Foundations of Business Data Analytics
Business Data Analytics is the newest addition to the CMA syllabus, added because a cost and management accountant is increasingly expected to work with data well beyond the general ledger — and this chapter builds the vocabulary the rest of the topic depends on.
1. Types of data
Data is first split into structured and unstructured form. Structured data fits neatly into rows and columns with a fixed, predefined format — a ledger, a sales register, a payroll table — and is what traditional accounting and costing systems have always used. Unstructured data has no predefined format — emails, call-centre recordings, social-media posts, scanned invoices — and requires different tools (text mining, natural-language processing, image recognition) to extract anything usable from it.
Semi-structured data sits between the two, carrying some organisational markers (such as tags in an XML or JSON file) without fitting a strict tabular schema.
Within structured data, every variable falls on one of four measurement scales, and this classification decides which statistical operations are even meaningful on it:
| Scale | What it captures | Example | Meaningful operations |
|---|---|---|---|
| Nominal | Named categories, no inherent order | Product colour, department name | Count, mode |
| Ordinal | Ordered categories, unequal or unknown gaps | Customer satisfaction rating (poor/average/good) | Count, mode, median |
| Interval | Ordered, equal gaps, no true zero | Temperature in Celsius, a date | All of the above, plus mean, addition/subtraction |
| Ratio | Ordered, equal gaps, a true zero point | Sales revenue, cost per unit, number of units produced | All arithmetic operations, including meaningful ratios |
The scale matters practically because it is a frequent error to compute a mean on ordinal data (such as averaging a 1-5 satisfaction score as if the gap between "2" and "3" were provably identical to the gap between "4" and "5") — a calculation that is arithmetically possible but statistically shaky, since ordinal scales do not guarantee equal spacing between categories.
2. Data quality — six dimensions
Data that is technically available is not automatically usable, and data quality is formally assessed against six recognised dimensions.
| Dimension | Question it answers |
|---|---|
| Accuracy | Does the data correctly reflect the real-world fact it represents? |
| Completeness | Are all required fields and records present, with no missing gaps? |
| Consistency | Does the same fact agree across different systems or records? |
| Timeliness | Is the data current enough to be useful for the decision at hand? |
| Validity | Does the data conform to the defined format, type and range rules? |
| Uniqueness | Is each real-world entity represented exactly once, with no duplicates? |
Data governance is the organisational framework — policies, defined ownership, and accountability — that keeps these six dimensions under control over time, rather than data quality being an accident of whoever last touched a spreadsheet. A named data owner is accountable for a dataset's overall quality and appropriate use, while a data steward carries out the day-to-day work of maintaining definitions, resolving quality issues and enforcing access rules on the owner's behalf.
Master data management (MDM) is the specific governance discipline of maintaining one single, authoritative version of core business entities — a customer, a product, a vendor — so that the same entity is not represented inconsistently across a company's sales system, its accounting system and its costing system.
3. The Big Data 5 Vs
Big data is conventionally described by five characteristics, all starting with "V", that distinguish it from a traditional, moderately sized dataset: Volume (the sheer quantity of data generated), Velocity (the speed at which it is generated and must be processed), and Variety (the mix of structured, semi-structured and unstructured formats).
The remaining two are Veracity (the trustworthiness and accuracy of the data, tying directly back to the data-quality dimensions above) and Value (the actual business benefit that can be extracted from it — data with volume, velocity and variety but no extractable value is not worth the cost of managing it).
4. The analytics maturity ladder — four types
Every analytics activity a business performs can be placed on a four-step maturity ladder, and each step answers a progressively harder question than the one before it.
| Type | Question it answers | Typical technique |
|---|---|---|
| Descriptive | What happened? | Reports, dashboards, summary statistics |
| Diagnostic | Why did it happen? | Drill-down analysis, correlation, root-cause analysis |
| Predictive | What is likely to happen? | Regression, forecasting models, trend extrapolation |
| Prescriptive | What should we do about it? | Optimisation models, simulation, decision analysis |
Each step is harder and more valuable than the one before it, but also depends on the ones below it being done well — a business cannot reliably predict next quarter's sales (predictive) if it cannot first accurately and consistently report last quarter's sales (descriptive), which is the practical reason most organisations' analytics maturity climbs this ladder in order rather than jumping straight to prescriptive techniques.
Worked Examples
Example 1. Classify each of the following as structured, semi-structured or unstructured data: (a) a company's monthly trial balance, (b) a folder of scanned supplier invoices, (c) a product catalogue exported as a tagged XML file.
(a) Structured. (b) Unstructured (until processed by, for example, optical character recognition). (c) Semi-structured.
Example 2. Identify the measurement scale of each variable: (a) employee department (Sales, Production, Finance), (b) a customer's ranking of three products from most to least preferred, (c) monthly sales revenue in rupees.
(a) Nominal. (b) Ordinal. (c) Ratio.
Example 3. A finance team notices its cost-centre codes are recorded as "CC-101" in one system and "101" in another for the same cost centre. Which data-quality dimension is being violated, and which governance role should be primarily accountable for fixing it?
Consistency is violated. The data steward is primarily accountable for resolving the definitional mismatch and enforcing a single agreed format across systems, under policies set by the data owner.
Example 4. A retail chain's analytics team reports last month's sales fell 12% (Report A), finds the fall was concentrated in one region due to a stockout (Report B), forecasts sales will recover next month once stock is replenished (Report C), and recommends reallocating inventory from an overstocked region to prevent recurrence (Report D). Classify each report by analytics type.
Report A: Descriptive. Report B: Diagnostic. Report C: Predictive. Report D: Prescriptive.
Example 5. Which of the Big Data 5 Vs is most directly concerned with whether a dataset can actually be trusted for decision-making?
Veracity.
Example 6. A dataset of customer records is missing the phone number field for 30% of its 10,000 records. Which data-quality dimension does this violate, and what percentage of records are affected?
Completeness; 30% of records (3,000 out of 10,000) are affected.
Example 7. Explain why computing an arithmetic mean on a five-point ordinal satisfaction scale (1 = Very Poor to 5 = Very Good) is statistically questionable, even though it is arithmetically possible.
An ordinal scale only guarantees the ranking order of categories, not that the gaps between them are equal — the difference between "Very Poor" and "Poor" is not provably the same size as the difference between "Good" and "Very Good." Averaging treats these unequal, unverified gaps as if they were numerically identical, so the resulting mean can be misleading; the median or mode is generally the more defensible summary statistic for ordinal data.
Summary
Data splits into structured, semi-structured and unstructured forms, and structured data further splits into four measurement scales — nominal, ordinal, interval and ratio — that decide which statistical operations are meaningful on it. Six dimensions (accuracy, completeness, consistency, timeliness, validity, uniqueness) define data quality, upheld through data governance, with data owners accountable for a dataset overall and data stewards handling its day-to-day maintenance; master data management keeps one authoritative version of core entities across systems.
Big data is characterised by five Vs — Volume, Velocity, Variety, Veracity and Value — and every analytics activity climbs a four-step maturity ladder from descriptive (what happened) through diagnostic (why) and predictive (what next) to prescriptive (what to do), each step depending on the reliability of the ones beneath it.
