Data Communication, Decision Support and Managerial Applications

Introduction

Data science creates value only when evidence is converted into understanding and action. A technically accurate model, query, or analysis may have little organisational value if decision-makers cannot interpret it, do not trust its data, or cannot identify what action should follow. Data communication is therefore not an ornamental final step; it is part of the analytical process.

This chapter examines how data professionals communicate evidence through dashboards, reports, and data stories. It also explains how analytical outputs are connected to key performance indicators (KPIs), operational decisions, recommendations, uncertainty, and limitations. The emphasis is on practical professional communication for technical and managerial audiences.

A successful communication product answers five questions:

1. What is happening?
2. Why is it happening?
3. How certain is that conclusion?
4. Why does it matter to the organisation?
5. What should be done next?

Learning Objectives

By the end of this chapter, students should be able to:

• distinguish dashboards, analytical reports, and data stories;
• select appropriate visualisations for different data types and audiences;
• design KPIs that are measurable, relevant, and resistant to misleading interpretation;
• connect evidence to operational and managerial decisions;
• communicate uncertainty, assumptions, data quality, and analytical limitations;
• formulate practical recommendations without overstating the evidence;
• evaluate ethical, governance, accessibility, and privacy issues in data communication; and
• develop a coherent dashboard or briefing for a technical or managerial audience.

1. Data Communication as a Decision Process

Data communication is the deliberate transformation of data and analysis into a form that enables a particular audience to understand a situation and make an informed decision. It includes selection, organisation, visual encoding, explanation, interpretation, and recommendation.

Communication is not the same as simply displaying data. A spreadsheet may contain more values than a dashboard, but it may be less useful if users cannot identify the important pattern. Communication requires purpose. The same dataset may support different products:

The communicator should identify the decision before selecting a chart. For example, a manager deciding whether to increase staffing needs information about demand, service capacity, waiting time, cost, and forecast uncertainty. A chart showing annual revenue alone is insufficient.

Descriptive, Diagnostic, Predictive, and Prescriptive Communication

Analytical communication commonly operates at four levels:

• Descriptive: What happened? Examples include monthly sales and average waiting time.
• Diagnostic: Why did it happen? Examples include regional differences, process bottlenecks, or changes in customer mix.
• Predictive: What is likely to happen? Examples include a demand forecast or probability of customer churn.
• Prescriptive: What action is advisable? Examples include reallocating staff or changing an inventory reorder point.

These levels should not be confused. A forecast is not a fact, and a recommendation is not automatically justified by a prediction. Each transition requires assumptions and evidence.

2. Audience, Purpose, and the Communication Brief

Before building a report or dashboard, create a short communication brief. It should state:

• the decision or question;
• the intended audience and its level of technical knowledge;
• the time horizon;
• the relevant KPIs;
• the action that may result;
• the evidence available;
• known data limitations; and
• the preferred delivery format and frequency.

A useful distinction is between the primary audience, who will make or authorise the decision, and secondary audiences, who may implement, audit, or challenge it. Managers generally need concise interpretation, implications, and options. Technical audiences need definitions, methods, data lineage, and validation details. A professional product can serve both by using layers: a clear headline view, expandable detail, and an attached methodology note.

Choosing the Level of Detail

Too little detail produces unsupported conclusions; too much detail obscures the decision. A practical structure is:

1. Headline: the main finding in one sentence.
2. Evidence: the most relevant metric or visual.
3. Interpretation: what the evidence means and does not mean.
4. Implication: why the result matters.
5. Action: the recommended next step, owner, and timing.

This structure is often called an evidence-to-action chain. It prevents a common failure in analytics: reporting numbers without explaining their operational significance.

3. Dashboards, Reports, and Data Stories

Dashboards

A dashboard is an interactive or static visual interface that monitors selected measures. It is most useful when users need repeated access to current information. A dashboard should support scanning, comparison, filtering, and exception detection.

A well-designed dashboard normally includes:

• a clearly stated purpose;
• a limited set of priority KPIs;
• consistent definitions and time periods;
• visible targets or thresholds;
• trends rather than isolated values;
• filters that answer meaningful questions;
• annotation of unusual events; and
• a visible refresh date and data-quality status.

A dashboard is not a collection of every available chart. Excessive visual elements create cognitive load and encourage users to focus on attractive but irrelevant measures.

Reports

A report provides a more stable and detailed account of evidence, methods, interpretation, and recommendations. Reports are appropriate for governance meetings, audits, project evaluation, and decisions requiring a documented rationale. A professional analytical report usually contains an executive summary, context, data and methods, findings, limitations, recommendations, and appendices.

Data Stories

A data story combines evidence with explanatory context and a logical sequence. It does not mean manipulating data to create an entertaining narrative. A responsible data story makes the reasoning visible:

• establish the business or social context;
• introduce the question;
• show the most important pattern;
• explain relevant causes or comparisons;
• acknowledge uncertainty and alternative interpretations; and
• conclude with proportionate actions.

A story may use a sequence of visuals rather than one crowded dashboard. It should preserve the distinction between observation, interpretation, and recommendation.

4. Selecting and Designing Visualisations

Visualisation converts values into visual features such as position, length, colour, size, and shape. For quantitative comparison, position on a common scale and length are generally easier to interpret than area, angle, or three-dimensional perspective.

Common Design Principles

Start with a meaningful comparison. A value of 8,000 has little meaning without a time period, target, denominator, or benchmark. Rates are often more informative than counts when group sizes differ. For a rate per 1,000 people:

The numerator is the number of events, and the denominator is the population exposed to the possibility of those events. The multiplier is selected for readability. The numerator and denominator must refer to compatible places and periods.

Use honest axes. Truncating a bar-chart axis can exaggerate differences, especially when bars are compared by length. A truncated axis may be acceptable for some line charts when the scale is clearly labelled, but the choice must not distort interpretation.

Use colour purposefully. Colour may distinguish categories, indicate a sequential magnitude, or signal a meaningful status. Red should not be used merely for decoration because it can imply danger. Use colour-blind-accessible palettes and redundant labels or symbols.

Show denominators and sample sizes. Percentages without counts can be misleading. A change from 1% to 2% is a 1 percentage-point increase but a 100% relative increase. Both statements are mathematically true, but they communicate different impressions.

Avoid decoration. Three-dimensional bars, excessive gradients, unnecessary icons, and ornamental backgrounds reduce accuracy and accessibility. The visual should make comparison easier, not compete with the evidence.

A Note on Correlation and Causation

A scatter plot may show that two variables move together, but association alone does not establish that one causes the other. A third variable, reverse causality, selection bias, or coincidence may explain the pattern. A data story should use causal language only when the design and assumptions support it.

5. KPIs and Performance Measurement

A KPI is a measure selected because it represents progress toward an important objective. Not every metric is a KPI. A KPI must have a clear definition, owner, data source, frequency, target or reference point, and decision use.

A KPI specification should include:

KPI Formulas

For example, customer retention rate may be defined as:

This formula assumes that customers can be consistently identified and that the period, customer definition, and inclusion rules are stable. It should not be used when customer records are incomplete or when “active customer” changes without documentation.

A service-level KPI might be:

The denominator excludes cases that are not eligible under the stated policy. Excluding difficult cases without a valid operational reason can artificially improve the KPI.

Leading and Lagging Indicators

A lagging indicator measures an outcome after it occurs, such as quarterly profit or customer complaints. A leading indicator is associated with future performance, such as preventive-maintenance completion or staff training. Leading indicators can support early intervention, but they are not guarantees of future outcomes.

KPI Failure Modes

KPIs can produce harmful behaviour when they are poorly designed. Employees may optimise the measured target while neglecting unmeasured quality. For example, rewarding call-centre speed alone may increase premature call termination. This is known as Goodhart’s law in practical form: when a measure becomes a target, it may cease to be a good measure.

A balanced measurement system combines dimensions such as quality, speed, cost, safety, equity, and customer experience. Targets should be reviewed for unintended consequences.

6. From Evidence to Operational Decisions

Decision support connects a finding to a choice among alternatives. The connection should identify the decision owner, available options, constraints, expected consequences, and monitoring plan.

A practical decision template is:

1. Situation: What is the current condition?
2. Evidence: What data supports the diagnosis?
3. Options: What feasible actions are available?
4. Trade-offs: What costs, risks, and benefits accompany each option?
5. Recommendation: Which action is preferred and why?
6. Control: What will be monitored, by whom, and when?

Thresholds and Alerts

Operational dashboards often use thresholds. A threshold should be linked to a policy, service standard, risk tolerance, or statistically meaningful signal. Arbitrary traffic-light colours can create false precision. A threshold crossing does not always mean that an intervention is required; users should check data freshness, sample size, seasonality, and exceptional events.

For example, a warehouse manager may investigate a stock-out rate above its agreed service limit. The dashboard should show the rate, number of affected orders, product categories, trend, and possible causes. It should not imply that every stock-out has the same cause.

Decision Trees and Options

A recommendation should distinguish evidence from judgement. Evidence may show that demand is increasing. Judgement is required to decide whether to add permanent staff, use temporary staff, redesign the process, or accept longer waiting times. A decision matrix can make trade-offs explicit by listing options against cost, feasibility, risk, customer impact, and expected benefit.

7. Communicating Uncertainty and Limitations

Uncertainty is not a weakness to hide. It is information about how confidently a result can support a decision. Sources include sampling variation, measurement error, missing data, model assumptions, data drift, changing behaviour, and uncertainty about future conditions.

Describing Uncertainty Clearly

Use plain language such as:

• “The estimate is based on a small sample and should be treated as indicative.”
• “The forecast assumes that recent seasonal patterns continue.”
• “The association is consistent with, but does not prove, an effect.”
• “Results may not generalise to locations not represented in the data.”

Where appropriate, uncertainty can be represented with an interval. A confidence interval describes a procedure that, under repeated sampling and its assumptions, would capture the relevant population parameter at a stated long-run rate. It is not strictly correct to say that a particular 95% confidence interval gives a 95% probability that the fixed parameter lies inside it.

For managerial communication, the technical interval should be accompanied by an interpretation and decision consequence. For example: “The estimated reduction is 6 percentage points, with an interval from 2 to 10 points; the direction is favourable, but the size remains uncertain.”

Limitations Checklist

A limitations section should address:

• data completeness and accuracy;
• representativeness and possible selection bias;
• definitions and changes over time;
• missing or imputed values;
• confounding variables;
• model performance and validation;
• forecast horizon and drift;
• privacy and access restrictions; and
• what the analysis cannot establish.

Limitations should be specific. “The data may have limitations” is not useful. “Fourteen percent of records lack a verified location, so regional comparisons may be biased” is actionable.

8. Worked Example: Turning a KPI into a Recommendation

A delivery company monitors late deliveries. In one month, 1,200 of 10,000 eligible deliveries were late. The late-delivery rate is:

The target is 8%. The previous month had 900 late deliveries from 10,000 eligible deliveries, giving 9%. The rate therefore increased by 3 percentage points, from 9% to 12%, rather than “increasing by 3%.” The relative increase is:

Further analysis shows that 700 of the 1,200 late deliveries occurred in one region, where 4,000 deliveries were made. The regional rate is 17.5%. The other regions had 500 late deliveries among 6,000 deliveries, or approximately 8.3%.

The evidence supports an investigation focused on the region, but it does not by itself identify the cause. Possible causes include weather, route design, staffing, vehicle availability, or a change in customer delivery windows. A responsible recommendation is therefore:

Conduct a two-week regional process review, examine route and staffing records, and test a temporary route adjustment. Monitor late-delivery rate, delivery volume, cost per delivery, and customer complaints before deciding whether to make the change permanent.

This recommendation is stronger than immediately purchasing vehicles because it is proportionate to the evidence, creates a learning period, and specifies follow-up measures.

9. Case Study: Illustrative Hospital Outpatient Dashboard

Background

The following is a fictional illustrative case. A hospital outpatient department wants to reduce patient waiting time without reducing clinical quality. Managers receive a monthly spreadsheet containing appointment volume, arrival time, consultation start time, cancellation, staffing, and patient-feedback data.

Problem

The department reports an average waiting time of 42 minutes against a target of 30 minutes. Senior managers want to know whether additional staff are required. The average alone does not show whether the problem affects all patients or is concentrated in particular clinics and periods.

Analysis

The data team defines waiting time as consultation start time minus recorded patient arrival time. It checks that both timestamps use the same time zone and excludes appointments cancelled before arrival. The team reports the median as well as the mean because a small number of extreme waits may influence the average. It also reports the 90th percentile to describe long waits.

The dashboard includes:

• total eligible appointments and data completeness;
• median and 90th-percentile waiting time;
• waiting-time trend by week;
• clinic and time-of-day comparisons;
• appointment volume and staffing level;
• cancellation and no-show rates; and
• patient-experience results.

The analysis finds that the overall mean is 42 minutes, but the median is 29 minutes and the 90th percentile is 78 minutes. Long waits are concentrated on Monday mornings in one clinic. That clinic also has a higher proportion of complex appointments. This is important: a staffing decision based only on the mean could overlook concentration and case mix.

Approach and Solution

The department pilots staggered start times and assigns an additional staff member during the Monday peak for four weeks. The dashboard is revised to show waiting-time distribution, case mix, staffing, and patient experience together. A data-quality note explains that arrival times are manually entered and may be delayed.

Outcome

After the pilot, the Monday 90th-percentile waiting time falls from 96 to 70 minutes, while the median changes little. Patient complaints decline, but the evidence is not sufficient to claim that staffing alone caused the improvement because the pilot coincided with a change in appointment scheduling. Managers decide to continue the arrangement temporarily and conduct a controlled comparison with another clinic where feasible.

Learning Points

• An average can conceal operationally important extremes.
• KPI definitions must be agreed before comparison.
• A dashboard should support diagnosis, not merely display status.
• Recommendations should account for case mix and data quality.
• A positive change after an intervention is not automatically causal evidence.
• Monitoring must include quality and equity, not only speed.

10. Technical, Ethical, and Governance Considerations

Reproducibility and Data Lineage

Every published KPI should be traceable to a source, transformation, calculation, and refresh time. Version-controlled definitions and documented SQL or analytical code reduce disputes and enable correction. If a dashboard changes its definition, the change should be announced and historical comparisons reviewed.

Privacy and Confidentiality

Dashboards can expose personal or sensitive information through direct identifiers, small groups, or combinations of variables. Apply data minimisation, role-based access, aggregation, suppression of very small cells where appropriate, and secure transmission. A manager may need a clinic-level rate, not a patient-level record.

Fairness and Representation

A performance dashboard can reinforce inequality if some groups are missing, measured differently, or affected by unequal access. Disaggregate important outcomes where lawful and ethically justified, while avoiding re-identification. Examine whether a KPI rewards outcomes that are easier to achieve for already advantaged groups.

Accessibility

Use readable typography, sufficient contrast, descriptive titles, direct labels, keyboard-accessible controls, and alternatives to colour alone. A chart should have a text explanation for users who cannot access the visual. Accessibility is part of analytical quality because inaccessible evidence cannot support inclusive decisions.

Artificial Intelligence in Communication

Generative AI can assist with draft summaries, chart descriptions, or audience-specific wording, but it should not be treated as an authority. Human reviewers must verify numbers, definitions, sources, confidentiality, and claims. AI-generated explanations may invent causes or express unjustified certainty. Sensitive data should not be entered into an unapproved system.

11. Best-Practice Workflow

A disciplined workflow is:

1. Define the decision and audience.
2. Confirm KPI definitions, data scope, and refresh status.
3. Inspect quality, missingness, outliers, and changes in collection.
4. Select comparisons and visual forms that match the question.
5. Build a clear headline and evidence-to-action sequence.
6. Add targets, denominators, uncertainty, and limitations.
7. Test the product with a representative user.
8. Check accessibility, privacy, and possible misinterpretation.
9. Document assumptions, sources, and version information.
10. Monitor whether the communication actually improves decisions.

A final review should ask: Could a reasonable reader draw a different conclusion? If so, the visual or explanation may require additional context.

Chapter Summary

Data communication converts analytical evidence into understanding and action. Dashboards support repeated monitoring; reports provide documented depth; and data stories organise evidence into an explanatory sequence. Effective communication begins with the audience and decision, not with a preferred chart type.

KPIs require precise definitions, appropriate denominators, targets, ownership, refresh information, and limitations. Visualisations should make comparison accurate, use colour and axes responsibly, and show relevant context such as time, sample size, and benchmarks. Evidence should be connected to decisions through explicit options, trade-offs, recommendations, and monitoring plans.

Uncertainty, data quality, and limitations must be communicated rather than hidden. A responsible recommendation distinguishes observation from interpretation and judgement. Privacy, fairness, accessibility, reproducibility, and governance are essential parts of professional data communication. The ultimate test of a dashboard or report is whether it enables a better-informed and proportionate decision.

Key Terms

Data communication: The transformation of data and analysis into information that supports understanding and action.

Dashboard: A visual interface used to monitor selected measures and detect patterns or exceptions.

Data story: A structured explanation that combines evidence, context, interpretation, and action.

KPI: A key performance indicator selected to represent progress toward an important objective.

Metric: A defined quantitative measure; not every metric is a KPI.

Leading indicator: A measure associated with future performance or emerging conditions.

Lagging indicator: A measure of an outcome after it has occurred.

Denominator: The reference population or base used to calculate a rate or proportion.

Percentage point: The arithmetic difference between two percentages.

Relative change: Change expressed in relation to the original value.

Data lineage: The documented path from source data through transformations to a reported result.

Data quality: The suitability of data for its intended use, including accuracy, completeness, timeliness, and consistency.

Uncertainty: Lack of complete confidence about a measurement, estimate, prediction, or conclusion.

Confidence interval: An interval produced by a statistical procedure with a stated long-run coverage property under specified assumptions.

Threshold: A defined level that triggers attention, investigation, or action.

Data story: A sequence that explains what the evidence means and why it matters.

Data minimisation: Collecting or displaying no more personal data than is necessary for the purpose.

Case mix: The composition of cases or customers with potentially different characteristics and difficulty.

Goodhart’s law: The practical warning that a measure may become less useful when it is made a target.

Review Questions

Short Answer Questions

1. What is the main difference between a dashboard and an analytical report?
2. Why should a KPI include a definition of its denominator?
3. Distinguish a percentage-point change from a relative percentage change.
4. Give two sources of uncertainty in a data analysis.
5. Why should colour not be the only method used to communicate status?

Long Answer Questions

1. Explain how audience, purpose, and decision context influence the design of a data communication product.
2. Discuss the principles of responsible visualisation, including axes, colour, labels, denominators, and decoration.
3. Explain how a KPI system can create unintended behaviour and how an organisation can reduce that risk.
4. Describe a complete evidence-to-action process from identifying a problem to monitoring a recommendation.
5. Discuss the ethical and governance requirements for a managerial dashboard containing sensitive information.

Analytical / Application Questions

1. A service has 240 late cases out of 4,000 eligible cases in January and 300 out of 4,500 in February. Calculate both monthly late-case rates, the percentage-point change, and the relative change. Interpret the result for a manager.
2. Design a dashboard specification for a university concerned about student retention. Identify five KPIs, their definitions, audiences, possible limitations, and one decision each KPI could support.
3. A manager claims that a new training programme caused productivity to increase because output rose after training. Explain at least three alternative explanations and propose an improved evaluation and communication plan.

Multiple Choice Questions with Answers

Question 1

Which product is generally most suitable for repeated monitoring of current performance and exceptions?

A. Operational dashboard

B. Research literature review

C. One-time interview transcript

D. Unstructured data archive

Correct Answer: A. Operational dashboard

Question 2

A rate is calculated as (events ÷ population at risk) × 1,000. What does the denominator represent?

A. The number of visualisations used

B. The population exposed to the possibility of the events

C. The target value multiplied by 1,000

D. The number of missing observations

Correct Answer: B. The population exposed to the possibility of the events

Question 3

A late-delivery rate rises from 9% to 12%. What is the percentage-point increase?

A. 3 percentage points

B. 3.3 percentage points

C. 25 percentage points

D. 33.3 percentage points

Correct Answer: A. 3 percentage points

Question 4

Which visualisation is usually most appropriate for showing change over time?

A. Line chart

B. Pie chart with many segments

C. Three-dimensional bar chart

D. Word cloud

Correct Answer: A. Line chart

Question 5

Why should a dashboard show a KPI’s refresh date?

A. To make the dashboard look more technical

B. To indicate how current the information is

C. To replace the KPI definition

D. To guarantee that the data is accurate

Correct Answer: B. To indicate how current the information is

Question 6

Which statement best describes the relationship between correlation and causation?

A. Correlation always proves causation

B. Causation can be established from any scatter plot

C. Association alone does not establish that one variable causes another

D. Correlation and causation are identical terms

Correct Answer: C. Association alone does not establish that one variable causes another

Question 7

A department has 1,200 late deliveries out of 10,000 eligible deliveries. What is the late-delivery rate?

A. 1.2%

B. 8.3%

C. 12%

D. 120%

Correct Answer: C. 12%

Question 8

Which is an example of a specific limitation statement?

A. The data may have limitations

B. The dashboard is modern and attractive

C. Fourteen percent of records lack a verified location, so regional comparisons may be biased

D. The result is definitely correct

Correct Answer: C. Fourteen percent of records lack a verified location, so regional comparisons may be biased

Question 9

What is a major privacy safeguard for managerial dashboards?

A. Displaying all raw records to every user

B. Using role-based access and data minimisation

C. Removing the refresh date

D. Using more decorative colours

Correct Answer: B. Using role-based access and data minimisation

Question 10

What is the main danger of treating a single performance measure as the only target?

A. It can encourage optimisation of the measure while neglecting unmeasured quality

B. It always decreases data volume

C. It prevents any numerical calculation

D. It makes all forecasts unbiased

Correct Answer: A. It can encourage optimisation of the measure while neglecting unmeasured quality