Data Literacy and Business Questions

Introduction

Data visualization and business intelligence (BI) are often presented as technology disciplines, but their value depends first on human judgement. A dashboard cannot compensate for a poorly defined question, an unsuitable metric, or unreliable data. Data literacy is the ability to read, understand, question, analyse and communicate data in a particular context. For a business analyst, it includes both technical competence and the ability to connect evidence with organisational decisions.

A useful analytical process begins before data are extracted:

1. Understand the decision or business problem.
2. Translate the problem into answerable questions.
3. Identify appropriate measures, dimensions and data sources.
4. Assess data quality and limitations.
5. Analyse and communicate results responsibly.
6. Support action and evaluate outcomes.

This chapter develops the foundations required for that process. It explains data types and measurement scales, distinguishes business questions from analytical questions, examines metrics and key performance indicators (KPIs), and introduces common limitations affecting data quality and interpretation.

Learning Objectives

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

• distinguish qualitative and quantitative data and common structured data types;
• classify variables according to nominal, ordinal, interval and ratio measurement scales;
• translate business requirements into clear analytical questions;
• distinguish dimensions, measures, metrics and KPIs;
• assess data quality using recognised dimensions;
• identify bias, missingness, uncertainty and other limitations;
• apply data-literacy principles to a practical BI scenario; and
• communicate findings without overstating what the data can support.

1. Data as Evidence for Decisions

Data are recorded observations, measurements or representations of events. Information is data interpreted in context. Insight is a well-supported understanding that can guide action. These terms are related but not interchangeable. A spreadsheet containing 20,000 transactions is data; a finding that repeat customers have a higher average order value is information; a decision to create a loyalty campaign based on a reliable analysis may reflect insight.

Data become useful only when their meaning, provenance and limitations are understood. For example, “sales” may mean invoiced revenue, cash received, gross merchandise value or revenue after returns. A dashboard that does not define the term may create disagreement even when all calculations are technically correct.

Business intelligence commonly combines:

• descriptive analysis: what happened;
• diagnostic analysis: why it happened;
• predictive analysis: what may happen; and
• prescriptive analysis: what action may be appropriate.

Data literacy is especially important because each level requires assumptions. A descriptive trend does not prove causation, and a forecast is not a guarantee. Responsible analysts state what the evidence supports and what remains uncertain.

2. Data Types and Structures

2.1 Qualitative and quantitative data

Qualitative data describe categories, attributes or meanings. Examples include product category, customer complaint type and department. They may be stored as text, but text storage does not automatically make a variable qualitative: a postcode stored as digits is still a label rather than a quantity suitable for arithmetic.

Quantitative data represent amounts or counts. Examples include revenue, delivery time, units sold and temperature. Quantitative data may be:

• discrete, taking separate countable values, such as number of orders; or
• continuous, potentially taking any value within a range, such as package weight or elapsed time.

The distinction matters because a count of 4 orders is meaningful, whereas an average of customer IDs is not.

2.2 Common data structures

A single business system may contain several structures. A BI model might combine customer, product and transaction tables with a calendar table. Analysts must understand relationships between tables because an incorrect join can duplicate transactions and inflate totals.

2.3 Identifiers, attributes and measures

An identifier distinguishes an entity, such as CustomerID or InvoiceNumber. It is not normally a measure, even when represented numerically. An attribute describes an entity, such as customer segment or product colour. A measure is a value that can be aggregated or analysed, such as units, cost or revenue.

Measures can be:

• additive across all relevant dimensions, such as sales revenue;
• semi-additive, additive across some dimensions but not others, such as account balance, which may be added across customers but not across time; or
• non-additive, such as percentages and ratios, which should generally be recalculated from their underlying components rather than summed.

For example, a monthly conversion rate should not normally be averaged without considering the number of visitors in each month. The overall rate is calculated from total conversions divided by total relevant visitors.

3. Measurement Scales

Measurement scales describe what can legitimately be inferred from values. Four traditional scales are nominal, ordinal, interval and ratio. The scale determines suitable summaries, charts and statistical operations.

3.1 Nominal scale

Nominal values label groups. “North,” “South” and “West” have no inherent numerical order. Coding them as 1, 2 and 3 does not make the codes quantitative. A bar chart or frequency table is appropriate; calculating an average region code is meaningless.

Binary variables are nominal variables with two categories, such as subscribed/not subscribed. They may be represented as 0 and 1 for modelling, but the coding convention must not obscure the underlying meaning.

3.2 Ordinal scale

Ordinal values have an order, but the distance between adjacent categories may not be equal. A five-point satisfaction scale from “very dissatisfied” to “very satisfied” is ordinal. The difference between categories 1 and 2 cannot automatically be assumed equal to the difference between 4 and 5. Analysts may report the distribution, median or percentage selecting favourable responses. Treating ordinal responses as interval data can be acceptable in some survey analyses, but it is an assumption that should be disclosed.

3.3 Interval scale

Interval data have equal differences between values but no meaningful absolute zero. In Celsius, 20 °C is not twice as hot as 10 °C. Differences, such as a 10-degree increase, are meaningful; ratios are not. Calendar years are another example: 2024 is not twice 1012 in a meaningful temporal sense.

3.4 Ratio scale

Ratio data have equal intervals and a meaningful zero representing absence of the measured quantity. A revenue of $200 is twice a revenue of $100, assuming the same currency and definition. Ratio-scale data support addition, subtraction, multiplication and division, subject to business context.

Measurement-scale classification is not merely academic. It prevents invalid analysis and helps select visual encodings. A nominal category should not be displayed as a continuous line merely because it has been assigned a numerical code.

4. From Business Requirements to Analytical Questions

A business requirement states a need or capability from a stakeholder’s perspective. An analytical question specifies what evidence must be examined to address that need. For example:

• Requirement: “Improve profitability in the retail division.”
• Weak question: “Can we analyse retail data?”
• Stronger questions: “Which product categories have declining gross margin by region?” and “How has discounting affected average order value during the last four quarters?”

A well-formed question identifies an object, measure, population, time period and often a comparison. The SMART principle—specific, measurable, achievable, relevant and time-bounded—can help, but analytical quality also requires clarity about definitions and decision use.

4.1 A requirements framework

Before building a dashboard, clarify:

1. Decision: What decision will the analysis support?
2. Decision owner: Who is accountable for acting?
3. Users: Who will view or operate the output?
4. Business process: Where does the issue occur?
5. Scope: Which products, customers, locations and dates are included?
6. Measures: What outcomes indicate performance?
7. Dimensions: By which categories should performance be compared?
8. Refresh and timing: How current must the data be?
9. Constraints: What privacy, access, budget or system restrictions apply?
10. Acceptance criteria: What would make the output useful and trustworthy?

A requirements interview should expose ambiguous terms. “Customer” might mean anyone who has registered, anyone who has purchased, or an organisation with multiple contacts. “Churn” might mean cancellation, inactivity for 90 days, or non-renewal at contract expiry.

4.2 Question types

Common analytical questions include:

• descriptive: How many units were sold last month?
• comparative: Which region grew fastest year on year?
• diagnostic: Which factors are associated with delayed delivery?
• trend-based: How has retention changed over six quarters?
• segmentation: Which customer groups generate the greatest contribution?
• exception-oriented: Which transactions violate approval rules?
• forecasting: What demand range is plausible next month?
• decision-oriented: Which intervention should be tested first?

The question determines the data, calculation and visualization. A trend question requires a valid time field and consistent periods. A comparison requires comparable denominators. A forecasting question requires historical patterns and explicit uncertainty.

5. Metrics, Measures and KPIs

A measure is a recorded or calculated quantity. A metric is a defined measure used to assess a phenomenon or process. A KPI is a strategically important metric linked to an objective, target and decision responsibility. Not every metric is a KPI.

For a subscription business, examples include:

• active subscribers: a count of customers meeting an activity definition;
• monthly recurring revenue: recurring revenue under a stated accounting or management definition;
• churn rate: customers lost during a period divided by the relevant customer base;
• customer acquisition cost (CAC): acquisition expenditure divided by acquired customers, under a defined attribution rule; and
• retention rate: the proportion of a defined starting population remaining active at the end of a period.

A generic rate is:

The numerator and denominator must refer to compatible populations and periods. If 80 of 1,000 eligible customers renew, the renewal rate is 80 ÷ 1,000 = 0.08, or 8%. Reporting “80 renewals” without the denominator may mislead when comparing segments of different sizes.

A percentage change between an initial value and a later value is:

This formula assumes the initial value is non-zero and that both values use the same definition and units. If the initial value is negative or close to zero, interpretation requires special care.

5.1 Metric design

A metric specification should record:

A target can change behaviour. If a call centre is rewarded only for short handling time, agents may end calls prematurely and reduce service quality. Balanced measurement combines efficiency, quality and outcome measures.

6. Data Quality and Analytical Limitations

Data quality is fitness for a stated purpose, not an absolute property. Data may be sufficiently accurate for a monthly overview but unsuitable for real-time fraud detection. Important dimensions include:

• accuracy: values represent the real-world entity or event;
• completeness: required records or fields are present;
• consistency: values agree across systems and definitions;
• timeliness: data are available and current enough;
• validity: values conform to rules, formats and ranges;
• uniqueness: records are not improperly duplicated; and
• integrity: relationships between records are preserved.

6.1 Common problems

Missing data may arise because a field was optional, a sensor failed, a customer declined to answer, or records were not integrated. Missingness can be concentrated in particular groups. Deleting incomplete records may therefore create bias. Analysts should report the extent and pattern of missingness and justify any imputation or exclusion.

Duplicates can inflate counts and revenue. Duplicate detection may use a transaction ID, but repeated legitimate events must not be removed simply because their values look similar.

Inconsistent definitions are common across departments. Finance may report recognised revenue while marketing reports booked sales. Both can be correct for different purposes; the issue is undocumented meaning.

Outliers may be errors, exceptional events or important signals. A very large order should be investigated rather than automatically deleted. An outlier can materially affect an average, so analysts may compare mean and median and explain the treatment used.

Sampling bias occurs when observed records do not represent the target population. Website survey respondents may be more dissatisfied or engaged than non-respondents. A large dataset can still be biased; volume does not guarantee representativeness.

Survivorship bias occurs when analysis excludes entities that disappeared. Studying only current customers may overestimate satisfaction because former customers are absent.

Selection and confirmation bias can enter when analysts choose only convenient data or search for evidence supporting a preferred explanation. Predefined questions, documented transformations and independent review reduce these risks.

6.2 Correlation and causation

If two variables move together, they are associated, but the relationship may result from coincidence, a common cause, reverse causality or measurement artefacts. For example, advertising expenditure and sales may rise together because both increase during seasonal demand. A chart alone cannot establish that advertising caused all additional sales. Causal claims require an appropriate design, such as a controlled experiment or carefully justified quasi-experimental approach.

6.3 Aggregation and the ecological fallacy

Aggregated results can conceal subgroup differences. Overall customer satisfaction may rise while satisfaction falls among small-business customers. This is why analysts should examine relevant dimensions and avoid assuming that a group-level relationship applies to every individual. Changing aggregation can even reverse an apparent pattern, a phenomenon often associated with Simpson’s paradox.

7. Visual and Communicative Implications

Data literacy affects chart choice and interpretation. Bar charts are suitable for comparing categories; line charts are suitable for ordered time; scatter plots help examine relationships between two quantitative variables; and maps are appropriate when geographic location is analytically relevant. Pie charts can be difficult to compare when categories are numerous or similar in size.

A visualization should make the question visible. It should identify units, time period, source and definitions. Axes should not exaggerate small differences, and colour should not imply an unsupported ranking. A dashboard should distinguish:

• actuals from forecasts;
• counts from rates;
• targets from observed results; and
• statistically or operationally meaningful changes from normal variation.

Accessibility is part of quality. Use sufficient contrast, descriptive labels and alternatives for users who cannot distinguish colours. Avoid relying on red and green alone, and do not disclose personal information through unnecessary detail.

8. Ethical, Governance and Privacy Considerations

BI projects may involve personal, behavioural or sensitive data. Analysts should follow applicable law and organisational policy, including principles of purpose limitation, data minimisation, access control, retention and accountability. A legitimate business interest does not justify unlimited collection or indefinite retention.

Important governance practices include a data catalogue, business glossary, lineage documentation, role-based access, quality checks and an escalation route for suspected errors. Personally identifiable information should be masked, aggregated or pseudonymised where possible. Pseudonymisation reduces direct identification but does not make data anonymous if re-identification remains possible.

AI-assisted analytics can accelerate summarisation, querying and anomaly detection, but generated outputs may contain fabricated explanations, reproduce historical bias or expose confidential data. Human review, approved tools, audit logs and clear accountability remain necessary. An analyst must be able to explain the source, transformation and limitations of a result.

9. Illustrative Case: Northstar Grocery Delivery

Background

Northstar Grocery operates 35 stores and offers home delivery. Managers report that customer complaints about late deliveries have increased. The operations director requests a dashboard “to show delivery performance and identify the problem stores.”

Problem definition

The analyst first clarifies that “late” means arrival more than 15 minutes after the promised time, not merely a long delivery. The decision is whether to change staffing and delivery-slot allocation. The population is completed home-delivery orders in the last six months; cancelled orders are excluded because no delivery occurred.

Data and metric design

The analyst combines order, promised-slot, dispatch and arrival tables. A data dictionary identifies the following fields:

• OrderID, used to test uniqueness;
• StoreID and delivery region;
• promised arrival timestamp;
• actual arrival timestamp;
• order status; and
• complaint indicator.

The primary metric is:

A second metric, median lateness among late orders, prevents a few extreme delays from being hidden by an average. The dashboard includes order volume because a rate based on 10 orders is less stable than a rate based on 10,000 orders.

Quality investigation

The analyst finds 3% of completed orders lack an arrival timestamp, and one store has timestamps recorded in local time while the central system uses UTC. After correction, 1,200 records remain ambiguous and are excluded from the primary calculation but shown in a quality indicator. Duplicate OrderIDs are found in a historical export and removed only after reconciliation with the source system.

Analysis and solution

The dashboard compares late rate by store, day of week, delivery slot and order volume. Store C has the highest rate, but its late orders are concentrated on Friday evenings. Further investigation shows that promised slots were calculated using weekday capacity assumptions even during a local market event. The immediate solution is to adjust capacity rules for event periods and test an additional driver shift on Friday evenings.

Outcome

After four weeks, the team compares the same metric definition with the preceding four weeks. Late rate declines from 14% to 9%, while order volume remains similar. The result is encouraging but does not prove that the staffing change alone caused the improvement; weather and demand conditions also changed. Managers therefore schedule a longer controlled comparison and continue monitoring complaint rate and delivery cost.

Learning points

The case demonstrates that a useful dashboard depends on precise definitions, compatible timestamps, quality checks, denominators, segmentation and cautious causal interpretation. The visual product is the final expression of analytical work, not its substitute.

Best-Practice Checklist

Before publishing an analysis or dashboard, ask:

• Is the decision and intended user explicit?
• Are terms, populations, dates and units defined?
• Are measures calculated at the correct grain?
• Are counts, rates and denominators shown together where relevant?
• Have missing, duplicate, invalid and inconsistent records been assessed?
• Are comparisons fair and based on comparable periods?
• Does the visualization match the measurement scale and question?
• Are uncertainty, assumptions and limitations visible?
• Are privacy, access and retention requirements satisfied?
• Can another analyst reproduce and explain the result?

Chapter Summary

Data literacy is the foundation of effective data visualization and BI. Data must be interpreted in context, and numerical appearance does not make a variable suitable for arithmetic. Qualitative and quantitative data, discrete and continuous variables, identifiers, attributes and measures serve different analytical purposes. Nominal, ordinal, interval and ratio scales determine which comparisons and calculations are defensible.

Business requirements should be translated into specific analytical questions linked to decisions, populations, measures, dimensions and time. Measures become metrics when they are defined for assessment, and become KPIs when they are strategically important, owned and connected to targets. Rates require compatible numerators and denominators, while percentages and ratios should not be aggregated carelessly.

Data quality is multidimensional. Accuracy, completeness, consistency, timeliness, validity, uniqueness and integrity must be assessed relative to purpose. Missingness, duplicates, bias, outliers, aggregation and correlation–causation errors can mislead even sophisticated dashboards. Ethical governance requires privacy protection, documentation, controlled access and human accountability, including when AI tools are used. Sound BI therefore combines technical calculation with disciplined questioning and professional judgement.

Key Terms

Data literacy: Ability to understand, question, analyse and communicate data in context.

Business intelligence: Processes and technologies that transform data into information for decision-making.

Qualitative data: Data representing categories, attributes or meanings.

Quantitative data: Data representing amounts or counts.

Discrete data: Quantitative data consisting of separate countable values.

Continuous data: Quantitative data that can take values across a range.

Nominal scale: Categories with no inherent order.

Ordinal scale: Ordered categories without an assumed equal distance between levels.

Interval scale: Ordered values with equal intervals but no meaningful zero.

Ratio scale: Values with equal intervals and a meaningful zero.

Dimension: A category or perspective used to group or filter measures.

Measure: A recorded or calculated quantity.

Metric: A defined measure used to assess a phenomenon or process.

KPI: A strategically important metric linked to an objective, target and responsibility.

Data quality: Fitness of data for a stated purpose.

Completeness: The extent to which required data are present.

Validity: Conformity with specified formats, rules or ranges.

Sampling bias: Systematic difference between observed data and the target population.

Data lineage: Record of a dataset’s sources, transformations and destinations.

Ecological fallacy: Incorrectly applying group-level relationships to individuals.

Review Questions

Short Answer Questions

1. Distinguish data, information and insight.
2. Why is a customer ID not an appropriate measure for calculating an average?
3. Explain the difference between ordinal and interval data.
4. What is the purpose of a denominator in a performance rate?
5. Name four dimensions of data quality.

Long Answer Questions

1. Explain the four measurement scales and discuss how scale affects analysis and visualization.
2. Describe a systematic process for translating a business requirement into analytical questions.
3. Distinguish measures, metrics and KPIs, using a business example.
4. Discuss how missing data, duplicates and inconsistent definitions can affect a dashboard.
5. Explain why correlation does not establish causation and identify two sources of misleading relationships.

Analytical / Application Questions

1. A retailer reports 900 returns from 18,000 orders in one quarter. Define an appropriate return-rate metric, identify its numerator and denominator, and state two limitations of comparing it across stores.
2. A satisfaction survey shows an overall improvement, but results are not reported by customer type. Design an analysis to test whether the improvement applies equally to all important segments.
3. Design a metric specification for late delivery performance, including the business question, formula, population, time logic, dimensions, quality checks and governance risks.

Multiple Choice Questions with Answers

Question 1

Which measurement scale describes payment method such as cash, card or mobile wallet?

A. Nominal

B. Ordinal

C. Interval

D. Ratio

Correct Answer: A. Nominal

Question 2

Which is the strongest analytical question for a reported decline in profitability?

A. Can we access the sales database?

B. Which product categories show declining gross margin by region?

C. Does the company have many products?

D. Can a dashboard be created quickly?

Correct Answer: B. Which product categories show declining gross margin by region?

Question 3

A business records 80 renewals from 1,000 eligible customers. What is the renewal rate?

A. 0.8%

B. 8%

C. 80%

D. 920%

Correct Answer: B. 8%

Question 4

Which measure is generally semi-additive?

A. Units sold during a transaction

B. Revenue by order

C. Account balance across time

D. Number of complaints

Correct Answer: C. Account balance across time

Question 5

Which data-quality dimension concerns whether values follow permitted formats and ranges?

A. Timeliness

B. Validity

C. Uniqueness

D. Completeness

Correct Answer: B. Validity

Question 6

Why should a monthly conversion rate not automatically be calculated as the simple average of daily conversion rates?

A. Rates cannot be displayed in charts

B. Daily denominators may differ

C. Conversion is always nominal

D. Monthly data cannot contain dates

Correct Answer: B. Daily denominators may differ

Question 7

Which statement best describes a KPI?

A. Any field stored in a database

B. A strategically important metric linked to an objective and target

C. A category used to filter a report

D. A unique identifier for a transaction

Correct Answer: B. A strategically important metric linked to an objective and target

Question 8

A postcode stored as digits should normally be treated as which kind of variable?

A. Ratio measure

B. Interval measure

C. Nominal identifier or category

D. Continuous measurement

Correct Answer: C. Nominal identifier or category

Question 9

Which situation is an example of survivorship bias?

A. Removing a duplicated transaction ID after reconciliation

B. Studying only current customers while ignoring customers who cancelled

C. Converting currencies using a documented exchange rate

D. Reporting missing values by segment

Correct Answer: B. Studying only current customers while ignoring customers who cancelled

Question 10

What is the most appropriate conclusion when advertising and sales rise together?

A. Advertising definitely caused all additional sales

B. The variables are associated, but causation requires further evidence

C. Sales must have caused advertising expenditure

D. The data are invalid because two variables changed

Correct Answer: B. The variables are associated, but causation requires further evidence