A persona taxonomy is the classification layer over a library of research personas: a small set of codes, drawn from published reference schemes rather than invented in-house, that says where a persona lives, what industry it works in, and where it can be reached. It is the difference between a folder of character sketches and a panel you can select from.
Most persona libraries have no such layer, and the cost shows up late. A team can read fifty personas and still be unable to answer “give me the suburban parents and the enterprise operations leads, and tell me which channels reach both”, because nothing in the documents is comparable across documents. Prose does not compose. Codes do.
What a taxonomy has to survive
Three tests, and a scheme that fails any of them is decoration.
Selection. Can you assemble a cohort by writing a filter rather than by reading every file? If the answer requires a human to interpret the persona’s prose, the library scales to about forty personas and then stops.
Repetition. Can the same filter, run in six months, produce the same cohort? That requires stable identifiers and content that does not change under you.
Translation. Can the cohort be described to someone outside the research team in terms they already use? A media planner does not want “urban creative types”. They want a segment code, an industry, and a channel.
The three layers worth having
Klingbar’s library carries three, and they were chosen because each answers a different question that the other two cannot.
Residential: Esri Tapestry
Consumer personas carry an Esri Tapestry segment — a geodemographic classification of US neighbourhoods at block-group level. The 2024 vintage the library pins carries 67 segments, organised into 14 LifeMode groups and six urbanization groups.
The reason this layer earns its place is that it is a residential anchor rather than an attitudinal one. Attitudes are what a persona document is for; where somebody lives, and what kind of neighbourhood that is, is a fact that geodemographic data already models better than a persona author can. It also travels: a Tapestry segment means the same thing to a media agency as it does to the research team, which is the translation test above.
Industry: NAICS 2022
Business personas carry a six-digit NAICS code from the 2022 revision — the Census Bureau’s current classification, whose national-industry level holds 1,012 six-digit codes. Six digits is a deliberate choice. Two-digit sectors are too coarse to distinguish a hospital’s procurement lead from a clinic’s, and anything coarser than the industry level puts personas in the same bucket whose buying processes have nothing in common.
The higher levels are not discarded, they are derived. Sector, subsector, and industry group all fall out of the six-digit code by truncation, so the library never stores a sector that disagrees with its own industry code.
Reach: channel propensity
Every persona in the library — consumer and business alike — scores all 19 marketing channels on a four-point scale: high, medium, low, or minimal. The channels run from the obvious social and search placements through podcasts, connected TV, out-of-home, direct mail, email, SMS, in-app push and print.
The rule that makes this layer usable is that no persona may omit a channel. A missing score and a minimal score are different claims, and a library that allows omissions forces every downstream reach calculation to guess which one it is looking at. Scoring all 19 every time costs the author about a minute and removes an entire class of silent error.
Authored codes, derived facts
The load-bearing design rule is that an author writes exactly one code per axis and the system computes everything else.
The persona file declares a Tapestry segment and, for business personas, a NAICS code. The loader derives the segment’s name, its LifeMode group, and its urbanization group; and the industry’s sector, subsector, industry group and description. None of those derived fields is writable in the file.
This matters more than it looks. Every hand-entered classification is an opportunity for a persona to claim a LifeMode that its segment does not belong to, and nobody notices, because the two fields are forty lines apart in a file nobody re-reads. Derivation makes that failure unrepresentable. The general principle is the same one behind an org chart written as code: declare the authority once, compute the consequences, and never let two copies of the same fact drift apart.
The second half of that rule is membership validation. A code that is not in the authoritative table is an error, not a warning — the library’s validation gate refuses a persona whose segment or industry code does not resolve. This is the check that catches transposed digits, and a typo in a NAICS code is not detectable any other way: 541511 and 541151 are both plausible-looking strings, and only one of them is custom computer programming services.
Naming the vintage
Both schemes here are versioned by their owners, and the versions are not compatible. NAICS is revised every five years; Esri revises Tapestry periodically and publishes the changes. A library that says “Tapestry segment 1A” without saying which Tapestry has recorded a code whose meaning may have moved.
So the vintage is part of the specification: 2024 Tapestry, 2022 NAICS, stated in the schema document and enforced by the embedded reference tables. When a scheme’s owner ships a new vintage, migrating is a deliberate project — remap the codes, re-validate every persona, bump versions — rather than something that happens quietly when a reference file is refreshed.
Identity and immutability
The taxonomy is only reproducible if the things it classifies hold still. Two rules do that work.
Identifiers are permanent and never reused. Every persona has a stable id — b2c-<generation>-<archetype>-NNN for consumer personas, b2b-<tier>-<role>-NNN for business ones — and no id is ever reassigned to a different persona. A study that recorded which personas took part recorded something that still means what it meant.
Published content is immutable. An edit does not rewrite a persona; it publishes a new version alongside the old one. Removing a persona from the registry archives it rather than deleting it, so the references a completed study holds do not dangle.
Together those give the repetition test a real answer: a cohort selected by filter last quarter can be reassembled exactly, and the documents its members were played from can be produced byte for byte.
What a taxonomy cannot do
It cannot make an invented persona representative. Classifying a synthetic subject as Tapestry 1A does not import the behaviour of the households in that segment; it records the author’s intent about who this subject is meant to resemble. The codes make a library selectable and describable. They do not make its contents evidence — that distinction is the subject of the guide on synthetic respondents.
It also cannot substitute for a well-written persona. A rich taxonomy over thin documents produces a panel that is easy to assemble and useless in a session.
Building one of your own
If you are adding a taxonomy to an existing library, the order that works:
- Pick published schemes, not private ones. A scheme somebody else maintains comes with definitions, updates, and an audience that already understands it.
- One authored code per axis. Everything derivable, derive.
- Name the vintage in the schema document, and pin the reference tables next to the code that reads them.
- Score reach exhaustively. Whatever your channel list is, every persona scores all of it.
- Make membership a build failure. A code that does not resolve should stop a release, not log a warning that scrolls past.
- Freeze published content and version edits, so a cohort is reproducible rather than merely re-selectable.
The result is a library where “who should be in this study” is a question with a written answer, and where the answer can be given twice. The persona library is browsable without an account if you want to see the shape of it before building your own.