Product Autopsy9 Stages~20 min read

Amazon Shopping

Follow one buyer from "just checking the price" to a life wired through Prime — and see the commerce machine running behind every click

Stage 1 of 9

Acquisition

Where do they come from — and at what cost?

Tomas did not go to Amazon. Amazon was already there. He Googled "best noise canceling headphones 2026" and the first three results were Amazon product listings — two organic, one a Google Shopping ad paid for by Amazon. Below those, five more results were review articles with Amazon affiliate links.

This is the quiet dominance of Amazon's acquisition machine. Half of all product searches in the US start directly on Amazon, not Google. For the other half that start on Google, Amazon is usually the first click anyway. The company spends $20B+ per year on marketing, but its real acquisition channel is something money cannot easily buy: default behavior. "Just check Amazon first" is a reflex for 200M+ Americans.

Amazon's cheapest acquisition channel is brand gravity. Over 55% of traffic comes directly — people who type amazon.com or open the app without any ad prompting them. The most expensive channel is Google Shopping, where Amazon competes with retailers for clicks at $1–3 per click. But the affiliate network is the quiet genius: millions of bloggers and review sites earn 1–10% commissions for sending traffic, meaning Amazon pays for acquisition only after a sale happens.

Every product page is a Google landing page. The engineering team manages structured data, canonical URLs, and page speed across 350M+ product pages. A 100ms slowdown in page load costs an estimated $1.6B in annual revenue. This is SEO treated as infrastructure, not marketing.

New users without personalization data see a deals-forward layout with category chips — an impression of abundance without overwhelming. The app home screen determines 30% of first-session behavior. For returning users, the recommendation engine takes over, driving roughly 35% of all purchases.

A customer acquired via Google Shopping converts to Prime at 35% within 90 days, versus 22% for social traffic. The debate is not just about cost per click — it is about downstream quality. Cheap acquisition of low-retention users destroys unit economics. The affiliate channel sidesteps this by paying only on conversion.

310M+Active Customers

55%Direct Traffic Share

$20B+Marketing Spend/yr

War Room

4 perspectives

"Should we bid more aggressively on Google Shopping for high-intent queries?" Data shows a customer acquired via Google Shopping converts to Prime at 35% within 90 days vs. 22% for social traffic. But CPC is $2.50+ and rising. The debate: pay more for higher-quality traffic, or lean into the affiliate network which is effectively free until conversion?

ENG

SEO at Amazon's scale is infrastructure. Every product page is a Google landing page. The team manages structured data, canonical URLs, and page speed across 350M+ product pages. A 100ms slowdown in page load costs an estimated $1.6B in annual revenue.

DATA

Attribution across the affiliate network involves millions of affiliate partners with different cookie windows, attribution models, and fraud vectors. The data team tracks click-to-purchase paths across 24-hour attribution windows while detecting click-stuffing and cookie manipulation at scale.

DESIGN

The app home screen determines 30% of first-session behavior. For new users, personalization data is thin. The team tests "trending near you" vs. "top deals" vs. "popular categories" for cold-start users. Current winner: deals-forward layout with category chips — it gives the impression of abundance without overwhelming.

Tomas found the Sony headphones. He taps "Buy Now" and hits a Prime trial offer at checkout. ↓

Stage 1 of 9 · Acquisition

Stage 2 of 9

Activation

Did the product actually deliver for them?

Tomas taps "Buy Now" at 10:22 AM Saturday. The checkout screen shows two options: free shipping in 5–7 days, or free next-day delivery with a Prime free trial. The headphones cost $278 either way. The only difference is when they arrive. He starts the trial.

Sunday morning, 10:48 AM. His phone buzzes: "Delivered — your package was left at the front door." He opens the door. Twenty-four hours from "my headphones broke" to "wearing new headphones." He did not drive anywhere, talk to anyone, or wait in a checkout line. That is activation. Not the account creation. Not adding the item to the cart. The moment Amazon's promise — "it's already here?" — becomes real.

That delivery was not magic. Amazon's demand forecasting model predicted the Sony WH-1000XM5 would sell 847 units in the Atlanta metro that week. 200 units were already at the ATL2 fulfillment center, 15 miles from Tomas's apartment. The headphones were practically at his door before he ordered them.

The Prime trial was not offered at signup or on the homepage — it appeared at the exact moment Tomas was deciding between 5–7 day shipping and tomorrow. The trial was not about Prime's value proposition. It was about resolving this specific purchase faster. 73% of Prime trial starts happen at checkout. The product places the offer at maximum intent.

The delivery photo confirmation replaced an anonymous "delivered" notification with an image of his package at his door. This reduces "where's my package?" support contacts by 30% and creates a small moment of delight — proof that the system worked.

Users who make 3+ purchases during the trial convert to paid at 93%. Amazon has 30 days to make him feel that canceling would be a loss. The playbook: surface "You saved $X with Prime" after every purchase, send Prime Video recommendations within 48 hours, trigger a "Your Prime benefits" email on Day 7.

73%Trial Starts at Checkout

93%Convert if 3+ Orders

<24hrFirst Delivery Promise

War Room

4 perspectives

ENG

Demand forecasting models run on 300+ features — seasonality, weather, local events, competitor pricing, search trends, social media signals. The model predicts SKU-level demand per fulfillment center per day. Getting this wrong by 10% means either stockouts or excess inventory. The system rebalances inventory across 110+ US fulfillment centers nightly.

"Prime trial conversion is our single most important metric." Every decision in checkout is optimized around trial start rate. The PM tracks trial-to-paid conversion at 30/60/90 days. The current debate: should the trial be 14 days instead of 30? Shorter trials force faster habit formation but convert at a lower rate. Data says 30 days wins — barely.

DATA

First-purchase reliability is an SLA. On-time delivery rate for first Prime orders is tracked separately from overall. If it drops below 97%, an automated alert fires to the logistics PM. A late first delivery reduces trial-to-paid conversion by 8 percentage points.

OPS

Amazon now delivers 60%+ of its own packages vs. UPS/USPS. The DSP (Delivery Service Partner) program manages 3,000+ independent contractors running delivery vans. Quality control across a fragmented fleet is the ops challenge — one bad driver experience can tank a first impression permanently.

Tomas's headphones arrived in 24 hours. He's impressed. His Prime trial is running. Now the machine starts building a habit. ↓

Stage 2 of 9 · Activation

Stage 3 of 9

Engagement

Is the product earning repeated attention?

Tuesday, 8:17 PM. Tomas is browsing Amazon — not because he needs anything specific, but because the app sent him a notification 20 minutes ago. It knew he bought headphones two days ago. It recommended the most-purchased accessory for that exact model. And it added urgency: a Lightning Deal timer. Tomas adds the case to his cart. Then he keeps browsing.

The recommendation engine generates 35% of all Amazon revenue — not by showing people what they searched for, but by showing them what they did not know they wanted. The algorithm runs on collaborative filtering at a scale no competitor can match: billions of purchase histories revealing hidden correlations. People who buy noise-canceling headphones also buy travel neck pillows. It sounds random. It converts at 12%.

By week three, Tomas has placed six orders. Coffee pods on Subscribe and Save (auto-recurring every month). A phone charger he saw in a Lightning Deal. A birthday gift he found through "Customers also bought." He opens the Amazon app 4–5 times a week — sometimes to buy, often just to browse. The app has become his default product discovery engine, replacing Google, retail store visits, and even thinking about where to buy things.

Lightning Deals drive 3x the conversion of standard discounts. The urgency mechanic — countdown timer plus claimed percentage — creates FOMO that drives impulse purchases without cannibalizing regular-price sales. The timer is not decorative. It is a behavioral trigger.

Subscribe and Save is the ultimate engagement flywheel. Users with 3+ active subscriptions order 4.2x more frequently than non-subscribers and have a churn rate of under 3%. The system builds invisible habit by removing the decision entirely — the product arrives whether or not the user thought about it.

The homepage is the highest-stakes real estate in e-commerce. Every pixel is a revenue decision. The team A/B tests widget order, recommendation placement, and deal formatting weekly. Current debate: should personalized recommendations appear above or below deals? Data says above for high-frequency users, below for occasional browsers.

35%Revenue from Recommendations

4.2xOrders (3+ S&S Subscriptions)

<3%Churn Rate for S&S Users

War Room

4 perspectives

ENG

The recommendation engine is Amazon's most valuable algorithm. Item-to-item collaborative filtering, running on a real-time graph of billions of purchase pairs. The system generates recommendations within 50ms per page load. It accounts for 35% of total revenue — more revenue-per-engineer than almost any system at any company.

"Lightning Deals drive 3x the conversion of standard discounts — but are we training users to wait for deals?" A/B test shows deal-exposed users buy more overall, not less at full price. The urgency mechanic drives impulse purchases without cannibalizing regular-price sales.

DATA

Subscribe and Save is the ultimate engagement flywheel. Users with 3+ active subscriptions order 4.2x more frequently than non-subscribers and have a churn rate of under 3%. The data team is building a model to predict which products a user will want to subscribe to based on reorder patterns.

DESIGN

"The homepage is the highest-stakes real estate in e-commerce." Every pixel is a revenue decision. The team A/B tests widget order, recommendation placement, and deal formatting weekly. Current debate: should personalized recommendations appear above or below deals? Data says above for high-frequency users, below for occasional browsers.

Tomas orders weekly now. But every transaction funds a business model far more complex than "sell things." ↓

Stage 3 of 9 · Engagement

Stage 4 of 9

Monetization

Is the business model real and sustainable?

Tomas buys a $40 Bluetooth speaker from a third-party seller. He thinks Amazon made $40 in revenue. The reality is far more interesting — and far more profitable than selling speakers. That speaker was sold by "SoundWave Electronics," a small brand using Amazon's marketplace. Amazon takes a 15% referral fee ($6), an FBA fulfillment fee ($5.40), a storage fee ($0.80), and $3.20 in ad spend the seller chose to run. Amazon collects ~$15.40 from a $40 sale — 38.5% — without touching the product.

But the speaker sale is just one revenue stream. Amazon's real money machine has four engines running simultaneously. Marketplace fees at 15% average referral, with 60% of Amazon sales from third-party sellers. Advertising at $50B+/year, growing 20%+ YoY. Prime membership at $139/year across 200M+ members. And FBA fees that turn Amazon's logistics network into a profit center.

Advertising is Amazon's fastest-growing segment and its profit engine. Sellers pay to appear at the top of search results. It is Google Ads for people who are already ready to buy. Average ROAS for Amazon ads is 4–5x, making it irresistible for sellers. Advertising is nearly pure margin — the inventory, logistics, and trust infrastructure already exist.

Prime membership is not just a shipping program — it is a commitment device. Members spend 2.3x more than non-members because they have already paid for the shipping. The annual fee turns an optional purchase decision into a sunk cost that biases every buying decision toward Amazon.

The ad auction system is a real-time bidding engine processing millions of auctions per second. Each search query triggers an auction where sellers bid for placement. The ranking combines bid amount, expected CTR, and relevance score. A $0.01 improvement in auction efficiency is worth hundreds of millions annually.

Average advertising cost as a percentage of revenue for sellers has risen from 4% to 8%+ in three years. If it gets too high, sellers leave for Shopify or direct-to-consumer. The data team monitors seller health metrics to find the maximum extraction point before defection — a real tension between platform profitability and ecosystem health.

$50B+Ad Revenue/yr

60%3P Seller Share of Sales

$139Prime Annual Fee

War Room

4 perspectives

"Ad load is at 25% of search results. Can we go to 30% without hurting conversion?" Every additional ad slot is pure profit. But if search results feel like an ad page, users lose trust and click less. The PM runs ad-load experiments weekly, watching both revenue and long-term search engagement. Current sweet spot: 25–28% depending on category.

ENG

DATA

Measuring the "advertising tax" on sellers. Sellers increasingly feel they must advertise to be visible. Average advertising cost as a % of revenue has risen from 4% to 8%+ in three years. If it gets too high, sellers leave for Shopify or DTC. The data team monitors seller health metrics to find the maximum extraction point before defection.

Prime price elasticity modeling. At $139/year, renewals are 97%. What about $149? $159? Every $10 increase is $2B+ in annual revenue. But the model shows price sensitivity spikes sharply above $159 — cancellations would offset gains. The sweet spot is a slow, annual increase that stays below the pain threshold.

Tomas has been a Prime member for three months. He's forgotten what it's like to pay for shipping. That's the point. ↓

Stage 4 of 9 · Monetization

Stage 5 of 9

Retention

Do users genuinely need this — or just like it?

Four months in. Tomas's Prime free trial converted to paid without him noticing — which is by design. He now has three saved addresses (home, office, parents' house), two saved payment methods, and five active Subscribe and Save items. His 1-Click ordering is enabled. His purchase history has 23 items.

He thinks about canceling Prime when he sees the $139 charge on his credit card. Then he does the math: 18 orders in the last three months, average shipping cost without Prime ~$6, that's $108 in shipping avoided. Plus 15% off Subscribe and Save items. Plus Prime Video, which he watches three nights a week. Canceling would feel like losing $200+ in value he's already gotten used to having.

He doesn't cancel. 97% of Prime members don't. The retention mechanics are layered and reinforcing. Without Prime, Tomas is a comparison shopper — checks Walmart, Target, Best Buy for every purchase. With Prime, he does not compare prices. He 1-Click orders. He is psychologically pre-committed to Amazon for everything.

The deepest retention lever is not Prime itself — it is accumulated data. Tomas's Amazon account knows his sizes, his taste in headphones, his coffee brand preference, his parents' address for gifts, and his reorder patterns. Switching to Walmart.com means losing all of that context and starting from zero. This is data gravity — the more you use Amazon, the harder it is to leave, not because of a contract, but because Amazon knows you better than any alternative.

If a weekly buyer has not purchased in 10 days, churn probability jumps 4x. The model triggers increasingly aggressive win-back sequences: personalized deal emails at Day 7, "Your Subscribe and Save items are waiting" reminders at Day 10, and a targeted $10 credit at Day 14. Recovery rate: 62% if caught by Day 10.

Subscribe and Save fulfillment reliability is an SLA. If a subscription delivery is late or the wrong item, the damage is disproportionate — the user questions the entire subscription model, not just one delivery. On-time rate for S&S orders is tracked at 99.2%, with automatic credits issued for any delay.

97%Prime Renewal Rate

2.3xPrime vs. Non-Prime Spend

62%Win-Back Rate by Day 10

War Room

4 perspectives

"The 'Your Savings' page is our best anti-churn tool." Showing users exactly how much they've saved makes the value tangible. The PM is testing surfacing this data in the app monthly vs. only when users visit the Prime page. Hypothesis: proactive savings reminders reduce cancellation intent by 15%.

ENG

1-Click ordering patent expired, but the implementation moat remains. Saved addresses, payment methods, delivery preferences, and purchase history create a checkout experience that takes 2 seconds. Rebuilding this state on a competitor takes weeks of purchases. The engineering team measures "checkout seconds" as a retention-correlated metric.

DATA

Churn prediction: the "dark period" detector. If a weekly buyer hasn't purchased in 10 days, churn probability jumps 4x. The model triggers increasingly aggressive win-back sequences: personalized deal emails (Day 7), "Your Subscribe and Save items are waiting" reminders (Day 10), and a targeted $10 credit (Day 14). Recovery rate: 62% if caught by Day 10.

OPS

Tomas is locked in. He orders everything on Amazon. Now his behavior starts creating value for Amazon without Amazon spending a dollar. ↓

Stage 5 of 9 · Retention

Stage 6 of 9

Referral

Does the product spread without paid marketing?

Friday evening. Tomas's coworker asks him where to find a good standing desk. Without thinking, Tomas opens Amazon, searches, and texts her a product link. "Found it on Amazon — this one has 12,000 reviews." Amazon does not have a traditional "give $20, get $20" referral program for retail. It does not need one. The referral mechanic is built into the product itself: the shareable product link.

Every product page has a share button. Every Amazon URL is a referral. When Tomas texts that link, his coworker lands directly on the product page with pricing, reviews, and a Buy Now button. There is zero friction between recommendation and purchase. Link-referred customers have 22% higher 90-day retention than paid-ad customers — at zero marginal cost.

Wish lists are a structural referral channel. Tomas shared his birthday wish list with his family. His parents, girlfriend, and sister all bought him gifts through Amazon — three people who might have shopped elsewhere, pulled into the Amazon ecosystem not by marketing, but by Tomas's product choices. A single birthday wish list generates an average of $400–600 in purchases from non-regular shoppers.

Wedding and baby registries are Amazon's most powerful unintentional acquisition channel. Amazon is now the #2 wedding registry platform in the US. A single registry exposes 100–200 guests to the Amazon buying experience. Each registry generates an average of $2,400 in purchases from people who may not be regular Amazon shoppers.

The Amazon Influencer Program formalizes what already happens organically: creators earn commissions on products they recommend via their own Amazon storefronts. Unlike Spark, which failed because it lacked creator monetization, the Influencer Program succeeds because it plugs into existing creator economics — they earn on top of the platforms they already use.

Every product URL is a deep link. The engineering team ensures that shared Amazon links render perfectly across iMessage, WhatsApp, Instagram DMs, and every messaging platform — with product image, star rating, and price visible in the preview card. That preview card is the ad. It costs nothing. It converts.

#2Wedding Registry Platform (US)

$2,400Avg Registry Revenue

22%Better 90-Day Retention (link-referred)

War Room

4 perspectives

"Registry is a customer acquisition channel disguised as a feature." Each wedding registry brings 80–120 unique purchasers into the ecosystem. The PM tracks "registry-sourced new customers" as an acquisition metric. Current focus: making registry creation easier and pushing Amazon registry partnerships with wedding planning platforms.

ENG

Universal link architecture. Every product link needs to work across app, mobile web, and desktop — with deep linking, attribution, and personalized pricing intact. The team manages 500M+ unique product URLs that need to resolve correctly in every context. A broken link is a lost sale and a broken referral chain.

DATA

Measuring organic virality. How many new customers come through shared links vs. paid channels? The data team tracks "link-sourced first purchases" and found that link-referred customers have 22% higher 90-day retention than paid-ad customers — similar to traditional referral programs but at zero marginal cost.

"Gifting occasions are our single best non-search acquisition surface." Holiday, birthday, and wedding traffic brings first-time buyers whose LTV often exceeds organic search acquisitions. The PM is testing gift-wrapping and gift messaging features as conversion levers for these occasions.

Tomas started buying headphones. Now Amazon wants a piece of his grocery budget, his pharmacy, his entertainment. ↓

Stage 6 of 9 · Referral

Stage 7 of 9

Revenue Expansion

Can the business grow without just adding users?

Month six. Tomas opens Amazon to buy paper towels and notices an "Amazon Fresh" section. Free 2-hour grocery delivery for Prime members on orders over $35. He adds milk, eggs, chicken, and vegetables. Total: $62. He just gave Amazon a piece of his $400/month grocery budget for the first time.

Two weeks later, his doctor prescribes allergy medication. He checks the price at CVS: $45/month. Amazon Pharmacy: $31/month, delivered to his door. He transfers his prescription. Then his Prime Video autoplay serves him a new series after he finishes The Boys. He's watching three hours a week now.

Tomas's Amazon spending: $120/month (electronics) → $380/month (electronics + groceries + pharmacy + Subscribe and Save). Same person. Same Prime membership. Revenue tripled. This is expansion done right: not new users, but existing users unlocking new value on the platform.

Buy with Prime is the most strategically significant expansion. It lets Shopify and other external stores offer Prime shipping and the Amazon checkout experience on their own websites. For merchants, it means higher conversion rates. For Amazon, it means taking a cut of commerce that happens outside Amazon.com. It is the play to make Amazon the infrastructure layer of all online commerce, not just the storefront.

Cross-category spend correlation is significant: users who start buying groceries on Amazon increase total platform spend by 35% — and the spend is additive, not substitutive. Pharmacy users have the highest retention rate of any category: 96% at 12 months. Each new category a user adopts deepens the relationship and raises the cost of switching.

Prime Video's ROI is measured in retention, not viewership. Prime members who watch Prime Video are 25% less likely to cancel. The $12B content budget is not competing with Netflix for subscribers — it is reducing Prime churn. Every hit show is measured by its impact on Prime renewal rates, not just audience size.

$574BNet Revenue (2024)

+35%Spend Lift When Grocery Added

96%12-Month Retention (Pharmacy Users)

War Room

4 perspectives

"Buy with Prime is our Trojan horse." If Prime checkout works on Shopify stores, Amazon captures purchase data from non-Amazon transactions. That data feeds the recommendation engine, ad targeting, and demand forecasting — even for products not sold on Amazon. The PM is tracking merchant adoption rate and conversion lift as the two key metrics.

ENG

Grocery logistics are fundamentally different from package delivery. Cold chain management, freshness windows, substitution logic, and batch picking at Whole Foods stores create engineering challenges that do not exist in warehouse fulfillment. The team is building a separate logistics stack optimized for perishables with 2-hour delivery SLAs.

DATA

Cross-category spend correlation: users who start buying groceries on Amazon increase total platform spend by 35% — and the spend is additive, not substitutive. Pharmacy users have the highest retention rate of any category: 96% at 12 months. The team is building "category expansion propensity" scores to target users most likely to adopt a new Amazon service.

Tomas buys everything on Amazon. But the company faces threats that could unravel even its deepest moats. ↓

Stage 7 of 9 · Revenue Expansion

Stage 8 of 9

Sustainability

Will this product still matter in 3 years?

Tomas does not know it, but the headphones he bought last month have a problem. The "Sony WH-1000XM5" listing he purchased from was a counterfeit — a third-party seller using a legitimate brand name with a knockoff product. He notices the sound quality is slightly off. He leaves a 2-star review. He wonders: can he even trust Amazon anymore?

This is Amazon's quality control crisis, and it is existential. When 60% of sales come from third-party sellers, Amazon does not control what is in the box. Counterfeits, misleading reviews, and fraudulent sellers erode the trust that makes "just buy it on Amazon" possible. An estimated 30% of Amazon reviews are fake or incentivized — a systemic problem that ML models combat in an arms race that never fully resolves.

The FTC antitrust lawsuit alleges Amazon uses its marketplace dominance to inflate prices, self-preference its own products, and force sellers to use (and pay for) FBA. If the case succeeds, it could force structural changes — separating the marketplace from Amazon's own retail, or restricting how search results are ranked. The outcome could reshape e-commerce for a decade.

The logistics cost paradox. Amazon spends $90B+ annually on logistics — more than FedEx and UPS combined. Same-day and next-day delivery is the product promise that keeps users loyal. But every faster delivery option costs more. Amazon is betting that automation (robotic fulfillment centers, drone delivery, autonomous vehicles) will eventually cut per-package costs by 40–60%. If that bet fails, the margin math breaks.

Seller fee increases are approaching a breaking point. The average Amazon seller's total fees (referral + FBA + advertising) now consume 45–50% of their revenue. If Shopify's fulfillment network reaches cost parity with FBA, the migration risk is real. The PM team models "fee ceiling" scenarios to identify the threshold where seller defection accelerates.

Project Zero uses ML to scan 10B+ listings for counterfeit signals. Brand Registry lets brand owners report fakes. The Transparency program prints a unique scannable code on every enrolled unit. But the whack-a-mole problem persists: shut down one fraudulent seller, another appears. The PM is pushing for proactive authentication — verify products before they ship, not after complaints.

$90B+Annual Logistics Spend

45–50%Total Fees as % of Seller Revenue

1B+Units Enrolled in Transparency/yr

War Room

4 perspectives

"Counterfeit detection is a trust-or-die problem." Project Zero uses ML to scan 10B+ listings for counterfeit signals. Brand Registry lets brand owners report fakes. But the whack-a-mole problem persists — shut down one fraudulent seller, another appears. The PM is pushing for proactive authentication: verify products before they ship, not after complaints.

ENG

Transparency program: every unit gets a unique code. Brands enroll products and Amazon prints a unique scannable code on every unit. Buyers scan to verify authenticity. The engineering challenge: generating, printing, and verifying billions of unique codes at fulfillment center speed without slowing throughput. Current coverage: 1B+ units per year.

"Seller fee increases are approaching a breaking point." The average Amazon seller's total fees now consume 45–50% of their revenue. If Shopify's fulfillment network reaches cost parity with FBA, the migration risk is real. PM is modeling "fee ceiling" scenarios to identify the threshold where seller defection accelerates.

DATA

Review manipulation detection. An estimated 30% of Amazon reviews are fake or incentivized. The ML model analyzes review velocity, reviewer history, linguistic patterns, and purchase verification to flag suspicious reviews. But the adversaries adapt as fast as the models improve — it is a perpetual arms race.

Tomas uses Amazon for shopping. But Amazon has quietly become the infrastructure layer underneath half the internet. ↓

Stage 8 of 9 · Sustainability

Stage 9 of 9

Ecosystem

Has the product become bigger than itself?

It has been a year. Tomas wakes up and says "Alexa, what's the weather?" His Ring doorbell shows him a delivery arriving. He watches The Boys on Prime Video while his Amazon Fresh groceries are being delivered. His Subscribe and Save coffee pods arrive every four weeks without him thinking about it. His prescriptions auto-refill through Amazon Pharmacy. His company runs on AWS.

Tomas is not an Amazon customer anymore. He lives inside Amazon's ecosystem. Shopping is just the front door. Behind it is an infrastructure empire that touches every layer of his life — and the internet itself. Users who interact with 4+ Amazon services churn at 1/5 the rate of shopping-only users.

AWS is the most important part of the ecosystem most customers never see. It generates $100B+ in annual revenue and funds the losses Amazon takes on free shipping, Prime Video content, and below-cost pricing in new categories. Without AWS's profits, the consumer business model would be unsustainable. AWS also runs Netflix, Airbnb, and millions of other companies — meaning Amazon profits from its competitors' growth.

Alexa and Ring create physical touchpoints in the home. "Alexa, reorder paper towels" is the zero-friction commerce endpoint. Ring cameras create a delivery ecosystem (package detection, delivery verification). These devices are not profitable on their own — they are distribution channels for Amazon's commerce and services.

Amazon stopped being a store a long time ago. It became the infrastructure layer of commerce — the pipes through which products, payments, and data flow, whether you are shopping on Amazon.com or a Shopify store using Buy with Prime and fulfillment powered by AWS. The long-term vision: even if Amazon.com market share shrinks, Amazon takes a fee on every e-commerce transaction. This is the Microsoft Office playbook — be the platform, not just the product.

The cross-ecosystem engagement score captures this reality. Users who interact with 4+ Amazon services (Shopping, Prime Video, Alexa, Grocery) churn at 1/5 the rate of shopping-only users. The data team builds a "platform depth" index and routes low-depth users to cross-product promotions. Target: increase average services-per-user from 2.1 to 3.0 within 18 months.

4+Services = 5x Lower Churn

$100B+AWS Annual Revenue

2.1→3.0Services-per-User Target

War Room

4 perspectives

ENG

Alexa's commerce integration is the voice-first checkout. "Alexa, reorder coffee" triggers a purchase flow with zero screens. The engineering challenge: voice recognition accuracy for product disambiguation, confirmation UX without visual feedback, and fraud prevention when anyone in the house can order. Voice-commerce adoption is low but growing 30% YoY.

"AWS funds the flywheel — but how long can we cross-subsidize?" AWS's operating margin is ~30%, while retail is ~3–5%. The PM debate: should retail ever need to be independently profitable, or is the cross-subsidy the strategy? If antitrust forces a separation, retail needs to stand on its own economics. The team models both scenarios.

DATA

Cross-ecosystem engagement score: users who interact with 4+ Amazon services churn at 1/5 the rate of shopping-only users. The data team builds a "platform depth" index and routes low-depth users to cross-product promotions. Target: increase average services-per-user from 2.1 to 3.0 within 18 months.

Amazon as commerce infrastructure. Buy with Prime, FBA, AWS, and Amazon Pay form a stack that lets any business use Amazon's infrastructure without selling on Amazon.com. The long-term vision: even if Amazon.com market share shrinks, Amazon takes a fee on every e-commerce transaction. Be the platform, not just the product.

Tomas started with broken headphones. A year later, he lives inside an ecosystem he would need months to unwind. ↓

The Full Picture

Tomas started as a guy with broken headphones. Nine stages later, he is a Prime member who defaults to Amazon for everything — electronics, groceries, prescriptions, entertainment — and lives inside an ecosystem he would need weeks to unwind from. That transformation was not luck. It was a commerce machine: PMs optimizing checkout conversion, engineers pre-positioning inventory across 110 fulfillment centers, data scientists predicting what Tomas wants before he knows he wants it, and a business model where every transaction funds the next layer of infrastructure. Understanding these nine stages is not academic. It is how you think about any platform that turns a single purchase into a lifelong commercial relationship.

Back to all autopsies