Executive Summary: Marketing Taxonomy Governance

The guide in five minutes, for a marketing leader deciding whether to fund the work. No implementation detail.

The problem in one sentence

If your analytics shows facebook, Facebook, fb and FB as four separate sources, every report that groups by source splits one channel into pieces, and every budget decision made from those reports rests on numbers nobody has added up.

What it costs

The visible cost is analyst time: hours spent merging spellings, decoding campaign names nobody remembers, and rebuilding reports that should have arrived clean.

The other cost is budget moved on split numbers. A composite of teams we’ve worked with shows how.

A 40-person B2B software company selling project management software spent about $180K a quarter on paid channels. Its LinkedIn traffic sat under four spellings: linkedin, LinkedIn, li and linked-in. In the Q3 pipeline review, Priya, the VP of Marketing, asked: “Should we shift budget from LinkedIn to Google next quarter?” Megan, the analyst, couldn’t say how LinkedIn had done without hours of cleanup.

The room made the call anyway. $25K moved from LinkedIn to Google.

Then the company wrote down a controlled vocabulary for each parameter and moved link-making into one builder whose dropdowns offered only those values. Within one quarter, duplicate source entries in its reports went from 47 to zero. With the history consolidated, LinkedIn turned out to have produced three times the pipeline per dollar of paid search.

The $25K had gone to the weaker channel.

“Data quality” is the polite name for this. It’s a budget allocation problem. Chapter 3, How a Value Behaves, tells the story in full.

What a marketing taxonomy is

Every tagged link carries a short letter about the campaign behind it: the channel, the source, the initiative, the product, the launch month. Each fact is something a report groups by. The link is one envelope for that letter. Others carry it too: the campaign name in an ad platform, the source fields on a lead in the CRM.

The taxonomy is the letter plus its rules: which dimensions you record, the approved values for each, what each value means, and who may add one. Defining it, encoding it and keeping it enforced is taxonomy governance.

The taxonomy is the asset, because it keeps paying out after it’s built:

• It outlasts tools. When GA4 replaced Universal Analytics in 2023, a team’s approved values and their meanings could move across unchanged.
• It compounds. Years of values that mean the same thing compare with each other.
• It’s a shared language. A new hire, an agency and an analyst read the same value and take the same meaning from it.

Links are disposable. The agreement about their values is what you keep. The Introduction builds the idea.

What good looks like

One month of traffic as analytics recorded it, beside the same month with the spellings merged:

Example numbers

As recorded

Source	Sessions	Conversions
google	4,291	312
Flagged: Facebook	1,843	87
Flagged: facebook	782	41
Flagged: fb	234	12
Flagged: LinkedIn	567	38
Flagged: linkedin	312	22
Flagged: li	89	4
Flagged: newsletter	1,204	95
Flagged: Newsletter	445	31
(direct)	8,432	201
(not set)	2,105	44

Eleven rows: three for Facebook, three for LinkedIn, two for the newsletter.

Spellings merged

Source	Sessions	Conversions
google	4,291	312
facebook	2,859	140
linkedin	968	64
newsletter	1,649	126
Flagged: (direct)	8,432	201
Flagged: (not set)	2,105	44

One row per source. The flagged two are visits analytics couldn't trace to a source.

Each channel now reads as one number. The last two rows don’t move: a visit with no source has nothing to merge. Question 4 below asks about that share.

Making the case

The pair above is the exhibit. It’s strongest from your own data: last quarter’s report as it stands, beside the same report with the duplicates merged and the untraceable share shown. Beside it, one decision the merged version would have changed.

The costs above add one more: the cleanup grows. Each new person, agency and campaign can bring its own conventions, so every month without rules adds variants. Some of the damage can’t be undone. A value that held two meanings, like social for paid and organic posts at once, can’t be split apart afterwards (Chapter 10, Ownership and Change).

Today the work happens after the fact: in cleanup hours, and in budget decisions made on numbers nobody fully trusts. Governance moves that work upstream, from cleanup that never finishes to rules set once and held where links get made.

A pilot is the smaller ask. A proposal to govern the whole organization at once competes with everything else on the roadmap. One channel for one quarter doesn’t: its approved values, its links made in one builder that offers only those values, and then that channel’s report before and after. It’s the same exhibit, produced on purpose.

Three objections come up, each with a reply:

• “This is too rigid.” Picking from a menu is quicker than typing a campaign name from memory, and nothing on the menu is misspelled.
• “We’ve always done it this way.” Pull up one source’s spellings, or a campaign name nobody can decode six months later. The way it’s always been done produced those rows.
• “It’s extra work.” It’s less work, earlier. Picking values takes moments. Cleaning up a quarter of inconsistent tagging takes hours, and part of it never comes clean.

The return: four metrics that make the case

The exhibit opens the conversation. Four before-and-after numbers can show whether the work is paying off:

• Fragmentation rate: distinct values in a parameter divided by the real number of things they name. For example, 31 utm_source values for 8 real sources is 3.9x; 1.0x means no duplicates.
• Cleanup hours: hours a quarter spent reconciling campaign data before it’s usable.
• Attribution coverage: the share of conversions with source, medium and campaign all present and approved.
• Time to answer: how long “which channel drives pipeline?” takes to answer.

Measured on today’s data and again after one governed quarter, they sit side by side like this. These are example numbers:

Metric	Before	After
Fragmentation rate	3.9x	1.1x
Cleanup hours	12 a quarter	2 a quarter
Attribution coverage	61%	94%
Time to answer	about 3 days	about 15 minutes

Numbers are half of it. A log of budget decisions the clean data made possible gives each quarterly update a line about money, not about data. The guide’s stories make sample entries:

• “LinkedIn produced three times the pipeline per dollar of paid search, which we couldn’t see while it sat under four spellings.”
• “Conversion campaigns on bedding in the US returned 4.2x the ROAS of awareness campaigns on the same products. We moved $200K of quarterly spend.”

What attribution means for your budget

A model decides who wins the credit for a sale; the taxonomy decides who is in the race. A paid ad shipped untagged reads as free under every model, and a campaign under two spellings splits its own credit under every model, so no choice of model repairs either (Chapter 11, Attribution).

Five questions for your team

You don’t need the implementation. You need five answers.

1. “Do we have one list of approved values, and where do links get made?” A list in a shared sheet is a rule people have to remember. A builder whose fields offer only the list’s values holds it at the moment a link is made. A free URL builder assembles a link correctly, but its fields are open: it takes Facebook as readily as facebook (Chapter 9, Where Links Get Made).
2. “Who owns our taxonomy?” A named owner who makes each change and keeps the record, alone or for a council that decides, and who runs the audit (Chapter 10, Ownership and Change). Without one, the rules still change, in whoever’s hands.
3. “Can we answer ‘which channel drives pipeline?’ without manual cleanup?” If it takes days of analyst work, each budget decision waits for it or goes ahead without it, the way the $25K did.
4. “What share of our traffic arrives untraceable?” Look at the (direct) and (not set) rows. Merging spellings won’t shrink them. Stride, the fictional shoe brand the guide follows, started with more than 30% of its traffic under (direct) / (none) (Chapter 1, What Do You Want to See?).
5. “Are we tagging links in every channel?” QR codes, partner links, podcast show notes and organic posts are links too. One that goes out untagged can’t say which campaign sent the click (Chapter 2, How Analytics Reads a Visit).

What the work is

The guide splits the work into five parts, always in this order:

Part	The job	Chapters
Dimensions	Decide what to measure, from the report backwards.	1
Rules	Approved values and naming conventions that encode the dimensions.	2, 3, 4, 6
Workflow	Links made in one place, from the approved lists.	5, 9, 10
Auditing	Catch drift before it reaches a decision.	8
Automation	Carry the same values into the CRM and the warehouse.	7, 11

Where links get made is the part with a range of costs, from a free builder whose fields are open to a platform with approvals. Chapter 9 sets each rung’s cost beside what it refuses.

---

For the whole picture, start with Chapter 1: What Do You Want to See?, or go to Chapter 3: How a Value Behaves if you’re ready to build.