A CONNECTED LEARNING PATH / 18 GUIDES
Data quality.
Follow a question, explore a worked example and choose one useful action. Read in order or start with the challenge in front of you.
Explore every topicDistinguish missing values from genuine zeros
Handle blanks and zeros according to their business meaning, preserve uncertainty and avoid silently changing totals or averages.
Measure completeness against the population you expected
Check completeness at the record and field level, explain the expected population and investigate missing-data patterns before reporting a percentage.
Resolve duplicate entities without deleting useful history
Distinguish duplicate records from legitimate repeated events and make entity matching reviewable before combining or removing information.
Write validation rules that respect legitimate exceptions
Use format, range and relationship checks to identify records for review without confusing valid-looking values with verified truth.
Show data freshness and reporting periods separately
Help report readers distinguish the period measured, the source update time and the dashboard refresh time before acting on an apparent change.
Choose identifiers that survive a system handover
Keep customer and transaction relationships intact by specifying identifier scope, preserving old keys and testing migration mappings.
Keep units and currencies visible during an import
Import measurements and money with explicit units, conversion rules and original values so apparently plausible totals remain interpretable.
Set reporting cut-offs before comparing time periods
Define event clocks, time zones and late-arrival rules before assigning records to days, weeks or reporting periods.
Build a category taxonomy people can maintain
Design categories with operational definitions, ambiguous-case examples and a controlled change process while preserving original source labels.
Check relationships between records
Find broken references between customers, orders and other linked records while distinguishing timing differences from genuine relationship errors.
Investigate outliers before correcting them
Distinguish unusual observations from verified errors by checking comparable populations, original evidence and the effect on reported measures.
Handle late-arriving data without silent revisions
Build an explicit revision policy that keeps event periods, arrival dates and previously published totals traceable when late records appear.
Make spreadsheet imports repeatable
Control spreadsheet structure, parsing and reconciliation so a new title row or locale change cannot silently shift imported meaning.
Agree a source-change contract
Create a practical agreement for field, status and meaning changes so downstream reports can assess compatibility before new data arrives.
Triage quality issues by their decision impact
Prioritise data issues by the decisions, populations and values they can distort rather than by error counts alone.
Use a data-quality score without hiding exceptions
Design quality summaries with explicit denominators, dimensions and critical exceptions so an attractive average cannot override a material limitation.
Validate accuracy with a source-based spot check
Check selected values against authorised supporting records, document selection and discrepancies, and keep conclusions within the sample’s evidence.
Compare consistency across sources without forcing agreement
Reconcile source differences through concepts, scope and timing before treating every unequal total as an error to eliminate.