CRO Deep Dive — Last 90 Days
internal · MUD\WTR growth
Site-wide CRO · last 3 months vs. the prior 3 months

You ran ~19% less traffic — and the store got better at converting it.

Sessions are down 19% and orders down 8%, but conversion rate is up 12% and revenue-per-visitor up 10%. The store is doing more with less. The whole gain is a paid-media efficiency and landing-page-routing story — not an on-site testing story. Almost none of the 25 A/B tests this quarter actually won. Here's exactly what moved, page by page and channel by channel.

WINDOW: Mar 8 – Jun 8 2026 (vs. Dec 8 – Mar 8) SOURCE: Intelligems (site, LP, channel, 25 experiments) 85% MOBILE
2.52%
Site conversion rate
▲ 12% vs prior 90d
$1.27
Net rev / visitor
▲ 10%
$2.40M
Net revenue
▼ 10%
47,490
Orders
▼ 8%
2.19M
Sessions
▼ 19%
Week over week · last 8 weeks (net revenue)
The last two weeks are real — but it's a traffic drop, not a conversion drop
Read this first: the 90-day headline above (conversion up vs. the prior quarter) is still true — but it hides a clear downtrend within the quarter. Revenue peaked late April (the Strawberry Matcha launch week) and has slid since. The most recent week is the lowest of the eight. Here's exactly what's driving it.
$163k3,261 ord
$206k3,982 ord
$192k3,934 ord
$196k4,027 ord
$161k3,337 ord
$174k3,430 ord
$154k3,104 ord
$129k2,607 ord
Apr 13
Apr 20
peak
Apr 27
May 4
May 11
May 18
May 25
Jun 1
Week ofVisitorsOrdersNet revenueConv rateRev/visitor
Apr 13150,4813,261$163,1122.17%$1.08
Apr 20 · peak125,6253,982$206,2233.17%$1.64
Apr 27130,8893,934$191,5103.01%$1.46
May 4142,3524,027$195,9092.83%$1.37
May 11136,5663,337$161,3852.44%$1.18
May 18138,9423,430$173,5202.47%$1.24
May 25119,6463,104$154,3462.59%$1.29
Jun 1 · latest115,2322,607$128,5142.26%$1.11

Note: the Jun 1 week ends today (Jun 8) and the final day or two may be slightly under-reported in Intelligems — but the downtrend is consistent across the prior 3 weeks, so it's real regardless. Pacific time.

What's actually driving the last 2 weeks down (last 14 days vs. the prior 14): revenue −16%, orders −16%, but conversion rate is essentially flat (−2%). So this is a volume problem, not an on-site conversion problem — and it's remarkably concentrated:

Rise 2 landing page is almost the entire story. Its traffic fell 34% and its conversion slipped 12% — together that cut Rise 2 orders 42%, which alone accounts for nearly the whole site-wide order decline. Rise 2 is fed mostly by Meta.
Paid Social (Meta) drove it at the channel level: traffic −15%, conversion −10%, orders −24%. That's roughly two-thirds of the drop. A handful of A/B tests are live on Rise 2 right now (CTA copy, the new "Sophia" page at 100%, "Spicy") — losing variants would depress its blended conversion exactly as we see; worth checking they aren't making it worse.
Homepage also softened (orders −20%).
Partial offset: the compare-listicle page grew (orders +21%), and the OG tin PDP held orders flat while improving conversion +34%. The healthy parts of the funnel are fine — the bleed is specifically Meta → Rise 2 volume.

What changed, in one read

You deliberately bought less traffic, and the traffic you kept converts much better. Paid sessions were cut hard — Paid Social −23%, Paid Search −29% — yet Paid Social orders rose 18% and its conversion rate jumped +50%. That single channel is responsible for almost the entire site-wide CVR gain. Net revenue only fell ~10% on a 19% traffic cut, and revenue-per-visitor — the number that actually matters for a paid, subscription business — is up 10%.

The uncomfortable part: this improvement did not come from your A/B testing program. Of 25 tests in the window, exactly one cleared a 90% confidence bar on any primary metric (and only barely). The conversion lift is coming from (1) sharper media buying and (2) a landing-page strategy that now routes paid traffic through Rise 2 and a new advertorial "compare" page instead of straight to the product page. Those are the things working. The on-site experiment engine is running fast but mostly producing noise — that's the fixable gap.

Top landing pages
Where traffic lands — and what each page did to conversion

Ranked by sessions over the last 90 days. CVR Δ and RPV Δ compare to the prior 90 days. Green = improved, red = declined. The story is concentrated: one page (Rise 2) drove most of the gain, one page (the OG tin PDP) drove most of the drag, and a brand-new advertorial page quietly became your #2 entry point.

Landing pageSessions 90dCVRCVR ΔRev/visitor Δ
/pages/rise-2#1 paid LP — the workhorse709,7582.20%▲ 40%▲ 41%
The win. Traffic down 28% but orders up 2% and conversion up 40% — the most valuable single change on the site. This is where the heavy landing-page testing and routing effort paid off.
/products/30-servings-tincore OG product page322,5163.85%▼ 30%▼ 29%
The drag. Orders down 36% — the single biggest negative in the business. Most likely a routing effect (high-intent buyers now arrive via Rise 2 / the compare page first, so the PDP-as-entry now sees colder traffic) rather than a broken page — see "The PDP question" below.
/pages/compare-listicle-og NEWadvertorial "compare" page274,7302.08%— new— new
Did not exist last quarter. Now your #2 entry point by volume — a listicle/advertorial page absorbing a large share of paid traffic. Converts at 2.08%; a clear new pillar of the acquisition funnel worth optimizing deliberately.
Homepageaggregated257,5153.24%▲ 72%▲ 68%
Traffic down 45% but conversion up 72% — same pattern as Rise 2: less, but far higher-intent, traffic.
/collections/shopmain collection46,7604.66%▲ 1%▲ 5%
/pages/rise-2-coffeecoffee-switch LP44,4711.36%▼ 31%▼ 32%
/products/coffee-starter-kit30,0073.20%▼ 31%▼ 27%
/pages/nourish+ nourish starter kit25,2350.74%▲ 77%*low base
/pages/self-care-ritualbeing sunset14,5111.52%▼ 88% trafficwinding down

*Nourish CVR moves on a tiny base (<0.8% conversion) — directional only. Highest-converting entry points overall are /collections/shop (4.66%) and the OG tin PDP (3.85%); the high-volume paid LPs (Rise 2, compare page) convert lower by design because they catch the coldest traffic.

Where it came from
Traffic was cut across paid — the survivors convert far better

By acquisition channel, 90d vs. prior 90d. The headline efficiency gain is almost entirely Paid Social. Two channels (Direct, Organic Social) dropped in a way that looks more like a measurement/attribution shift than real demand — flagged below, verify before acting.

ChannelSessions 90dSess ΔOrders ΔCVRCVR ΔRPV Δ
Paid Social (Meta)959,228▼ 23%▲ 18%2.27%▲ 50%▲ 46%
Paid Search350,678▼ 29%▼ 17%3.74%▲ 19%▲ 18%
Direct263,081▼ 4%▼ 28%1.53%▼ 27%▼ 26%
"Other" (untagged)265,720▲ 7%▼ 12%2.61%▼ 20%▼ 24%
Paid Other118,356▼ 30%▼ 31%1.08%▼ 11%▼ 10%
Email101,926▲ 6%▲ 14%5.00%▲ 11%▲ 18%
Organic Search82,167▼ 3%▼ 1%3.89%▲ 5%▲ 5%
SMS28,943▼ 7%▼ 7%6.05%▲ 3%▲ 5%
Organic Social20,528▼ 64%▼ 91%1.25%▼ 74%▼ 73%
The signal: Paid Social is the engine of the whole quarter. Meta traffic was cut 23%, yet it produced 18% more orders at a 50% higher conversion rate and 46% higher revenue-per-visitor. Email ($2.86 RPV) and SMS ($3.23 RPV, the highest on the site) remain your most efficient channels and are stable-to-up.

The flag: Direct conversion fell 27% on basically flat traffic, and Organic Social orders collapsed 91%. Conversion rates don't usually move that hard from real behavior alone — these patterns are consistent with an attribution/UTM change (sessions getting re-bucketed, e.g. into "Other," which grew). Treat both as a data-integrity question to confirm before reading them as lost demand.
The funnel
Mobile-first, and the leak is at the top — not at checkout

All 2.19M sessions, last 90 days. 85% arrive on mobile. The biggest loss happens immediately: two of every three sessions bounce on the landing page. Once a visitor actually starts checkout, 61% finish — checkout itself is healthy. The opportunity is upstream: landing-page engagement and the browse-to-cart step.

Sessions
2,192,051
Bounced on landing
1,448,426 · 66%
Added to cart
82,688 · 3.8%
Started checkout
77,849
Converted
47,263 · 2.16%

Read: 66% landing bounce and the browse→add-to-cart step are where the money leaks — not checkout (which closes 61% of starts). On a site that's 85% mobile, that points the next round of CRO at mobile landing-page engagement (above-the-fold proof, message match, fewer attention leaks) rather than cart/checkout mechanics.

The PDP question
Did the product page break, or did the traffic just move?

The OG tin PDP (/products/30-servings-tin) lost 30% of its conversion rate and 36% of its orders — the single biggest negative in the data. Worth understanding precisely, because the answer changes what you do about it.

Hypothesis A — traffic routing (most likely)

Paid traffic that used to land directly on the high-intent PDP now gets routed first through Rise 2 and the new compare-listicle advertorial. So the PDP-as-a-landing-page now receives colder, top-of-funnel traffic and converts lower — while the actual orders show up credited to Rise 2 (whose conversion soared) and the compare page. This fits the data: Rise 2 and Homepage conversion are up 40–72% over the same period.

Hypothesis B — the page regressed

The PDP saw heavy redesign testing this quarter (the "Omni PDP," product-name, and buy-box experiments). If a losing variant shipped, the live page itself could have degraded. The evidence argues against this: those PDP tests lost — Product Name v3 (−16.5%) and Product Selector Name v2 (−22%) were both beaten decisively by control — so control should still be live. But it's worth a 10-minute confirmation that no loser was shipped.

Resolve it cleanly: two "Simplified PDP" tests are running right now (started Jun 1 & 4) and trending mildly positive (+3–6%, not yet significant). Let them finish, confirm the current PDP is the control build, and segment the PDP's conversion by traffic source to confirm the routing story. Don't "fix" the PDP on the assumption it broke — the numbers suggest it didn't.

The A/B testing program
25 tests ran. One won. That's the real problem to fix.

High velocity, low rigor. The standard for a winner is ≥95% probability-to-beat-control on revenue-per-visitor (90% is "keep running," not "ship"). Against that bar, almost nothing this quarter qualifies — mostly because tests were stopped after 1–4 days, before they could read. The good news: the gains we did see came from elsewhere (media + routing), so the store improved anyway. The bad news: you're not yet learning from the tests.

CLOSESTCoffee & OG Product Name — "New Name": +7.7% conversion rate at ~90% confidence (revenue/visitor +4.5%, not significant). The only result that cleared any bar — treat as a weak-but-real win worth re-confirming, not a done deal. May 4–14
CLEAR LOSERProduct Selector Name v2: −22% rev/visitor, 10% p2bc. Control crushed it. Do not ship. May 14–18
CLEAR LOSERProduct Name v3 (Omni PDP): −16.5% rev/visitor, 11% p2bc. Do not ship. May 26–Jun 1
CLEAR LOSERRise 2 sticky CTA on mobile: −11.3% rev/visitor, 18% p2bc. A repeat of a prior loss — stop re-testing sticky mobile CTAs. May 18–21
RUNNINGSimplified Omni PDP & Simplified OG-only PDP: trending +3–6%, not yet significant. The most promising open thread — let them run to power. started Jun 1 & 4
RUNNINGRise 2 CTA copy: +3.6% rev/visitor, early. started Jun 1
INCONCLUSIVE~17 of 25 tests — landing-page redirect swaps (Super LP, Sophia, Spicy, Switchers, Fresh), hero updates, side-cart redesign, flavor-bundle ABCs. Most ran too few days or on too-low a conversion base to read. Several leaned negative (Hero Update −7.5%, Founder-ads-vs-Fresh −13.4%, OG-vs-Super-LP −5.8%).

Read: the pattern across the quarter is consistent — generic hero/CTA swaps and product-name changes lose or fail to read; the genuinely different angles (advertorial routing, PDP simplification) are the only things trending positive. The program's velocity is an asset; its statistics are not. Fix the discipline and this engine starts compounding.

So what — where to point the next 90 days
Five moves, in priority order
DO NOW Protect what's actually working

The CVR/RPV gain is a media-efficiency + landing-page-routing story (Meta CVR +50%, Rise 2 +40%, the new compare page). Make sure whoever drove the paid-media tightening and the Rise 2 / compare-page routing knows it worked and keeps the formula. This is the win — don't let a "traffic is down" panic reverse it.

DO NOW Verify the Direct & Organic-Social drops are real

Direct CVR −27% and Organic Social orders −91% look like attribution/UTM re-bucketing (note "Other" traffic grew). Have analytics confirm tracking didn't change before treating these as lost demand. If it's measurement, the real numbers are better than they look.

FIX Give the testing program statistical teeth

Stop killing tests in 1–4 days. Set one rule: a test runs until it reaches ~200 orders/arm and ≥95% probability-to-beat-control on revenue-per-visitor, or it's called inconclusive — never "winner." Don't ship the three clear losers above. This single change turns a noisy engine into a learning one.

RESOLVE Close out the PDP question

Confirm the live OG-tin PDP is the control build (no loser shipped), let the two "Simplified PDP" tests finish, and segment PDP conversion by source to prove the routing explanation. Then decide deliberately — don't redesign a page that the data says didn't break.

NEXT Aim the next CRO round at mobile landing engagement

The leak is a 66% mobile landing bounce, not checkout. The highest-leverage on-site work is above-the-fold proof and message-match on Rise 2 and the compare page (see the companion Rise 2 audit), plus deliberately optimizing the brand-new compare-listicle page that's now your #2 entry point and has never been tuned.

Self-review
Sources queried (Jun 8 2026): Intelligems site-wide snapshot · landing-page breakdown · traffic-channel breakdown · conversion funnel · all 25 in-window A/B tests analyzed for winner / lift / significance
Comparison basis: every Δ is last 90 days (Mar 8–Jun 8) vs. the prior 90 days (Dec 8–Mar 8), Intelligems' own attribution.
Held honest: weekly time-series endpoint returned empty for this store → trends shown period-over-period, not week-by-week (no fabricated trend line). Direct & Organic-Social moves flagged as possible tracking shifts, not asserted as demand. Experiment hypotheses aren't logged in Intelligems (descriptions blank) → test intent inferred from names + targeting, not invented.
Statistical bar stated, not blurred: winners require ≥95% p2bc on RPV; the one ~90% result is labeled borderline, not a win. No fabricated numbers. Dates Pacific; figures tabular.