures installed purely for information are read as definitions of which aspects of a job matter. His three cases have been repeated ever since: the single measure that narrows attention, the multiple measures that create trade-offs, and the composite index that buries the value judgements inside one authoritative number. Bouckaert and Balk (1991) recorded the shift from belief in quantified measures to awareness of their dysfunctional effects, describing thirteen diseases of productivity measurement. Bevan and Hood (2006) then showed it at scale. The English public health care system of the 2000s combined targets with what they called an element of terror, and its most visible number was the four-hour accident and emergency target. Hospital trusts met it in ways that moved the waiting rather than shortened it, such as keeping patients in ambulances until the clock had started. The regime rested on synecdoche, the taking of a part to stand for a whole, and a panel is a set of parts chosen to stand for a whole. It records neither the workaround nor its cost.

The first decision is whether the panel is for compliance or for learning

West and Blackman (2015) examined four years of work on employee performance management in the Australian Public Service. The programme had moved its focus from preventing poor performance towards achieving high performance. West stressed that the framework existed, and that implementation and accountability were now the problem. Blackman argued that there had been too much emphasis on having a system, and not enough on determining what it was for. The disagreement is the one a dashboard owner faces, and it is not about metrics. It is whether the panel is a compliance instrument or a learning instrument, and that decision changes which numbers belong on it.

A panel is made of charts, and each chart has a task

None of this removes the ordinary design problem inside the panel. Sarikaya and Gleicher (2018) surveyed the literature to catalogue scatterplot-specific analysis tasks. They examined how data characteristics influence design decisions, and surveyed scatterplot-like designs, connecting tasks, data and design to derive challenges and example best practices. They began from the observation that the traditional scatterplot fails to scale as data grows, which applies to every tile. The governance argument does not excuse the design argument. A tile chosen for visual variety rather than for a named task spends the panel's scarce attention on a question the organisation did not need answered.

The contested question: does measuring improve performance or move it?

One side holds that measurement improves performance, and its best evidence is firm-level. Brynjolfsson and colleagues (2011) found the five to six per cent productivity difference, the same pattern in other performance measures, and argued with instrumental variables against reverse causality.

The other side holds that the measured number becomes the objective and the objective stops being the point. Ridgway (1956) argued by cases drawn from mid-century industry. Bouckaert and Balk (1991) catalogued the failures. Bevan and Hood (2006) documented a national regime built on targets and terror. On that account the panel's main effect is on behaviour around the panel, and the firm-level correlation is silent about it.

A middle position, implied by Dowding and colleagues (2015) and West and Blackman (2015), is that both readings outrun the evidence. The answer depends on what the panel is attached to. A panel that informs a decision and a panel that triggers a sanction are different instruments, and the literature rarely records which one was studied.

The dispute cannot be settled by more correlational work, because the two sides measure different things at different scales. What would settle it is a study that installs the same panel under two regimes: one with a target or sanction attached, and one for information only. It would hold the data and the display fixed, record the decision that followed, and audit whether the reported behaviour changed. That separates the display from the incentive, which is the variable the field keeps leaving out. Until it exists, the honest move is to name which regime the panel is entering.

The two sides rest on different evidence. The improvement side rests on one firm-level survey of 179 companies (Brynjolfsson, Hitt and Kim, 2011). The dysfunction side rests on case studies from mid-century industry (Ridgway, 1956) and a catalogue of thirteen measurement diseases (Bouckaert and Balk, 1991). While the dispute is open, a designer should write down what the panel is for, who owns it and what consequence attaches to it before a tile is drawn. The strongest objection is that no study here installs the same panel under two regimes, so the claim that a dashboard's failures are governance failures is inferred from the measurement literature rather than measured on a panel.

Where this is heading

Bach and colleagues (2022) published their pattern groups as a shared, editable library rather than as a paper alone. Dashboard guidance is therefore moving from individual opinion to a common vocabulary with named trade-offs. Sarikaya and colleagues (2019) close their framework with research directions for dashboard design, implementation and use. Dowding and colleagues (2015) ask for the higher quality studies their review could not find. West and Blackman (2015) argue that the next gain comes from organisations stating what high performance is in their own context. The trajectory is not a better tile. It is a panel whose purpose, owner and consequence are written down, so that its failure can be attributed to the right thing.

The sources this unit rests on

Each note states what the cited work did and found and what it adds to the argument. Identifiers and metadata were checked on 2 November 2026.

• Bach, B. et al. (2022). Dashboard Design Patterns. IEEE Transactions on Visualization and Computer Graphics, 1-11. doi:10.1109/tvcg.2022.3209448 Reviewed 144 dashboards to derive eight groups of design patterns and ran a two-week workshop with 23 participants. Found the groups combine into narrative, analytical and embedded genres, and that dashboard guidance had been surprisingly thin. Design knowledge here is recent, not settled.
• Bevan, G. and Hood, C. (2006). What's Measured is What Matters: Targets and Gaming in the English Public Health Care System. Public Administration, 84(3), 517-538. doi:10.1111/j.1467-9299.2006.00600.x Examined the English governance regime of the 2000s that combined targets with what the authors called an element of terror, using evidence on reported successes, measurement problems and gaming. Argued it rests on synecdoche, taking a part to stand for a whole. Names the failure at national scale.
• Bouckaert, G. and Balk, W. (1991). Public Productivity Measurement: Diseases and Cures. Public Productivity & Management Review, 15(2), 229-235. doi:10.2307/3380763 Traced the shift from belief in quantified measures to awareness of their dysfunctional effects. Set out thirteen diseases of productivity measurement. An early statement that measurement has effects of its own.
• Brynjolfsson, E., Hitt, L. M. and Kim, H. H. (2011). Strength in Numbers: How Does Data-Driven Decisionmaking Affect Firm Performance? SSRN Electronic Journal. doi:10.2139/ssrn.1819486 Surveyed the practices and information technology investments of 179 large publicly traded firms, comparing data-driven adopters with the rest. Found adopters had output and productivity five to six per cent higher than expected, with the same pattern in other performance measures and instrumental variables against reverse causality. A firm-level result about a posture, not a display.
• Dowding, D. et al. (2015). Dashboards for improving patient care: Review of the literature. International Journal of Medical Informatics, 84(2), 87-100. doi:10.1016/j.ijmedinf.2014.10.001 Searched seven databases for 1996 to 2012, retrieved 122 full texts and included eleven. Found heterogeneity in setting, users and indicators, and improvement associated with easily accessible dashboards such as a screen saver, while concluding that higher quality studies were needed. Availability counted for more than analytic power.
• Provost, F. and Fawcett, T. (2013). Data Science and its Relationship to Big Data and Data-Driven Decision Making. Big Data, 1(1), 51-59. doi:10.1089/big.2013.1508 Argued that data science is hard to define because it is intertwined with big data and data-driven decision making. Recommended naming fundamental principles rather than fixing the boundary. Supports reading data-driven as a management posture.
• Ridgway, V. F. (1956). Dysfunctional Consequences of Performance Measurements. Administrative Science Quarterly, 1(2), 240-247. doi:10.2307/2390989 Argued, across single, multiple and composite measures, that quantitative measures have undesirable consequences for performance, and that even informational measures are read as definitions of what matters. The paper predates indexed abstracts, so this is stated from sources that quote it at page level. The founding statement of measurement dysfunction.
• Sarikaya, A. and Gleicher, M. (2018). Scatterplots: Tasks, Data, and Designs. IEEE Transactions on Visualization and Computer Graphics, 24(1), 402-412. doi:10.1109/tvcg.2017.2744184 Surveyed scatterplot-specific tasks, the influence of data characteristics on design, and scatterplot-like designs, deriving challenges and best practices. Began from the observation that the traditional scatterplot fails to scale. Each panel tile inherits the task-matching problem.
• Sarikaya, A., Correll, M., Bartram, L., Tory, M. and Fisher, D. (2019). What Do We Talk About When We Talk About Dashboards? IEEE Transactions on Visualization and Computer Graphics, 25(1), 682-692. doi:10.1109/tvcg.2018.2864903 Systematically reviewed the literature on dashboard use and constructed a design space, identifying dashboard types by design goals, levels of interaction and practice. Found dashboards among the most common uses of visualisation and unlike exploratory tools in context. The field's account of them is recent.
• Srinivasan, A., Purich, J., Correll, M., Battle, L., Setlur, V. and Crisan, A. (2025). From Dashboard Zoo to Census: A Case Study With Tableau Public. IEEE Transactions on Visualization and Computer Graphics, 31(9), 6085-6099. doi:10.1109/tvcg.2024.3490259 Models a dashboard as a graph of content blocks and relations, then applies it to 25,620 dashboards from Tableau Public.
• West, D. and Blackman, D. (2015). Performance Management in the Public Sector. Australian Journal of Public Administration, 74(1), 73-81. doi:10.1111/1467-8500.12130 Reported four years of work on employee performance management in the Australian Public Service, refocused from preventing poor performance to achieving high performance. West stressed implementation; Blackman argued there had been too much emphasis on having a system and too little on what it is for. Frames the first decision as purpose.
• Wexler, S., Shaffer, J. and Cotgreave, A. (2017). The Big Book of Dashboards: Visualizing Your Data Using Real-World Business Scenarios. Wiley. doi:10.1002/9781119283089 A practitioner reference of dozens of worked dashboard scenarios across industries and across print, desktop, tablet, smartphone and conference-room display, with a section on practical and psychological factors. Argues from experience, and is cited here as evidence of what practitioners are told.