Data quality work establishes whether records still represent the customers, transactions, assets or events a business intends to analyse. Basic cleaning can correct formats and duplicates, but it cannot explain why information is missing or why teams record the same event differently. Marketways traces errors back to definitions and operating processes before correcting them. The resulting data provides a defensible basis for reporting, modelling and action.
The decision this method supports
We use Data Quality, Cleaning & Auditing to help clients answer: What must be corrected, documented or excluded before the data can support a decision?
How the method works
Data quality work examines whether recorded information still represents the real event, customer, asset or transaction that the analysis is meant to describe. Cleaning may correct formats and duplicates, but auditing also asks how the data was created, which definitions changed and whether a correction would remove meaningful variation.
A business example
A company may hold customer records from several branches. One branch records a returned item as a negative sale, while another records a separate return transaction. Combining the fields without reconciling the definitions would distort revenue, return rates and any model trained on the data.
How the client uses the result
Reliable data prevents managers from acting on duplicate customers, inconsistent definitions or records that no longer represent the business. Marketways also examines why a record exists or is missing, because cleaning cannot repair a biased data-generating process. The immediate benefit is greater confidence in reports and models; the longer-term benefit is less time spent reconciling competing versions of the truth.
What we deliver
We produce a documented dataset, quality assessment, measure dictionary, analytical-readiness decision or management view. A manager should be able to see what the information represents, which limitations remain and what action the resulting view can support.
Limits and complementary methods
Cleaning improves consistency, but an aggressive correction can erase real differences. The audit trail must preserve what changed and why.
Selected methods and techniques
We select from these established methods according to the decision, evidence and operating conditions.
- Availability / performance / quality decomposition: Separate lost effective production into availability, performance and quality components using stated and internally consistent bases. Define planned production time, running time, ideal cycle time or rate, total units, good units, exclusions and the asset and period covered; then calculate the three factors and reconcile their combined effect without counting the same loss twice. The method performs the decomposition, the Overall Equipment Effectiveness framework supplies the common metric structure, and the resulting Availability / Performance / Quality Breakdown records the populated calculation. The factors locate types of loss but do not explain their operational causes.
- Complaint / case-data analysis: Analyse complaint, service-case or recovery records to identify recurring issues, severity, journey stage, resolution time, outcome, reopening and escalation. Define the case and exposure population, remove or link duplicate records, document coding rules and changes in logging practice, and compare counts with a suitable denominator such as transactions or active customers where available. Separate the frequency of recorded complaints from the prevalence of customer problems: many people do not complain, some complain repeatedly and channels differ in what they capture. Case patterns can locate problems and recovery gaps but do not represent all customers or show that recovery caused a later outcome.
- Confidence provenance analysis: Determine what produced a stated confidence level, including the initial model score or prior, later evidence, repeated or dependent signals, questions, tool results and update rules. Separate final confidence from evidence newly earned during the process and from empirical calibration. The analysis can expose decorative reasoning or confidence inflation; provenance alone does not say whether the conclusion is correct.
- Critical-to-quality analysis: Translate what customers or stakeholders value into specific, measurable characteristics the system must deliver, with targets and tolerances. It links needs to measurable performance requirements, with targets and tolerances validated against operational capability as well as stakeholder preference. The characteristics are only as good as the evidence about what stakeholders actually value.
- Data-drift detection: Check whether the data a model receives have changed compared with the data used as its reference. Data drift concerns the inputs; concept drift concerns changes in how those inputs relate to outcomes.
- Data-flow exposure analysis: Trace where data, especially sensitive or personal data, travels through a system and identify the points where it could be exposed, copied, retained unnecessarily or reach parties without a need to see it. It specialises data-flow modelling for protection. It depends on data flows being mapped completely.
- Data-flow modelling: Model how data moves through a system: its sources, the processes that transform it, where it is stored and where it goes. It exposes duplication, gaps, delays and privacy exposure in information handling. It models data movement, not the physical databases used.
- Data-minimisation analysis: Examine each data item a system collects, uses or keeps and decide whether it is needed for a stated purpose, at what level of detail and for how long, removing or reducing data that is not needed. It reduces exposure and privacy risk. It relies on purposes being defined clearly.
- Decision provenance analysis: Trace material parts of a decision, such as its frame, assumptions, evidence, alternatives and recommendations, to their documented origins and subsequent changes. Use timestamps, version history and independent pre-exposure records where available, while marking mixed or unknown origins. The method establishes lineage and sequence; a claim that a source influenced judgement requires additional evidence.
- Frame audit: Examine whether the representation of a decision matches the underlying objective. Make explicit the decision question, success criteria, boundaries, assumptions, time horizon, included and excluded possibilities, and who introduced them. A frame audit can recommend reframing when the wrong problem is being solved; it is distinct from comparing alternatives inside a frame already accepted.
- Missing-evidence analysis: Identify evidence absent from an argument, model or decision and link each gap to the claim or action it could change. Assess materiality, whether the evidence can be obtained and what status the conclusion should have if it remains missing. The issue is absent support. Low data volume is a different problem.
- Panel-data analysis: Analyse repeated observations of the same people, organisations, locations or other entities over time. Use the repeated structure to distinguish changes within an entity from persistent differences between entities, while accounting for common time effects and dependence between observations from the same entity. Define the panel, timing and comparison; inspect missing periods, entry, exit and attrition; and state any weighting or model assumptions. Panel data can strengthen temporal comparison, but repeated observation alone does not establish causality. A series of different samples drawn at each date is repeated cross-sectional data, not a panel.
Parent method family
Data Foundations & Business Intelligence explains how this method connects to adjacent methods and relevant services.
Related service families
These service families contain business questions supported by this method. Service pages link to the wider method family so readers can understand the complete analytical approach.