How many call categories should I actually use?

Design for the decision, not for completeness

The instinct is to build a category for every kind of call anyone can name. That produces forty categories, most holding a handful of calls, none of which changes a decision. It also guarantees inconsistent classification, because the boundaries between forty things are inevitably fuzzy.

The better test for adding a category is whether someone would do something different because a call landed in it. New service demand, existing job follow-up, billing, and solicitation each trigger different actions. Two flavors of the same demand type usually do not.

A two-level structure that survives

Keep the top level small, mutually exclusive and stable. Put everything volatile underneath.

• Level one: what kind of contact this was. New demand, existing customer service, scheduling change, billing, warranty, solicitation, employment, other.
• Level two: the specifics. Trade, system type, symptom, urgency. This is where seasonal and marketing-driven detail belongs.
• Tags rather than categories for cross-cutting facts. Emergency, commercial, repeat contact and competitor mention can apply to several level-one categories at once and should not compete with them.

The other bucket is a health indicator

Every taxonomy needs an other category, and its size tells you whether the taxonomy fits the business. A small residual is normal. A large one means a real call type is not represented, and the fastest way to find it is to read thirty calls from that bucket rather than to theorize.

Reviewing the residual quarterly is a cheap habit that keeps categorization honest. It also tends to surface changes in the business before anyone reports them, which is one of the quieter benefits of running call analysis across every conversation instead of a sample.

Changing the taxonomy breaks history, so plan for it

The moment you split a category or merge two, every trend line that crosses the change becomes misleading. There are two acceptable responses: reclassify the historical calls under the new scheme, which is feasible when the transcripts are retained, or mark the change date on every chart so nobody reads the step as a real movement.

Retaining transcripts is what makes the first option possible, and it is a strong argument for keeping the underlying text rather than only the derived labels. That principle applies well beyond calls; it is the same reason we keep raw joined records behind reporting and behind revenue analysis, so a definition change can be applied backwards instead of forwards only.