Tags: analytics commerce concept

Channel Taxonomy

Date: 2026-08-16


The rules turning raw sources into the handful of channels anyone reports on. Every tool ships a default set, nobody reviews it, and it quietly decides which channel gets credit for what.


What it is

A channel taxonomy is the classification mapping traffic sources to reporting groups.

RAW SIGNAL                              CHANNEL

source=google, medium=cpc          →   Paid Search
source=google, medium=organic      →   Organic Search
source=facebook, medium=paid       →   Paid Social
source=facebook, medium=referral   →   Organic Social
source=klaviyo, medium=email       →   Email
source=(direct), medium=(none)     →   Direct
source=partner-site.com            →   Referral

The mapping is rule-based, applied at query or ingestion time, and almost always inherited from a tool’s defaults.

Why the defaults fail

Every tool ships a channel grouping. It’s a reasonable general-purpose guess and it’s wrong for most retail sites in predictable ways:

  • Email lands in Referral or Direct when links lack parameters, or when a webmail client strips them
  • Paid social falls into Organic Social when the platform’s auto-tagging isn’t recognised
  • Affiliate traffic looks like Referral, so it isn’t separable from genuine editorial links
  • Your own app or emails count as external traffic
  • Price comparison and marketplaces land in Referral alongside blogs, which are entirely different commercially
  • Branded and non-branded search are merged, hiding the single most important distinction in paid search reporting

That last one matters most. Branded search converts far better and is largely demand you already created — merging it with non-branded makes paid search look uniformly efficient and hides where the incremental value is. See Incrementality Testing.

Designing your own

Three principles:

Group by decision, not by source. The channels should be the units you’d move budget between. If Referral contains affiliates, comparison sites and blogs, it’s not a channel — it’s a bucket.

Separate paid from organic, always. Same platform, entirely different economics. facebook / cpc and facebook / referral are not one channel.

Split branded from non-branded search. Two lines in the taxonomy, and it changes how paid search is read permanently.

A workable retail set:

Paid Search — Brand        Organic Search
Paid Search — Non-brand    Organic Social
Paid Social                Email
Paid Shopping              Referral
Display                    Affiliate
                           Marketplace
                           Direct

Where it should live

In the warehouse, not in the tool. A tool’s channel grouping is applied at ingestion and often can’t be restated — change the rule and history doesn’t update, so you get a seam.

Defined as SQL over source, medium and campaign, the rule is version-controlled, reviewable, and retroactive: fix a mapping and every historical report corrects. That single property is the strongest argument for Warehouse-First Analytics that most teams actually feel.

Maintaining it

  • A default rule that flags rather than absorbs. Unmatched traffic should land in “Unclassified” and be visible, not be swept into Referral. An unclassified bucket that grows is a new source nobody told you about
  • Review it when a channel is added, not quarterly
  • One taxonomy across every report. Two definitions of “Email” in two dashboards is Metric Drift with extra steps
  • Depends entirely on parameter discipline upstream — a taxonomy can only classify what the tagging provides — UTM Governance
  • Annotate changes. Reclassifying affiliate out of Referral moves revenue between channels overnight, and it will be read as performance — Annotation and Change Logs

What it can’t fix

The taxonomy classifies what arrives. It can’t recover a source that was never recorded — traffic with no referrer, stripped parameters, or an identifier that expired lands in Direct regardless of how good your rules are. See Direct Traffic and Lost Referrers.