A maintainable taxonomy groups records according to a stated analytical purpose, with clear inclusion rules and examples. Preserve source labels, provide an explicit unknown category and version the mapping. Test ambiguous cases with the people applying it before treating category totals as comparable.
Start with the decisions categories must support
A useful category scheme reflects distinctions people need to act on. An expense taxonomy for reimbursement approval may differ from one for transport planning. List the questions the classification should answer, then identify which distinctions change those answers. Avoid copying every spelling from a source system into a permanent category. Equally, do not compress different activities into one group simply because a short chart legend looks cleaner.
Choose the unit being classified. A single invoice may contain accommodation, rail travel and a software purchase. Classifying the entire invoice by its largest line loses the smaller categories. If line-level analysis is required, establish whether reliable line records exist. Where only document-level data is available, state that limitation instead of manufacturing precise allocations from a free-text summary.
Write rules that resolve overlapping meanings
Give each category an identifier, plain definition, inclusions, exclusions and boundary examples. A label such as travel is insufficient when another category is transport. Specify whether travel is a parent group and transport a child, or whether they represent mutually exclusive concepts. If a record can receive several tags, distinguish that tagging system from a classification intended to produce additive totals.
Keep unknown separate from other. Unknown means there is insufficient information to classify; other means the activity is understood but outside named categories. Set a review trigger for a growing unknown share, using a denominator such as eligible expense lines. That trigger prompts investigation rather than automatically proving poor staff performance. Preserve original labels and a mapping version so a disputed classification can be traced and revised.
Maintain comparability when the scheme evolves
Trial the rules on difficult records with two reviewers. Investigate disagreement by reading the definitions, not just taking a majority vote. A disagreement may reveal that the source lacks necessary detail. Record the decision and add a new boundary example where useful. Allow source owners to propose changes, but give one role responsibility for approving the shared taxonomy and its effective date.
When splitting a category, decide whether history can be reclassified from retained evidence. If it cannot, show the break in the series rather than implying a clean trend. The linked government reference provides quality background. The method here makes classification choices explicit; it does not claim a universal industry taxonomy or suggest that a neat set of categories establishes the accuracy of the underlying expenses.
For hierarchical groups, verify that child categories roll up to the intended parent exactly once. A transport line tagged both Taxi and Local Travel should not be added twice when calculating the parent total, even if both descriptive tags remain useful.
Fictional expense file: Travel, Transport and Taxi
A fictional scenario to explain the method, not a customer result or a market benchmark.
A fictional organisation imports 100 expense lines worth 10,000 units. The source labels 40 lines Travel, 30 Transport and 30 Taxi. Review shows that Travel includes rail and hotels, while most Transport and Taxi lines describe local passenger journeys. Five lines lack sufficient detail. The team defines accommodation and passenger transport as separate reporting groups.
Line details support 2,000 units of accommodation and 7,500 of passenger transport. Five poorly described lines worth 500 units remain unknown. The report shows the full 10,000 total and a 5% unknown-value share, rather than forcing every ambiguous line into transport.
The mapping retains original labels and records why Transport and Taxi were grouped. If later evidence identifies the five lines, reviewers can reclassify them and explain the revision. The fictional numbers illustrate category design; they do not imply that those labels have the same meaning in every organisation.
TAKE THIS INTO YOUR NEXT CONVERSATION
A practical next-step checklist.
- Define the purpose and unit of classification.
- Write inclusions and exclusions for overlapping labels.
- Separate unknown from understood other cases.
- Version mappings and explain historical breaks.
Common questions.
Can categories and tags be used interchangeably?
Only if their counting rules match. Multiple tags can create overlapping totals that a mutually exclusive classification avoids.
Should every unknown record block publication?
Assess its value, concentration and decision impact. Publish a limitation where appropriate and assign follow-up.
Further reading & source context.
These primary references provide background. They are not evidence for the fictional example above and do not endorse Accuralytix.
Found an error or a missing qualification? Send an editorial correction.
Check my data readiness.
Use a practical tool to turn this idea into a clearer starting point.
Check my data readiness