The Analytics Decision

YouChews · 27 Aug 2026 · second pass, against what is actually wired

Hassan recommended Mixpanel. Carlos recommended Microsoft Clarity. We already run Google and Shopify. This is the call, and it is graded against Brett's four tests: simple · fast to do · fast for the customer · repeatable by anybody.

I went back and checked the real wiring rather than trusting yesterday's write-up. Two things in yesterday's memo were wrong, and one of my conclusions has changed. Both are corrected below.

1. What we actually have

On the website right now (read off the live page today)

What	Status	Notes
Shopify's own analytics	Live	The system of record. Sessions, carts, checkouts, orders.
Google tag `GT-P8ZTDFHQ`	Live	Installed by Shopify's Google & YouTube app.
Google Analytics 4 `G-NBF6MZRJH4`	Live	Property "YouChews Inc" (467518887).
Google Ads `AW-16667033421`	Live	Conversion tracking.
Klaviyo onsite script	Live	`company_id=T8yfNm`. Loading on the storefront.
Google Tag Manager	Not installed	See correction 1.
Session replay / heatmaps	Not installed	No Clarity, no Hotjar.
Product analytics	Not installed	No Mixpanel, Amplitude, PostHog, Heap.

The Shopify Google app already configures eleven GA4 e-commerce events for us: page_view, view_item, view_item_list, add_to_cart, remove_from_cart, view_cart, begin_checkout, add_shipping_info, add_payment_info, purchase, search. Nobody has to build an event schema. It is already there.

In the data pipeline (Windsor into our own daily report)

Ten connectors live: Meta, Google Ads, Google Merchant, GA4, Shopify, Amazon Ads, Amazon Seller, Instagram, Klaviyo, Walmart. Mixpanel is available in Windsor's catalog and is not connected.

And the funnel is already being pulled every day — tools/kpi_pull.py requests GA4 sessions, add_to_carts, checkouts, ecommerce_purchases, purchase_revenue, bounce_rate daily and again broken out by source. It also pulls Shopify customer order history all the way back to 1 Jan 2025 with customer IDs on every order. We are collecting the funnel and the retention data today; we just are not showing it.

2. What the data says — and this is the part that decides it

The funnel (Shopify, last 30 days)

Step	Sessions	Of previous	Of all
Landed on the site	33,311	—	100%
Added to cart	2,369	7.11%	7.11%
Reached checkout	1,746	73.7%	5.24%
Bought	1,302	74.6%	3.91%

Net sales $48,052.31 · 1,345 orders · AOV $35.96 · 87.9% of sessions are mobile.

30,942 people (92.9%) never put anything in the cart. That is the whole game.

Our counting is already broken in two places, and a third tool would inherit it

GA4's raw event counts for the same 30 days:

Event	GA4 count	What it should roughly be
session_start	28,490	Shopify says 33,311 sessions
view_item	13,560	—
add_to_cart	1,647	Shopify: 2,369 sessions with cart adds
begin_checkout	659	Shopify: 1,746 reached checkout
purchase	1,163	Shopify: 1,302 orders
view_cart	37	effectively not firing
scroll (90% page depth)	2,014	7.1% of sessions

GA4 reports more purchases than checkout starts (1,163 vs 659). That is structurally impossible — you cannot buy without starting a checkout. GA4 is seeing roughly 38% of our checkout starts and about 89% of our purchases. Its add_to_cart is light too.

This is the single most important fact in this document, because Mixpanel would be fed by the same client-side events. A third counter does not fix a fire-rate problem; it reproduces it, and then we have three systems that disagree. We are already fighting utm_source split across meta and facebook, placement written into the medium field, and 25% of orders with no source at all.

Three findings that came out for free, from tools we already own

79% of all traffic lands on the homepage. 22,484 of 28,490 sessions land on /, converting at 3.90%. /products/coffee-chews gets 5,005 sessions and converts at 4.28%. /pages/military is the standout on intent — 210 sessions, 9.52% add-to-cart. Sending paid social to the homepage instead of the product page is a well-known conversion tax. Honest caveat: this is not proof — people who land on a product page may be coming from more specific ads. It is a test worth running with Carlos, not a settled fact.
Returning visitors convert 2.6x better, and there are almost none of them. New sessions: 24,042, buying at 3.74%. Returning: 2,717, buying at 9.72%. Returning traffic is only 9.7% of sessions. The money is in getting people back, and Klaviyo is the machine that does that — its script is live on the site, we pull its data every day, and our rollup currently throws that data away (zero references to Klaviyo in kpi_rollup.py).
Only 7.1% of sessions reach 90% page depth. People are not reading the page they land on. That is a design and speed problem, and it is exactly what a heatmap answers.

3. Two corrections to yesterday's memo

Correction 1 — we have no Google Tag Manager. I called GT-P8ZTDFHQ a GTM container. It is a Google tag ID, not a Tag Manager container (those look like GTM-XXXXXXX), and there is no GTM container on the site at all. This matters practically: there is no clean place to drop a new script. Anything we add means a theme edit or a Shopify app. That raises Mixpanel's setup cost and is a point in favour of using Clarity's official Shopify app rather than pasted code.

Correction 2 — I gave Mixpanel too much credit on retention. Yesterday I said cohort retention was the question that would justify Mixpanel. That was too generous. GA4 has cohort exploration, Windsor exposes GA4's cohort fields to us directly, and more importantly Shopify is the better source anyway — we already pull every order back to Jan 2025 with a customer ID attached. Shopify knows who actually bought, permanently, with no cookie decay. Mixpanel would be reasoning over browser cookies on a site where 85.6% of visitors are strangers. The retention question does not need Mixpanel.

4. What I recommend, in order

Each is graded against Brett's four tests.

1. Make Shopify the official funnel number, and put it in the daily report. Shopify counts carts, checkouts and orders inside its own system — it cannot under-fire. GA4 becomes the slicer (by source, landing page, device, new vs returning), never the source of absolute funnel counts. The five-line query that produced the table above becomes a permanent band in the 5:30am report and a tile on Mission Control. Simple: yes. Fast to do: about an hour, and it is mine. Customer speed cost: zero, no new script. Repeatable by anybody: yes — it lands in our own report, not in a third-party UI somebody has to learn.

2. Test sending paid social to the product page instead of the homepage. 79% of traffic lands on /. This is Carlos's lane and it is a recommendation, not a change. One campaign, destination swapped, control kept live. No tool required. Simple: yes. Fast: same day. Customer cost: zero. Repeatable: yes.

3. Install Microsoft Clarity — as a time-boxed diagnostic, not a permanent fixture. Carlos is right that this is the missing instrument: it is the only thing on the list that shows us what the 30,942 people saw. Free forever, no traffic cap, official Shopify app, roughly 30 minutes. Two conditions I would put on it, both because 87.9% of our traffic is on a phone:

Measure page speed before and after. Session replay watches the page continuously, and that is real work on a cheap phone on cellular. If the page's load time moves, Clarity comes off. Being honest: Clarity is not free of customer cost, it is just cheap.
Run it 2 to 4 weeks, extract the findings, then decide whether it stays. A diagnostic you leave running forever is just another script.
Keep text-masking on so no customer data lands in recordings.
Known limit on our plan: we are on Shopify Advanced, not Plus, so Clarity will see the storefront but not the checkout pages. It aims squarely at the 92.9% no-cart leak, which is the big one. The 25% checkout drop-off (about $15,970 a month) will have to be diagnosed from Shopify's numbers plus someone buying a pack on their own phone and watching where it hurts.

Simple: yes. Fast to do: 30 min. Customer cost: small but real, so measure it. Repeatable: yes, the recordings are just there.

4. Use the Klaviyo data we already pay for and currently discard. Returning visitors convert 2.6x better and are only 9.7% of traffic. Klaviyo's script is already live, both its tables are pulled every refresh, and the rollup drops them on the floor. This is the highest-value analytics work available and it requires no new tool at all. Simple: yes. Fast: half a day. Customer cost: zero. Repeatable: yes.

5. Mixpanel: not now. Reasons, ranked: our client-side events are demonstrably unreliable and it would inherit them; there is no Tag Manager to install it cleanly through; 87.9% mobile makes another SDK a real page-weight decision; and the two questions it would answer are already answered better by Shopify.

5. What Mixpanel is genuinely good for

Hassan's instinct is not wrong. It is right for a different shape of problem, and it is worth saying exactly what Mixpanel is better at, because none of this is a knock on the tool.

Asking questions you did not plan for. Mixpanel keeps raw events and lets you build any funnel or segment after the fact. GA4 makes you set things up in advance and is clumsy about retroactive questions.
Speed and self-serve — and this is Brett's own "repeatable by anybody" test. A non-analyst can build a funnel in Mixpanel in about a minute. GA4's Explorations are genuinely hard and most people give up. On that axis Mixpanel beats GA4 outright. Our answer is to make the repeatable surface our own daily report, which we control, rather than any third-party UI — but Hassan's point stands.
No sampling. GA4 samples big explorations. Mixpanel does not.
One identity across web and app. If there is ever a YouChews app, Mixpanel stitches app and web behaviour into one user. Nothing we have does that.
Raw event export without standing up BigQuery.

Why that does not fit us today, in one line: mobile apps hand you a logged-in user ID on day one and have dozens of meaningful in-product actions. Our storefront has a four-step funnel, eleven events already standardised, and 85.6% of visitors are strangers who never come back. Mixpanel's power is in analysing identified, returning users doing many things. We have 2,717 returning sessions a month and four things to do.

6. When to introduce it

Not on a date — on triggers. Any two of these and I would pilot it on the free tier, which costs nothing but attention:

There is a login. A subscriber account portal, a rewards programme, a personalisation quiz. The moment there is a user ID at the door, Mixpanel's identity model works and its cohorts become real rather than cookie-guesses.
There is an app, or web plus app that need one view of the same person.
Returning sessions pass roughly 25 to 30% of traffic. Today: 9.7%. Below that there is not enough repeat behaviour to analyse.
There are more than about ten meaningful actions worth measuring behind that login.
Somebody owns it weekly. Mixpanel is only valuable if a person builds and reads boards. Hassan is the natural owner, and it is a SaaS tool, so being remote is no obstacle at all.
And never before the counting is fixed. Adding a third client-side counter while begin_checkout under-fires by 2.6x makes reconciliation harder, not easier.

On cost, so it is not a factor anyone has to guess about: at our volume we would sit inside Mixpanel's free 1M-events-per-month tier (their pricing moved to per-event in Feb 2026). Their autocapture click-tracking could blow past that, so it would need capping. The expense was never dollars — it is attention and a page-weight budget on a phone.

The one assumption most likely to be wrong

That the aggregate funnel is what we need. If what we actually want is the individual journey — "this person saw the Military ad, came back three times over nine days, then bought" — Mixpanel does beat GA4 at that, because GA4 samples, lags two to three days on attribution, and is poor at per-person paths. Our counter-argument is that the order is the identity anchor and Shopify already holds it. If the plan involves a login in the next quarter, this whole call should be re-run.

Bet #282 (open): the binding constraint is the 92.9% nobody can see, not the counter. Clarity names at least one concrete, testable page defect within 14 days. Wrong if after 14 days we cannot name a page change to test, or if a question we actually need answered needs per-user cross-session sequencing that Shopify and GA4 provably cannot do.

All figures pulled live on 27 Aug 2026. Shopify funnel and sales: 30-day window to today. GA4 events, landing pages, device and new-vs-returning: 27 Jul to 25 Aug, held two days back because GA4 attribution restates for about 48 hours. Site tag stack read directly off youchews.com. Mixpanel and Clarity pricing and capability from their own current documentation.

← All documents