CRO & Experimentation Sleep & Wellness D2C Independent Research Puffy.com · 2026

Selling sleep,
not just a mattress.

An independent conversion-rate audit of Puffy.com, diagnosing why a feature-led landing page misses two of its highest-intent buyer moments, and specifying two evidence-backed A/B tests, each modeled against a real revenue bar, to close the gap.

2tests
Independently specced experiments, hypothesis through revenue model
21–31%
Projected relative CVR uplift range, decomposed by mechanism
$4.2M+
Conservative annual incremental revenue projected, per test
$3.375Mbar
Minimum revenue threshold each hypothesis had to clear
01 — Background

Why I picked apart a mattress company's landing page.

Puffy is a US direct-to-consumer sleep brand selling mattresses, bedding, and bedroom furniture online. This is not a client engagement or a role I held, it's an independent research case study I built to demonstrate how I approach conversion-rate optimization as a product manager: starting from a live site, not a hypothetical brief.

I chose a high-consideration, low-frequency e-commerce category on purpose. Mattresses are a considered purchase - buyers don't shop them often, price points are high relative to average online baskets, and the sales narrative usually leans on product specifications (foam density, certifications, trial periods) rather than the buyer's actual motivation for being on the page that day. That gap between what the page says and why the visitor is really there is exactly the kind of problem CRO work is supposed to find.

My method for this audit followed four steps: (1) walk the live landing page and PDP experience the way a first-time visitor would, noting what the page leads with and what it leaves out; (2) build two independent evidence bases - sleep-health research and marriage/new-household market data - to ground any hypothesis in more than intuition; (3) write each hypothesis in a testable form, with an expected outcome and a revenue threshold, exactly as I would for a real experimentation roadmap; and (4) model the revenue impact transparently, stating every assumption so the logic could be swapped for real analytics the moment they were available.

Scope of this case study: two fully-specified A/B tests - a health-stakes hero section and a new-home bundle offer - each carried from hypothesis through rationale, control/variant spec, uplift decomposition, and a revenue model built to clear a stated $3.375M annual incremental revenue bar.


02 — Diagnosis

A page selling a product, not solving a problem.

Puffy's landing experience is built around comfort and credibility: a hero image of the mattress, a "Shop Now" call to action, and feature-forward copy - CertiPUR-US certification, foam composition, the length of the sleep trial. It's a clean, trustworthy presentation. It's also entirely product-first: it assumes the visitor already knows they want a mattress and just needs reassurance about quality.

That assumption breaks down for a large share of traffic. Two buyer realities go completely unaddressed on the page as it stands today:

01
The messaging-to-motivation gap
Most visitors aren't shopping for foam density - they're shopping because something in their life isn't working: back pain, exhaustion, a sleep-satisfaction score they'd never admit out loud. The page never names the problem it solves, so it never earns the urgency a considered purchase like this needs to convert cold traffic.
02
The basket-composition gap
The entire funnel is merchandised around a single mattress purchase. There's no path for the visitor who isn't just replacing a mattress but furnishing an entire new home - a newly married couple, a first-time renter - who would rather solve five furnishing problems in one purchase than shop them one at a time.

Both gaps point to the same underlying diagnosis: the page optimizes for people who have already decided what they want, and does very little for the much larger set of visitors who are still deciding why they're here. The scale of that second group is not small - it's a documented, well-researched public-health and life-stage reality:

70M
Americans struggle with sleep issues
ScienceDirect
50%
of adults grade their own sleep an "F"
NSF Sleep in America, 2025
60%
of adults don't get enough sleep
NSF 2025 Poll
2.1M
new US marriages annually - a recurring furnishing audience
US marriage statistics

Neither audience is a niche. They're just invisible to a landing page that only speaks in product features.


03 — My Approach

A repeatable framework: hypothesis, evidence, spec, model.

Rather than pitch "improve the messaging" as a vague direction, I held both tests to the same discipline I'd expect on a real experimentation roadmap.

01
Anchor every hypothesis to a named segment and a specific driver
Not "improve conversion" but "cold paid-search visitors with an unaddressed pain point" and "newly married couples in a furnishing mindset." A hypothesis that doesn't name who it's for and what behavior it changes isn't testable.
02
Build the evidence base before writing a headline
Every claim in both tests - the health risks of poor sleep, the size of the newlywed/renter segment, the savings customers get from bundling - is sourced from public research, not assumed. The creative comes after the evidence, not before it.
03
Decompose the uplift into named mechanisms, not one guessed number
"25% uplift" means nothing on its own. I broke each projection into five contributing mechanisms - problem recognition, message-match, decision friction, and so on - each with its own defensible range, then combined them non-additively into a realistic band.
04
Model revenue with a stated conservative and upside scenario
Both tests are held to the same $3.375M annual incremental revenue bar, using the same traffic and baseline conversion assumptions, so they can be compared on equal footing rather than each using whatever numbers make the pitch look best.
05
Sequence by reach and effort, not by whichever idea is most exciting
A bigger projected number doesn't automatically ship first. I scored both tests on audience reach, implementation effort, and validation risk before deciding which one earns the first test slot.

Two hypotheses, two different levers

Test 1 · Jump to section 04

Health-Stakes Hero Section

Reframes the landing page from "here's our mattress" to "here's what poor sleep is doing to you, and here's the fix" - targeting the broadest share of visitors, including cold paid traffic.

Test 2 · Jump to section 05

New-Home Bundle Offer

Adds a curated bundling path for newlyweds and first-time renters - shifting the mission from "I need a mattress" to "I need to furnish my new home," raising both CVR and basket size.


04 — Test 1

Give visitors their health stakes before their comfort options.

Hypothesis
Adding a Health-Issue Hero Section above the current hero - naming the health risks linked to poor sleep and connecting them to a targeted mattress solution - will make the page feel personally relevant before it asks visitors to evaluate product specs.
Expected outcome
25% relative CVR uplift Min. $3.375M annual revenue Longer landing-page engagement Higher primary-CTA click-through

Why this test: the evidence base

The current page talks about comfort. It never talks about consequences. Sleep research gives a very concrete, very large pool of visitors a reason to care immediately:

10%
of Americans wake up with pain daily
Ortho Bethesda
48%
less low-back pain on a medium-firm mattress
ScienceDirect
70M
Americans (21% of the population) struggle with sleep
ScienceDirect
50%
of adults grade their own sleep satisfaction an "F"
NSF 2025

The health risks poor sleep creates

These aren't soft correlations - they're the kind of numbers that make a health-first message feel earned rather than manipulative:

Health conditionRisk increaseSource
Type 2 Diabetes3× higherJohns Hopkins Medicine
Heart disease48% increasedJohns Hopkins Medicine
Obesity (if <5 hrs/night)50% higherJohns Hopkins Medicine
Dementia33% increasedJohns Hopkins Medicine
Colorectal cancer36% increasedJohns Hopkins Medicine
Major depression (insomniacs)4× higherHarvard Health
Brain aging3–5 years fasterJohns Hopkins Medicine

What a better mattress actually fixes

The message only works if the product genuinely delivers - so the hero can't just raise alarm, it has to close the loop with a proven mechanism:

BenefitImprovementStudy
Extra sleep per night+42 minutesComfort study, CW Williams
Sleep quality improvement+55%Systematic review, ScienceDirect
Back pain reduction−48%Low back pain patients, ScienceDirect
Sleep quality after bedding replacementImprovedBedding study, Sleepycat

A secondary, not primary, motivator: poor sleep is also linked to lower sexual wellbeing in survey data (CBS News-cited research), and a subset of mattress features - bounce, edge support, motion isolation - address this directly. I chose not to lead with it: it broadens relevance for a portion of visitors without being necessary as the hook, and leading with health and pain relief keeps the primary message defensible and universal.

Control vs. variant

Control
  • Standard hero: product image + "Shop Now" + comfort messaging
  • Feature-focused copy: CertiPUR-US, foam type, trial length
  • No connection between the product and a personal health stake
  • Single CTA, framed around the product, not the problem
Variant
  • Health-issue flag banner at the top of the page naming the risk
  • New hero section replacing the current one, framed around the health stake
  • Health solutions section below the hero connecting symptom → mattress fix
  • Quick CTA section giving a single, low-friction path to purchase

Uplift decomposition

The 25% headline number is a combination of five separate effects, treated as multiplicative rather than purely additive - landing on a realistic 21–31% relative uplift:

Immediate problem recognition
+6–8%
Visitors see their pain reflected immediately, so the page feels personally relevant from the first scroll.
Paid-traffic message match
+5–7%
The page aligns better with ad intent and reduces bounce from mismatch - relevant for the 80% of visits that arrive via paid channels.
CTA engagement
+2–4%
A single, clear purchase action reduces hesitation and decision fatigue.
PDP continuation rate
+5–7%
More visitors move from landing page to product page because the mattress is framed as a solution, not a spec sheet.
Purchase confidence
+3–5%
Connecting the mattress to real health benefits strengthens the rationale to buy now rather than "think about it."

Revenue model

01
Baseline
Monthly mattress PDP visitors: 100,000. Baseline conversion rate: 2.0%100,000 × 2.0% = 2,000 orders/mo
02
Variant, after a 25% relative uplift
New conversion rate: 2.5%100,000 × 2.5% = 2,500 orders/mo — an incremental 500 orders per month.
03
Applying average order value
Conservative scenario at $700 AOV: 500 × $700 × 12 = $4.2M/yr. Stronger scenario at $800 AOV: 500 × $800 × 12 = $4.8M/yr.
Conservative annual revenue
$4.2M
Stronger AOV scenario
$4.8M
Case threshold
$3.375M — cleared

05 — Test 2

Shift the mission from "a mattress" to "a whole new home."

Hypothesis
If Puffy adds a dedicated new-home / new-couple bundling section - curated packages like Starter Bedroom, Complete Bedroom, and Full Home - more first-time home and rental shoppers will convert, because the site shifts from selling a single mattress to selling a complete, affordable move-in solution.
Expected outcome
25% relative CVR uplift Higher average order value Higher bundle attachment rate Min. $3.375M annual revenue

Why this test: the audience is large and recurring

~2.1M
new US marriages every year
Marriage statistics
~175K
new marriages per month - ongoing monthly demand
Marriage statistics
60–70%
of newlyweds rent before they buy a home
Housing research
100–125K
new couples renting and furnishing monthly
Derived estimate

That's a large, recurring audience actively in a "setup and furnish" mindset every single month - close to ideal conditions for a bundle offer.

Why bundling works for the customer

Customer benefitFindingSource
Cost savingsBundles save 15–30% vs. separate purchasesMarket research
Example bundle pricingFull house packages sold for $1,299 with added freebiesFurniture World Outlet
Mattress + frame bundlesBundles save money vs. separate purchasesHarmony Beds
Shipping efficiencyBundling reduces shipments and per-item delivery costSweetnight

For a new couple juggling move-in costs, deposits, and time pressure, a bundle gives them a simple answer: buy everything you need in one place, and save money doing it.

Why bundling works for Puffy

Company benefitImpactMarket evidence
Higher average order value+20–40% revenue per customerFurniture/bedding sales research
Market size opportunity$123.9B furniture/bedding market, growing to $147.2B by 2027Industry report
Cross-sell successBundles increase cross-category salesRetail pricing analysis
Inventory efficiencySlower-moving items paired with mattressesIndustry best practice
Lower marketing costOne bundle ad sells multiple products at onceRetail pricing analysis
Customer retentionBundle buyers more likely to return for other home needsFurniture packages research

Control vs. variant

Control
  • Landing page stays focused on mattress-led merchandising only
  • Mattress hero, mattress benefits, mattress CTA
  • Standard single-product exploration path
  • No path for a "furnish my whole home" shopping mission
Variant
  • New bundling section for newly married couples and first-time renters
  • Curated packages: Starter Bedroom, Complete Bedroom, Full Home
  • Shifts the shopper from "I need a mattress" to "I need a complete solution"
  • Matches the real shopping mission of a move-in or furnishing mindset

Uplift decomposition

Five mechanisms, non-additive, landing on a realistic 20–31% relative uplift:

Segment relevance
+6–8%
The offer speaks directly to new couples and new-home renters instead of a generic shopper.
Higher perceived value
+5–7%
Bundles feel more complete and more economical than assembling separate purchases.
Lower decision friction
+2–4%
Users choose from curated packages instead of assembling products individually.
Stronger AOV lift
+5–7%
Larger packages increase basket size and total spend per order.
Better paid-traffic alignment
+2–5%
Paid visitors see an offer that fits a move-in or furnishing mindset rather than a single-SKU ad.

Revenue model

01
Baseline
Monthly mattress PDP visitors: 100,000. Baseline conversion rate: 2.0%100,000 × 2.0% = 2,000 orders/mo
02
Variant, after a 25% relative uplift
New conversion rate: 2.5%100,000 × 2.5% = 2,500 orders/mo — an incremental 500 orders per month.
03
Applying average order value
Conservative scenario at $700 AOV: 500 × $700 × 12 = $4.2M/yr. Stronger scenario at $800 AOV: 500 × $800 × 12 = $4.8M/yr - and this test carries additional AOV upside from bundle attachment that a single-SKU test doesn't.
Conservative annual revenue
$4.2M
Stronger AOV scenario
$4.8M
Case threshold
$3.375M — cleared

06 — Prioritization

Same revenue bar, different sequencing logic.

Both tests clear the $3.375M threshold at identical assumed traffic and AOV, so the projected number alone can't break the tie. I sequenced them using reach, effort, and what each test validates for the other.

DimensionTest 1 · Health HeroTest 2 · Bundle Offer
Audience reachBroadest - applies to essentially all landing-page traffic, including 80% cold paid searchNarrower - newlyweds and first-time renters specifically
Implementation effortLower - new hero and messaging layer on an existing pageHigher - new merchandising logic, bundle pricing, and package inventory
What it validatesWhether problem-first messaging beats feature-first messaging at allWhether a narrower, higher-AOV audience responds to a bigger commitment
VerdictShip firstShip second

Sequencing logic: I'd run Test 1 first because it validates message-market fit across the broadest visitor segment and requires the lower implementation lift. Test 2 follows once the health-first positioning is proven - it targets a narrower but potentially higher-AOV audience, and a successful Test 1 also derisks the idea that problem-first (rather than product-first) framing works on this page at all.


07 — Revenue Model

How I stress-tested the numbers before trusting them.

I don't have access to Puffy's live analytics, so every dollar figure in this case study rests on a stated, auditable assumption set rather than real traffic data. That's a limitation I want to be explicit about rather than hide: the modeling method is sound, but the inputs are illustrative until validated against actual site data.

AssumptionValueWhy I chose it
Monthly mattress PDP visitors100,000A round, mid-market traffic figure for a national DTC mattress brand's core product page
Baseline conversion rate2.0%Consistent with typical PDP-to-purchase rates for considered e-commerce categories
Relative CVR uplift modeled25%The midpoint of each test's 20–31% decomposed range, used consistently across both tests
AOV scenarios$700 conservative / $800 strongerBrackets typical mattress-category order values without assuming best-case bundle attachment
TestConservative (annual)Stronger AOV (annual)
Test 1 · Health Hero$4.2M$4.8M
Test 2 · Bundle Offer$4.2M
+ untracked AOV upside from bundle attachment
$4.8M

What this model is - and isn't: it's a transparent, defensible way to reason about scale before committing engineering time to a test. It is not a guarantee. The first real deliverable of either test in production would be replacing every assumption in the table above with Puffy's actual analytics.


08 — Risks & Ethics

Where a health-fear headline could go wrong.

High
Overstating causation in health messaging
Poor sleep correlates with diabetes, heart disease, and depression risk - a mattress upgrade doesn't cure any of them. Copy must frame the mattress as improving sleep quality, which is linked to better outcomes, never as treating or preventing a disease. Mitigated by: legal/medical copy review before launch, and framing every claim as "linked to" rather than "causes."
High
Unverified traffic and conversion assumptions
The entire revenue model rests on assumed, not measured, traffic and baseline CVR. Mitigated by: treating every dollar figure in this case study as directional, and requiring real analytics access as the first step before any engineering resourcing decision.
Medium
Fear-based tone eroding brand trust
Leading with health risk can read as manipulative if the tone tips into alarmism, especially for a brand whose base personality is comfort and warmth. Mitigated by: A/B testing tone variants (clinical vs. empathetic framing) and tracking bounce rate and time-on-page as trust guardrail metrics, not just CVR.
Medium
Bundle offer cannibalizing higher-margin single-item sales
A cheaper bundled mattress-plus-frame sale could replace what would have been two separate, higher-margin purchases. Mitigated by: tracking margin-blended AOV, not just revenue-blended AOV, as a required guardrail metric alongside the primary conversion metric.
Low
Sample-size risk on the narrower bundle segment
Test 2 targets a smaller share of traffic than Test 1, which means longer runtime to reach statistical power. Mitigated by: sizing the test window against the segment's actual traffic share before launch, covered in Appendix C.
Low
Novelty effect inflating early results
A new hero section or bundle path may see an initial lift purely from being new, which fades over time. Mitigated by: running each test for a minimum of two full business cycles and monitoring for lift decay before calling a result.

09 — Reflection

What running this exercise reinforced about CRO as a PM.

01
A hypothesis is a testable link, not a headline idea
"Add a health section" is a creative idea. "Cold paid-search visitors will convert 6-8% better if they see their pain point recognized before the product" is a hypothesis. The difference is whether it names a segment, a mechanism, and a measurable outcome - and that difference is what makes a test worth building instead of just an opinion worth having.
02
Decomposing uplift disciplines optimism
It's easy to say "this could lift conversion 25%." It's much harder - and much more honest - to defend five separate mechanisms and show your work for each one. Naming the mechanisms is what turns a guess into a model, and it's the difference stakeholders can actually interrogate and trust.
03
A revenue model is only as credible as its stated assumptions
Without access to Puffy's real analytics, the honest move was to say so plainly rather than present the numbers as more certain than they are. Transparency about what you don't know is part of the deliverable, not a weakness to hide from a pitch.
04
Prioritization is a second decision, not a byproduct of the first
Both tests clear the same revenue bar. The choice of which one ships first came down to reach, effort, and what each test would tell me about the other - not which projection had the bigger number attached to it.
05
Emotional relevance only works if the product truth backs it up
Health-stakes messaging is only defensible because a better mattress genuinely reduces pain and improves sleep quality - the evidence in section 04 is the reason the framing isn't manipulative. Borrowed urgency without a real underlying mechanism is a liability, not a growth lever.
Appendix — The Evidence

Sources & Measurement Plan

Every statistic cited in this case study, plus the testing and measurement design I'd use to actually run these experiments.

Appendix A — Source References
TopicSource
Daily pain from sleep positionOrtho Bethesda
Mattress firmness & back pain reductionScienceDirect
Sleep issue prevalenceScienceDirect
Sleep satisfaction surveyNational Sleep Foundation, 2025
Diabetes, heart disease, obesity, dementia, brain-agingJohns Hopkins Medicine
Depression & insomnia riskHarvard Health Publishing
Extra sleep from a comfort upgradeCW Williams Community Health
Mattress effect on sleep qualitySleepycat
Sleep & sexual wellbeing correlationCBS News
Mattress features relevant to intimacyNaplab, Onsen
Bundle vs. bed-frame savingsHarmony Beds
Full-home package pricing exampleFurniture World Outlet
Shipping/bundling efficiencySweetnight
Furniture/bedding market sizeIndustry sales report, 2022–2027
Bundle cross-sell and retention effectsDataWeave, Neolivin
Appendix B — Testing & Measurement Plan

The statistical scaffolding I'd put around both tests before calling any result - illustrative, using the same assumed 100,000 monthly PDP visitors stated in Section 07.

ElementTest 1 · Health HeroTest 2 · Bundle Offer
Primary metric Landing page → purchase conversion rate Landing page → purchase conversion rate
Guardrail metrics Bounce rate, time on page, PDP click-through rate Margin-blended AOV, single-item cannibalization rate
Minimum detectable effect 15% relative lift (conservative vs. the 21–31% modeled range) 15% relative lift (conservative vs. the 20–31% modeled range)
Approx. sample size ~37,000 visitors per arm (95% confidence, 80% power) ~37,000 visitors per arm within the newlywed/renter segment
Estimated runtime ~3 weeks at 100K monthly PDP traffic, split evenly Longer - depends on what share of traffic is newlyweds/renters; needs segment-level traffic data before committing to a launch date
Minimum test duration 2 full business cycles, to rule out novelty effect 2 full business cycles, to rule out novelty effect