Attribute tagging and why taxonomies drift
Attribute tagging is labelling items with the characteristics a business wants to analyse by: format, flavour family, claim, price tier, occasion. It is how a brand asks a question the standard category hierarchy cannot answer.
It is also work that has to be maintained, and a taxonomy nobody maintains gives confident answers about a structure that no longer matches the market.
What makes a good attribute?
That it is observable, mutually exclusive within its dimension, and stable enough to be worth defining. An attribute that requires a judgment call on every item will be applied inconsistently by whoever happens to be tagging.
Define the values before tagging anything. A dimension with an open-ended list of values is a free-text field with extra steps.
Where does drift come from?
New products that do not fit existing values, values that blur as the market changes, and different people tagging to different interpretations. All three accumulate quietly.
The symptom is a growing "other" bucket. When a value that should be small starts absorbing a meaningful share of the category, the taxonomy has stopped describing the market.
How do you keep it current?
Review on a cycle, with a rule for introducing a new value and a rule for retiring one. Both rules matter: a taxonomy that only grows becomes unusable, and one that never grows misses everything new.
Record when a value was introduced. A share series for a value that did not exist for part of the history is not a series.
What breaks when a value is redefined?
Every comparison across the change. Redefining a price tier or a format boundary retrospectively rewrites history; doing it prospectively creates a break. Neither is free, and the choice should be deliberate and documented.
The usual best practice is to keep the old value alongside the new for a transition period so both series exist.
Who should tag?
Someone close enough to the market to judge and disciplined enough to follow the definitions. That combination is rarer than it sounds, and the most common failure is tagging delegated to whoever has capacity.
Where volumes are large, a documented process with sample auditing beats either heroic individual effort or unreviewed automation.
How do brand and market taxonomies differ?
A brand's taxonomy serves its own strategy; a provider's serves a general market view. They will not match, and forcing one onto the other loses information.
The workable arrangement is to hold both: the provider's standard hierarchy for comparability and your own tags for strategy, mapped rather than merged.
What should be documented?
Each dimension, its values, the definition of each value, the date it was introduced, and worked examples of edge cases. The edge cases are the part that makes tagging consistent between two people.
How do you audit tagging quality?
Sample and re-tag independently, then compare. Agreement rates below a comfortable level mean the definitions are ambiguous rather than that the taggers are careless, and the fix is in the definitions.
Frequently asked questions
How many attributes is too many?
More than are used. Every dimension is maintenance, and a taxonomy with dimensions nobody analyses by is cost without return.
Can a provider tag to my taxonomy?
Some will, as a service, and it is worth asking. The trade-off is that maintenance then sits outside your control, so the definitions have to be unusually clear.
What about attributes derived from the name?
Useful as a first pass and unreliable as a final answer. Product names are marketing text and they change without the product changing.
How do I handle an item that fits two values?
Change the dimension so it cannot, or define a precedence rule. An item in two values of one dimension makes every total double-count.
Should the "other" bucket be allowed?
Yes, and monitored. It is a useful safety valve and a growing one is a signal, which is exactly why it should be reported rather than hidden.
Does tagging help with competitor analysis?
Considerably — it is the main way a brand sees the market through its own strategic lens rather than the provider's standard one. That is usually the whole reason for doing it.
How often should the taxonomy be reviewed?
Annually at minimum, and whenever the "other" bucket grows or a genuinely new segment appears. Waiting until a report looks wrong is waiting too long.
Can two dimensions overlap?
They can and it creates confusion. Dimensions should be conceptually independent, so that a cross-tabulation of two of them is meaningful.
What is the first thing to fix in a neglected taxonomy?
The definitions document. Most tagging inconsistency traces back to definitions that were never written down, and writing them is cheaper than re-tagging.
### How should a taxonomy be versioned?
With a version number, a date, and a change log listing values added, retired and redefined. Any analysis should record which version it used.
Without versioning, a share series across a taxonomy change is a comparison between two different structures presented as one series. That is the single most common way a tagging system quietly produces wrong answers. Standards practice around product identification, such as GS1, is a useful model for disciplined value sets.
How many people should be allowed to tag?
Few, and trained. Tagging spread across many hands produces inconsistency that no definitions document fully prevents, and consolidating it is usually cheaper than auditing it.
Working with retail data? See NRS Insights data solutions or contact NRS Insights.