Skills CX operations

Designing a contact driver taxonomy

A taxonomy exists to answer one question: what should we fix to remove this contact? Most taxonomies cannot answer it, because they were built from the helpdesk's default categories or the org chart rather than from why customers actually get in touch. The result is a category list that describes support and prescribes nothing.

CX operationsDemand and capacityPlaybookAny helpdeskRead-only
Installnpx rulebase-skills install cx-contact-driver-taxonomy

When to use it

Reach for this when someone says any of these — they are the phrases the skill itself triggers on:

  • build a contact reason taxonomy
  • our ticket categories are useless
  • too many tickets are tagged Other

How it works

The method, in the order the skill runs it. The full procedure — tables, worked examples and the edge cases — is in the skill itself.

  1. The distinction that makes a taxonomy useful

    A topic is what the contact was about. A driver is the cause you could remove.

  2. Step 1: derive it bottom-up from real conversations

    Do not start from a template, a vendor list, or the helpdesk defaults. Read conversations and write down why each one happened.

  3. Step 2: structure it in two levels, not four

    Two levels. 15–40 drivers total. Deep hierarchies fail in a specific way: agents select a level-one value and stop, so level three is empty and useless. Beyond roughly 20 visible options at the point of selection, accuracy collapses regardless of how good the definitions are.

  4. Step 3: test every candidate driver

    Drop or rewrite anything that fails.

  5. Step 4: instrument the "Other" rate as your quality metric

    The share of contacts landing in "Other" is the single best measure of taxonomy health. Target under 5%; above 10% the taxonomy does not fit its contacts.

  6. Step 5: normalise before you compare anything

    Raw contact counts per driver are almost always misleading, because they track business growth. A driver rising 20% while the customer base rises 25% is improving.

  7. Step 7: version it, and never renumber

    Taxonomies must evolve, and every change breaks comparability if handled badly.

  8. AI labelling changes the economics, not the principles

    100% labelling coverage becomes affordable, which means you can afford more granularity than a human-selected dropdown allows. It does not relax the design rules — it tightens two of them.

Related skills

Free and open source, and vendor-neutral — it reads the conversations from whichever helpdesk you already run. Browse all 149 skills · connect your helpdesk over MCP · source on GitHub

Review every conversation. Act on what it finds.

AI for customer operations, built for financial services. Specialist agents chase every issue to resolution and every stalled customer to activation.

Rulebase dashboard