Week 6 · Web Metrics and Google Analytics II

Design a five-layer measurement plan against a live Google Analytics property, then go layer by layer and find out — honestly — what it can and cannot answer. This is your Measurement Plan and Gap Audit.

At a glance

Time	70 minutes, in pairs
Work with	A partner
You need	The Google Analytics demo account (Google Merchandise Store — your facilitator projects the access link) + an agent (Claude or Manus) for Step 5
You hand in	A one-page measurement plan and gap note, plus a supervision note (portfolio evidence)
Graded on	Specificity — a numeric target, every event labelled with how it arrives, and a gap note that names an event, not a feeling

The scenario

You are not given a file this week — you are given a live property. Google publishes a permanently available demo account with real traffic from its own merchandise store. It needs no administrator rights and costs nothing, and it is the same property most GA certification practice material uses, so this lab doubles as preparation for Assessment 2.

Work in your capstone role, advising the Director of Digital Growth at Northstar Living, on your own team’s data if it helps. There is no data dictionary for this lab, and that is deliberate — part of the work is finding out what the property does and does not collect, which is exactly what happens when you inherit somebody else’s analytics setup.

Before you start

Confirm all three, and pair up with a working group rather than losing lab time to access problems if any of them fail:

• Reports loads without an error
• The Reports snapshot shows a non-zero user count
• The date selector is set to the last 28 days

Step 1 — Write the question first · 10 min

Write one decision question — not a topic, a decision. The test: would a plausible answer change what somebody does on Monday?

Too vague: “How is the website performing?” Better: “Should we move next month’s promotion budget from desktop display to mobile?”

My scenario is...
My decision question is...
Dimension(s) I need...
Metric(s) I need...

Step 2 — Build the five-layer measurement plan · 15 min

Fill all five layers, one page, before you open the tool:

• Layer 1 — Who. Which audience attributes matter to this decision? Name three to four (device, first-time vs returning, geography, language, traffic channel).
• Layer 2 — What. Four to six events on the journey to the outcome, in order. For each, note whether it would be automatically collected, an enhanced measurement event, a recommended event, or a custom event.
• Layer 3 — Why. The single key event for this decision, with a target stated as a number. A target with no number is a hope.
• Layer 4 — Access. Which attributes would show whether the outcome is worse for part of your audience? Interface language and assistive-technology use are both audience attributes, and both are invisible unless somebody chose to collect them.
• Layer 5 — Quality. Two falsifiable checks that would tell you the data is wrong. “Every begin_checkout should be followed in-session by either a purchase or an abandonment” is a check; “the data should be accurate” is not.

Step 3 — Take the plan into the property · 15 min

Layer	Where to look
1 · Who (device, geography)	Reports → User → Demographics / Tech
1 · Who (channel)	Reports → Acquisition
2 · What (events)	Reports → Engagement → Events
3 · Why (key event)	Admin → Events → key event toggle

Go layer by layer and mark each item present, absent, or present but not usable, with one line saying why.

Expect gaps

The demo property collects far more than most real properties, and it will still not have everything your plan asks for. That is the normal result, not a failure of your plan. A measurement plan that only asks for what already exists is not a plan — it is a description.

Step 4 — Find the (other) row · 10 min

Go to Reports → Engagement → Pages and screens, and scroll to the bottom of the table.

• Found an (other) row? Note what share of total views sits inside it, and what that means for anyone asking about a specific low-traffic page.
• No (other) row? Widen the date range to twelve months, or add a secondary dimension, and look again — both increase the number of distinct rows the query has to return.

Also check the data-quality icon near the report title. Hover it, and note whether the report is complete, thresholded, or sampled.

The (other) row held ...% of views.
The data quality indicator said...

Why this step exists

You now know something most people using this tool do not: the bottom of a report is not the bottom of the data.

Step 5 — Supervise an agent · 15 min

🔗 On your agent platform

Open Claude or Manus and paste your decision question and five-layer plan.

Here is my decision question and my five-layer measurement plan. [Paste them.]
For each item in Layer 2, tell me whether Google Analytics 4 collects it
automatically, through enhanced measurement, as a recommended event, or only
as a custom event. Cite the Google documentation for each answer. Do not
guess; where you are not certain, say so.

Expect a table that is mostly right, with at least one confident error — recommended and custom events are the two an agent is most likely to confuse, because the distinction is a naming convention rather than a technical one.

Pick two rows and check them against the Google documentation yourself, using the sources listed on the last slide of the unit. Do not accept a citation you have not opened.

The agent said...
I checked...
It was [correct / wrong] because...

Fluency is not accuracy

The agent is fluent about this tool because the documentation is public and abundant. The failure mode here is not gibberish — it is a plausible sentence with the wrong category in it, exactly the kind of error that survives into a client deck.

Step 6 — Write the gap note · 5 min

Three sentences, no more:

1. The one thing your plan needs that this property cannot answer today.
2. What would have to change for it to answer it — name the event and the collection type.
3. Who would have to do it, and roughly when in the project it would have to happen.

Verify & submit

• My decision question would change somebody’s action if answered
• All five layers are filled and my key event has a numeric target
• Every Layer 2 event is labelled with its collection type
• I checked at least two of the agent’s collection-type claims against Google’s own documentation
• I looked for the (other) row and recorded what I found
• I noted the data quality indicator on at least one report
• My gap note names an event, not a feeling

Then attach a one-paragraph supervision note: which agent platform you used, one claim you verified and the documentation page you checked it against, one thing the agent got wrong (or correctly flagged as uncertain), and one question about this property the agent could not answer from documentation alone.

Save your measurement plan with your other unit evidence for the Individual Analytics Portfolio before you leave.

Debrief prompts

Worth raising with your pair or the room: What did your plan ask for that the property could not deliver? Did anyone’s (other) row hold more than a few per cent — and what would you now say to a colleague who called a page “no traffic”? Where did the agent sound most confident and turn out to be wrong?

Capstone connection

This lab already ran in capstone terms — you worked as the analytics team advising the Director of Digital Growth at Northstar Living. The certification sits on the same ground you covered today; the mid-trimester break is the sensible week to take it.

Companion links: Unit 13 app · lesson plan