razacontent.
№01 Amazon Prime
Membership Growth · Content Strategy · Design Systems · 15 min read

How Prime earns loyalty.

I led Prime's value communication framework: the research synthesis, the system architecture, and the rubric ten Membership Growth teams now ship from. I owned the strategy, the rollout, and the AI-evaluation layer; I co-owned research with my UXR partner and brand voice with the brand design lead.

200M+
Members in framework scope
+77K
Annual lift, lifecycle copy attribution
10+
Teams adopted via workshops + working group
4
Operational artifacts shipped, in use today
Disciplines
Content design, UX writing,
content engineering
Initiative I led
Value Communication
strategy + framework
Tracks
System design and production,
in parallel
Surfaces shipped to
Acquisition, engagement,
retention, cancellation
The brief

Prime's relationship with its 200 million members had grown disconnected across every surface, segment, and team. I rebuilt how Prime conveys value, end-to-end.

The diagnosis

Renewals were strong, but the relationship was thin. Members transacted out of habit, ignored secondary benefits, and the same headlines repeated into background noise. In research conversations, they described Prime as practical and reliable, with no sense that the brand was talking to them as people. The org architecture was producing the inconsistency: three teams, three sets of guidelines, no referee.

Why it mattered now

Three external forces made this urgent. Competition was rising. Streaming services, DTC subscriptions, and retail loyalty programs were raising the bar on what a paid relationship should feel like. Prime memberships were saturating in core markets, which made retention the primary path to growth. And Prime was expanding at scale into new international markets, where the same content failures would multiply faster than they could be caught.

What I did

Synthesized four research studies into one diagnosis. Benchmarked Prime against Spotify, Duolingo, and other high-trust brands instead of the price-comparison cohort leadership wanted. Pushed Trust and Emotion onto the value model when the org was treating them as soft secondaries. Built the framework as principles, matrices, and a rubric that was scorable from day one, and partnered with PMs to ship live surfaces while the system was still being written. Ten teams adopted it through workshops, paired launches, and a working-group cadence.

The team I led this with
UX Research
Senior researcher partner who co-owned the synthesis and ran the validation studies
Brand
Brand design lead who held me to Charm & Disarm voice and made the framework speak Amazon
Product
Two PM partners across membership growth who pressure-tested every artifact in real launches
Engineering
AI personalization eng who shaped the rubric so it would later score LLM output
Data Science
Helped me design the framework with LLM-grade engineering principles, so structure, rubric, and matrices were ready for machine reasoning
Legal
Reviewed framework guidance against Amazon policy, ensuring value claims and tone choices were defensible globally
My role
Initiator. Author. Scaler. I framed the problem, wrote the strategy, owned the artifacts, and ran the rollout.
Part 01 / The framework

A 0→1 content strategy for how Prime earns trust.

Five chapters: how I diagnosed the problem, how I built the framework, what I shipped, what landed, and what I'd do differently next time.

Chapter 01

Nobody had named the problem yet, so I started by naming it.

At the time, three teams inside Amazon held three different opinions about what Prime should sound like, and members were hearing all of them at once. Research kept arriving from every direction, but each team was reading its own version in isolation, which meant the conflict never got resolved at the level it needed to.

I started by reading every value-communication-adjacent research study I could find. Four major ones existed, completed over eighteen months across three different teams, and while each was a useful finding on its own, reading them together was where the verdict emerged. So I synthesized them and brought leadership a single diagnosis.

FOUR STUDIES, READ TOGETHER How members felt about Prime. STUDY 01 · LOVE LETTERS Members felt convenience but also overconsumption and pressure to buy. CONNECTED FORGIVENESS MANIPULATED OVERWHELMED STUDY 02 · MILESTONE CONCEPT Communication reached too few members, lacked relevance, made no emotional bond. CURIOUS MISTRUST UNAPPRECIATED STUDY 03 · PRIME PASSENGERS Power users were gaming Prime, timing subscriptions around free trials. VALUABLE IRRELEVANT CONFUSED STUDY 04 · VALUE PERCEPTION Prime described as habit. Practical. Transactional. Without emotion. PRAGMATIC TRANSACTIONAL CONDITIONALITY
Fig. 01 My synthesis of four parallel research studies. Connected, manipulated. Curious, mistrustful. Valuable, irrelevant. Pragmatic, conditional. Each pair sat in a different deck. Together they describe the same relationship in trouble.

Then I mapped the internal landscape. Three sets of guidelines were quietly competing for authority across Prime surfaces: Amazon's RIO standards, the Brand voice (Charm and Disarm), and Prime's own retention strategies. Each was strong on its own. Together, they left a gap nobody owned: how Prime talks about its value to members.

THE INTERNAL DIAGNOSIS Three guideline systems. No referee. AMAZON · RIO Acquisition · engagement · retention BRAND · CHARM & DISARM Customers rule · perfect is boring PRIME · CONTENT Lead with intent · inspire trust GAP value comm ↑ THE SPACE NO ONE OWNED. Writers picked the map that suited the brief. Members heard Amazon talking to itself.
Fig. 02 The internal diagnosis. Three guideline systems with useful overlap and a dangerous gap. Writers picked the map that suited the brief. Members heard the inconsistency.

I brought both findings to leadership in a single document. The recommendation: Prime needed a framework that owned the gap between strong benefits and the weak communication around them.

Prime kept telling members what they saved. It rarely told them they mattered.

There was also a value bias to address. Prime's communication leaned hard on shipping speed and the headline benefits. Members already knew about shipping; saying it again was noise. The dozens of secondary perks had low adoption and felt irrelevant when surfaced. The new system needed to be customer-backward, built from what members noticed and needed.

The counterfactual was concrete: with no shared model, the benefits would have kept shipping and the language would have kept fragmenting, with each team optimizing to its own loudest stakeholder. The framework's job was to make that fragmentation visible and replace it with a shared one.

Chapter 02

I built the framework by winning the first fight.

The first decision was about who Prime was being measured against. Senior leadership wanted a benchmarking cohort of Costco and Walmart Plus, which would have anchored the framework to a price-comparison story. I argued for a different cohort: AllTrails, Nike, Starbucks, Thrive Market, Headspace, Target, Duolingo, Spotify. The brands members trust, and the standards Prime was being measured against whether the org said so or not.

I made the case in a deck and a working session. A Costco comparison would have optimized for price-per-benefit and produced more of the same headline-savings copy. A high-trust brand comparison would optimize for the relationship Prime was failing to build, which was the actual problem in the diagnosis. Spotify was the threat we hadn't named. The audit was approved, and that decision shaped everything downstream because it set the bar the framework had to clear.

From there, I designed the system to be answerable to four inputs at once, because anything less would have made it opinion and anything more would have made it unreadable. Every principle, tone, and rubric question started from what members noticed and needed, so the structure inherited the customer-backward stance from the inputs.

FOUR INPUTS, ONE ARCHITECTURE Value Communication FRAMEWORK Values WHAT PRIME STANDS FOR Sentiments WHAT MEMBERS FEEL Guidelines HOW BRAND IS WRITTEN Market WHAT GREAT BRANDS DO
Fig. 03 Four inputs, one architecture. The framework is only as honest as the inputs it answers to. I argued each one in. Two were obvious to leadership. Two I had to defend.

What the high-trust audit returned.

I documented the patterns from real product surfaces across the audit cohort: the language, the timing, and the trade-offs each brand made. Four behaviors showed up in every one, and they became the backbone the framework's principles were tested against.

WHAT HIGH-TRUST BRANDS DO Four behaviors. Every brand. Every time. 01 · TRUST Build trust and relationships. Speak with transparency. Set expectations. 02 · CELEBRATE Celebrate and add delight. Acknowledge beyond transactions. Add joy. 03 · RELEVANT Relevant and timely. Value appears in interactions, fits current goals. 04 · EMPATHETIC Connect to daily lives. Make abstract concrete. Make complex simple.
Fig. 04 Four behaviors showed up consistently across every brand audited. None of them were about better marketing copy. All of them treated members like adults whose attention had to be earned.

I documented the patterns from real product surfaces: the language, the timing, the trade-offs each brand made. Below: the source material the audit was built on.

App screens from AllTrails, Nike, Starbucks, Thrive Market, Headspace showing trust patterns.
04aTrust: transparency, agency, social proof, clarity.
App screens from Nike, Target, Starbucks showing celebration patterns.
04bCelebration: recognition without an offer attached.
App screens from Target Circle 360, Thrive, Headspace, Instacart showing relevance patterns.
04cRelevance: value translated into the member's context.
App screens from Duolingo, Spotify, Claude showing empathy patterns.
04dEmpathy: structural, beyond a voice layer.

From four inputs to one architecture.

The architecture mapped five Value pillars through three Content Principles to nine Tones, read left to right. Writers don't pick one principle and ignore the others; they pick where the emphasis sits.

VALUES PRINCIPLES TONES Monetary Convenience Access Trust Emotion Intuitive EASY TO GRASP Pertinent RIGHT TIME, RIGHT PLACE Empathetic RECOGNITION + RESPECT Comprehensible Efficient Tangible Adaptive Contextual Relevant Dependable Considerate Celebratory A SCALE OF EMPHASIS. WRITERS PICK WHERE TO PUT WEIGHT.
Fig. 05 The full framework architecture. Five values. Three principles. Nine tones. A scale of emphasis. The framework tells writers where to put weight, then leaves the words to them.

The five Value pillars cover what Prime delivers. The first three (Monetary, Convenience, Access) were familiar to the org. Trust and Emotion were the fight. Research showed they carried more retention weight than savings did. I argued the case in research review. Both pillars were added.

The strategic call: add Trust and Emotion to the value model. Research showed they drove retention harder than savings. That decision reshaped every principle and tone that followed.
FIVE VALUE PILLARS What Prime delivers. $ Monetary Prime helps save money Convenience Prime saves time Trust Prime is reliable Access Prime has exclusives Emotion empathetic + delightful ↑ TRUST & EMOTION ADDED. RESEARCH SHOWED THEY CARRIED MORE RETENTION WEIGHT THAN SAVINGS.
Fig. 06 The five Value pillars. Trust and Emotion were added after research showed they carried more retention weight than monetary savings did. Adding them changed the framework. It also changed which content reviews UX got pulled into.

The three Content Principles describe how to write about that value. The argument: emotion creates consistency, confidence, familiarity, and the relationships retention depends on.

WRITING FOR EMOTION CREATES Writing for emotion Consistency SHARED VISION Confidence CONVICTION + IDENTITY Familiarity STRONG PERSONALITY Relationships EASY TO IDENTIFY
Fig. 07 The argument for emotion as a content discipline. This slide opened every framework workshop I ran.
Intuitive
Monetary + Convenience
ComprehensibleEfficientTangible
Members grasp what Prime does without translation. Copy connects benefits to real-world habits and shows, instead of tells.
Pertinent
Convenience + Access
AdaptiveContextualRelevant
Prime evolves with the member's journey. Content anticipates needs, surfaces value at the right moment, personalizes based on behavior.
Empathetic
Emotion + Trust
DependableConsiderateCelebratory
Prime delivers on what it says. Builds confidence through consistency, respects member context, creates moments of recognition that aren't tied to an offer.

The architecture also had to be legible to every person who'd ever touch Prime content: writers, PMs, marketers, designers, eventually AI agents. So I named the surfaces explicitly. Where value communication lives, and who sees it.

WHERE VALUE SHOWS UP Point in line In the moment of benefit usage e.g. CHECKOUT Aggregated Cumulative value over time e.g. DASHBOARD Milestones Anniversary or achievement events e.g. 1-YR THANK YOU Everywhere Members are always evaluating AMBIENT WHO SEES IT Non-Prime Interested but not members ACQUISITION New Just starting their Prime journey ONBOARDING Tenured Experienced members RETENTION Returning Engaging infrequently RE-ENGAGEMENT ↑ 4 CONTEXTS × 4 SEGMENTS = 16 DECISIONS A WRITER MAKES BEFORE A SINGLE WORD.
Fig. 08 Four contexts × four segments. Sixteen content decisions a writer makes before drafting a single word. Naming this matrix made content reviews shorter and content briefs more decisive.
Chapter 03

Four artifacts turned the framework into a practice.

A framework on a slide doesn't ship. It needed four operational artifacts to scale beyond me, and I built each one with the people who'd use it.

Artifact 01 / The matrices.

Two operational matrices translated the principles into specific content moves for every audience-and-surface combination. Aggregated Value covered the savings dashboards and benefit summaries, while Milestone & Recognition owned anniversaries, achievements, and welcome moments. Together they answered the question every writer kept asking the brief: which tone for which surface, for which member, at which moment?

The Aggregated Value matrix mapping member segments against value pillars with prescribed principles and tones.
Fig. 09 Aggregated Value matrix. A writer landing on a benefit dashboard for a tenured member knows from this matrix that they're writing toward Pertinent (Contextual, Relevant). The matrix gives them the assignment in copy: personalize savings summaries to reflect their unique usage and priorities.
The Milestone and Recognition matrix mapping member segments against value pillars for celebration moments.
Fig. 10 Milestone & Recognition matrix. The 1-year anniversary surface for a tenured member maps to Empathetic (Celebratory & Delightful). The matrix is the assignment: honor their loyalty with genuine celebration that reflects years of partnership.

The framework at copy level. A 30-day welcome card for a new member, before and after:

Surface 01 · 30-day welcome card · New member
Before Welcome to Prime! Enjoy fast, free shipping on millions of items, plus exclusive deals and entertainment.
After A month in. We're glad you're here. So far Prime has saved you 14 hours and $47, and it's only getting started.
→ Empathetic (Celebratory & Delightful) layered with Intuitive (Tangible). Honors the milestone, makes value concrete, opens forward.
Surface 02 · Annual savings dashboard · Tenured member
Before Your Prime Savings: $312 saved this year on shipping and exclusive deals.
After Three years with Prime. This year alone, your membership returned $312, 47 hours of shipping time, and 184 movies and shows. Your top-used benefit: same-day delivery on weekday afternoons.
→ Pertinent (Contextual + Relevant) leading, Intuitive (Tangible) supporting. The time anchor ("three years") taps recency and salience to ground the member in a habit they already trust, and the dual-channel proof (dollars + hours + movies) lets one reading carry both abstract savings and concrete usage. Same-day delivery as the "top-used benefit" leans on existing affordance strength instead of introducing new ones.
Surface 03 · Re-engagement card · Returning member (90+ days inactive)
Before We miss you! Come back to Prime and rediscover your benefits. Free shipping, exclusive deals, and more await.
After It's been a while. No pressure. Whenever you're back, your benefits are still here, and three new ones launched while you were away.
→ Empathetic (Considerate + Dependable) leading, Pertinent (Adaptive) supporting. Lowers the emotional cost of returning by removing guilt cues, and the "three new ones launched while you were away" line adds positive novelty without the loss-aversion pressure of a deadline. Treats absence as adult behavior, which is what makes the return feel like a choice rather than a recovery.

Artifact 02 / The writing process.

Four steps with checkpoints, designed to scale to non-writers.

WRITING FOR VALUE COMMUNICATION Four steps. Anyone can run them. 1 Place IDENTIFY THE PILLAR Pick the value pillar and the tone scale. 2 Write APPLY THE MATRIX Use guidelines, VC matrix, and AI tools to draft. 3 Score RUN THE RUBRIC Audit content with screen-level and flow-level rubrics. 4 Verify UX REVIEW Check in with a UX writer for review and edits.
Fig. 11 Place identifies the value pillar and tone. Write applies the matrix. Score audits with the rubric. Verify routes through a UX writer for review. Four steps a marketer or PM can follow without me in the room.

Artifact 03 / The rubric.

The rubric was the artifact I co-designed most closely with UX research. Working with research solidified my conviction: Prime was failing emotionally, and A/B testing alone missed it. A/B optimizes for clicks. It narrows success to single-point conversions, hides emotional response, and rewards false positives that can look like wins quarter over quarter. The framework needed a measurement layer beyond clicks. The rubric became that layer.

Structurally, it is a four-quadrant scoring instrument covering Value Clarity, Audience Alignment, Context & Timing, and Communication Style, with five points possible per quadrant for a total of twenty. Each quadrant is built from explicit yes-or-no questions a non-writer can answer.

VALUE COMMUNICATION RUBRIC Twenty points. Four quadrants. Anyone can score. VALUE CLARITY _/5 → Is the value pillar visible? → Is value quantified where possible? → Does it lead with customer value? → Are value verbs used effectively? → Are limitations clear? AUDIENCE ALIGNMENT _/5 → Right messaging for the segment? → Addresses real needs / pain points? → Framed for their journey stage? → Personalized without distraction? → Customer-backwards, everyday value? CONTEXT & TIMING _/5 → Is value prioritized in the hierarchy? → Does placement make sense? → Is key info scannable? → Discovery balanced with function? → No competing messages? COMMUNICATION STYLE _/5 → Empathetic tone without pressure? → Follows Prime voice guidelines? → Concise and jargon-free? → Accurate, builds trust? → Respects member privacy and intent? TOTAL _/20
Fig. 12 Twenty points. Four quadrants. Anyone can score, including AI. Once "good" was measurable, content reviews stopped being subjective and AI review became possible. (Part 03 picks that thread up.)

What the rubric looks like in use. The same surface (a 30-day welcome card), scored before and after the framework took hold:

Draft 01 · Pre-framework
"Welcome to Prime! Enjoy fast, free shipping on millions of items, plus exclusive deals and entertainment."
Value Clarity
3
Audience
2
Context
3
Style
3
Total: 11/20
Fig. 13 The rubric in use. Same surface, same brief, seven points of measurable improvement. The score is the conversation. The score is also what AI can be trained on.

Artifact 04 / The scaling plan.

Then the part nobody talks about: socialization. I ran sessions across content marketing, retention, acquisition, and the brand team. The pattern was consistent. Agreement in the room, drift the moment people went back to their briefs. The framework needed three commitments to take hold.

SCALING THE STRATEGY From doc to discipline. 01 Elevate standards Move from sharing value to telling customer- backwards stories. RUBRIC + AI FEEDBACK 02 Develop conviction UX writers shouldn't just advise. They should convince teams of direction. EVIDENCE-BACKED REVIEWS 03 Evangelize Teach. Defend. Revise in public. The system has to feel alive in use. WORKSHOPS + CHANGELOGS
Fig. 14 Three commitments that took the framework from a doc to a discipline: Elevate Standards. Develop Conviction. Evangelize.

The realization curve: how value moves through time.

The framework also accounted for time, since members don't experience value all at once. They move through a curve of promised, perceived, and proved, and each stage requires different content moves for different segments. This map became the briefing tool I gave PMs before any new surface review.

VALUE REALIZATION FRAMEWORK Promise → Belief → Proof. NON-PRIME NEW TENURED RETURNING Potential PROMISED Lead with functional proof Introduce multi- benefit potential Promise sustained innovation Reframe value via curiosity + FOMO Perceived BELIEF Establish credibility Reinforce connection + habit Reaffirm personalized value Reignite via novelty + belonging Realized PROOF N/A Show tangible return on member. Reward loyalty, show consistency Prove ROI via re-engagement ↑ EACH SEGMENT INTERNALIZES VALUE DIFFERENTLY. THE COMMS HAVE TO MEET THEM WHERE THEY ARE.
Fig. 15 Value moves from promise to belief to proof. The framework tells the writer which stage a member is in, and what content move that stage needs.

The hardest decisions I made.

Frameworks read clean in retrospect, but the decisions that built them never do. Three I'd defend in any interview:

Decision log Three calls that changed the framework.
Decision 01 Collapse four principles to three. An earlier draft had a fourth principle for Trust. I argued in research review that trust and emotion belonged inside Empathetic rather than alongside it. Three principles were defensible in a workshop. Four were a list.
Decision 02 Audit Spotify over Costco. Senior leadership wanted price-comparison. I argued for high-trust brands instead. Spotify was the relationship Prime was failing to build. Won the case with a deck and a working session.
Decision 03 Build the rubric before the principles were finalized. Conventional sequence is principles → matrix → rubric. I built the rubric in parallel because measurability was the only thing that would let the framework scale to AI. The rubric forced the principles to stay specific.
Chapter 04 · What landed

From a strategy doc to a working content system across membership growth.

1
Framework owned by membership growth
4
Operational artifacts shipped
10+
Cross-functional teams aligned on the rubric
200M+
Global members in the framework's scope

The framework graduated from a strategy doc into a working operating system for membership growth content. Three concrete signals it had taken hold:

The moment I knew the framework had landed: a PM I'd never trained walked into a content review with the rubric already filled out. She'd scored her own draft a 14 out of 20 and brought specific questions about Pertinence and Empathetic tone. The vocabulary had moved from my deck into her brief without me in the room. That's the only adoption signal that matters.

Inside the first two quarters, the framework informed copywriting across 16 surfaces in acquisition, onboarding, retention, and re-engagement, reaching every Prime member at the moments that mattered most. The rubric proved durable enough that engineering used the same 20-point scoring layer for the AI evaluation work in Part 03 without rewriting it, which is the carryover that turns a framework into infrastructure. The harder business numbers, what happened to retention, AURORA conversion, the Milestones launch, and the AI personalization engines, live in Parts 02 and 03. Part 01 made all of that possible.

Note on metrics. Adoption counts (10+ teams, 200M+ members in scope) are operational and verifiable through documentation. Annual lift figures attributed to lifecycle copy are Amazon-internal modeled reads from the membership growth cohort. Specific methodology and cohort sizing are subject to NDA.

Chapter 05

Here's what I'd do differently.

Reflection
A content system becomes legitimate the moment a PM uses it without you in the room.

The lesson I'm still chewing on: shipping the framework was the easy half. Convincing twelve teams it was theirs was the other half. I'd start the socialization plan before the doc is finished next time. Pairing with one PM on one launch convinces the next ten faster than any deck can.

Two things I'd do differently. First, build the AI scoring layer in parallel with the rubric design. By the time we got to AI evaluation in Part 03, the rubric was shaped by human reviewers and cost us a quarter of re-tuning. If engineering had been at the rubric design table from week one, the framework would have been multimodal-ready from day one.

Second, hand off earlier than felt safe. I held onto the matrices for a quarter longer than I needed to because I wanted them right. The matrices got better the moment two writers started iterating on them in parallel without me. The hardest staff-level move is letting your work get changed in ways you wouldn't have changed it.

Up next
Part 02 of 03
The framework, applied to membership growth.
Part 02 / The work

How the framework decided what to say on every screen.

The framework is the brief. The matrices are the rules. The rubric is the score. This is how all of it shipped to 200 million members across the surfaces they use.

Chapter 01

The framework expanded into four member journeys.

The framework gave Prime a shared model for value. The next move was turning that model into journeys: one for every kind of member Prime had, calibrated to where they were in their relationship.

Non-Prime is a skeptic who has tried other subscriptions and needs proof that Prime is worth the mental real estate. New is in their first 30 days, looking for one genuine win they can point to before renewal becomes a question. Tenured is the renewing member at risk of taking the relationship for granted because nothing on the surface acknowledges the years they've stayed. Returning is the lapsed member the system needs to invite back without making them feel guilty for leaving. Four different reasons to engage, and four different writing jobs.

SEGMENT STRATEGY Each segment, its own value story. SEGMENT PRIMARY PILLARS MESSAGING OBJECTIVE Non-Prime CONSIDERATION Monetary + Convenience Inform & Inspire Articulate breadth and ROI to reduce skepticism New ACTIVATION Convenience + Access + Emotion Guide & Empower Help discover benefits, form habits, gain early wins Tenured RETENTION Emotion + Trust + Monetary Reinforce & Reward Reassure through proof of ongoing value and trust Returning RE-ENGAGEMENT Access + Emotion + Trust Re-engage & Reignite Remind of forgotten benefits, showcase new value ↑ FOUR SEGMENTS. FOUR DIFFERENT JOBS THE COPY HAS TO DO.
Fig. 16 Four segments. Four jobs the copy has to do. Each one anchored to its own primary pillars and its own messaging objective, so the framework could speak differently to each member without contradicting itself.

Each segment also moves through three stages of value realization. Members start with potential (what Prime promises), build perception over time (what they believe), and end with realization (what they can prove). The pillar voice matrix gave writers a target quote per stage, in member language. The copy had to earn those quotes.

VALUE PILLARS · MEMBER VOICE What members say at each stage. PILLAR POTENTIAL PERCEIVED REALIZED Monetary "Savings are available to me" "Prime saves me money each month" "I saved $87 on shipping this month" Convenience "Prime can make shopping easier" "Life runs smoother with Prime" "My orders arrived early 8 times this month" Access "Members get exclusive things I might want" "Being a member feels like part of something" "I got early access to a deal I wanted" Emotion "Prime cares about members like me" "I feel recognized and valued" "I'm celebrated and recognized as a member" Trust "Prime promises reliability" "Prime is transparent and reliable" "I trust Prime" ↑ MEMBER VOICE GUIDED EVERY DRAFT. THE COPY HAD TO EARN THESE QUOTES.
Fig. 17 Five pillars × three realization stages, expressed as member voice. The matrix turned abstract pillars into testable language targets.
Chapter 02

Five objectives shaped every screen.

Five Value Communication Objectives became the shared brief for every surface. Designers, PMs, and writers used them to evaluate any new screen against the framework. If a surface couldn't answer at least two of these, it wasn't ready to ship.

VALUE COMMUNICATION OBJECTIVES Five jobs the framework asks every screen to do. 01 Clarify and unify Prime's value story Communicate the full breadth through clear, consistent messaging that connects benefits into one ecosystem. 02 Contextualize value across the lifecycle Adapt to member intent and stage so messaging always feels timely, personal, and relevant. 03 Show tangible "return on membership" Highlight time saved, money saved, and enjoyment gained to make value visible and emotionally rewarding. 04 Simplify and amplify benefit comprehension Use visual and linguistic hierarchy to make Prime's breadth feel intuitive and easy to navigate. 05 Reflect inclusivity and global relevance Localize value messaging for regional benefit sets and marketplace nuance, while maintaining global brand essence and trust.
Fig. 18 The five objectives. Unify the story. Contextualize across lifecycle. Show return on membership. Simplify comprehension. Reflect global relevance. Every screen that shipped earned its place against this list.
Chapter 03

Here's the work, surface by surface.

Member journeys live on shared surfaces. The Homepage, Search, Cart, and Email all carry every segment, but each segment needs the surface to do a different job. The framework's design decision was to keep the surface architecture consistent across segments, while letting the value moves shift segment by segment.

Surface 01 / Homepage.

The Homepage is the first place a relationship lives or dies, and it has to do a different job for every kind of member who lands on it. For non-members, it has to inspire; for new members, it has to orient; for tenured members, it has to recognize them. Same canvas, three different greetings.

Non-Prime Perceived Value Messaging journey: Homepage, Search, and Detail Page screens with framework annotations showing Awareness & Invitation, Savings re-enforcement, Contextual Value Surfacing, and Rational Conversion.
Fig. 19 Non-Prime Homepage. Lead with tangible savings and frictionless convenience. The pillar focus is Monetary + Convenience, the messaging objective is Inform & Inspire. Search, Detail, and CTYP carry the same logic forward, building a continuous case from awareness to commitment.
New to Prime Potential Value Messaging Homepage with personalized welcome, benefit incentive, and curated discovery.
Fig. 20 New to Prime Homepage. Same canvas, different work. Welcome and Orientation lead. The tone shifts to celebratory and empowering. Pillars become Convenience + Access + Emotion. Objective becomes Guide & Empower. Members feel seen the moment they land.
Tenured Prime Realized Value Messaging Homepage with anniversary moment and benefit re-surfacing.
Fig. 21 Tenured Homepage. The work shifts from persuasion to recognition. Anniversary milestones, benefit re-surfacing, and emotional reaffirmation. Pillars become Emotion + Trust + Monetary. Objective becomes Reinforce & Reward.

Surface 02 / Search and detail.

Search and detail pages are where intent meets value. The framework treated these surfaces as proof moments. Every Prime badge, every savings callout, every shipping promise had to earn its real estate by being timely and contextual.

New to Prime Search, Detail Page, Cart, and Checkout screens with framework annotations: Value Guidance, Membership re-enforcement, Reinforce Ease & Savings, Simplify & Confirm.
Fig. 22 New member journey across Search, Detail, Cart, and Checkout. Each surface does a discrete job: Value Guidance on Search, Membership Re-enforcement on Detail, Ease and Savings reinforcement in Cart, Simplify and Confirm at Checkout. The journey compounds proof at every step.
Tenured member Detail Page, Cart, and CTYP screens with Quantify at decision moment, Aggregate lifetime impact, and Customer Value Metrics annotations.
Fig. 23 Tenured member journey across Detail, Cart, and CTYP. Decision moments quantify accrued value ("$42.50 saved this year"). Cart aggregates lifetime impact. Order confirmation surfaces customer value metrics across savings, deliveries, and orders.

Surface 03 / Cart and checkout.

The framework's most personalized work showed up in cart and checkout. These are the surfaces where members pay, and where Prime's value has to make its strongest case.

Non-Prime CTYP and Order Confirmation screens with Reinforcement at Commitment, Broader Benefit Framing, and Post-Purchase Reflection annotations.
Fig. 24 Non-Prime cart and post-purchase moments. Even non-members see Prime's value clearly: $6.99 missed savings on this order, $230 potential savings on delivery last year, $46 missed savings on exclusive deals. The framework named the cost of not joining without nagging about it.

Surface 04 / Email and account.

Email and account surfaces are where members spend time outside the buying flow. The framework treated these as recognition moments. Less selling, more reflection. Members revisit these surfaces precisely because they want to feel the membership is working for them.

New member CTYP, Order Email, Welcome Email, and Value Recap screens with Celebrate + Educate, Reinforce reliability, Reinforce next steps, and Prove payoff annotations.
Fig. 25 New member email and account moments. Order emails celebrate and educate. The welcome email bridges joining and believing. Value recap surfaces tangible returns ($230 saved, $46 in deals). Each touchpoint converts data into celebration.
Tenured member Account, Orders, Member Email, and Membership Recap screens with Reflect progress, Embed ease, Convert data into celebration, and Anchor longevity annotations.
Fig. 26 Tenured member account and recap moments. Account hub reflects progress. Order history embeds ease. Member emails turn transaction data into a story of achievement. Membership recap anchors longevity in measurable proof.
Chapter 04 · What landed

A framework that became a working content system, with the business signal to back it.

+77K
Member growth from cohort-targeted content
11→17
Avg rubric score, legacy vs framework copy
12+
Surface types touched across the funnel
5×3
Pillars × stages, every screen tested against

The +77K is a number worth unpacking because it didn't come from a single launch. It came from the compounding effect of cohort-specific content iterations across three of the segments shown in Chapter 03. The Non-Prime journey rewrote how value showed up in Search, Detail, and CTYP, replacing generic "fast, free shipping" lines with contextual proof points like missed savings callouts ("$6.99 savings on this order with Prime delivery") and broader benefit framing on non-eligible items. The New member journey shifted from generic welcomes to celebratory orientation, with the Welcome Email and the 30-day Value Recap doing the heaviest lifting. The Returning member journey rebuilt re-engagement messaging around novelty and belonging, rather than guilt and discount urgency.

Each iteration was scored against the rubric before it shipped, then re-scored against actual member behavior afterward. Average rubric scores moved from a baseline of around 11 out of 20 on legacy copy to consistently 16 to 18 out of 20 on framework-aligned content. The cohorts that received the framework-aligned content showed measurably stronger acquisition and reactivation signals than holdout cohorts running on legacy copy. The +77K reflects the net new-member delta attributable to those framework-driven cohorts during the first month of staged rollout, measured against a held-out control group.

I am careful about the attribution because Prime growth has many drivers, and the framework was one ingredient inside a quarter that included pricing changes, catalog expansions, and seasonal demand. What I will defend in any interview is the methodology: rubric scores were instrumented, cohort assignment was randomized, and the control held. The framework's contribution to the +77K is real and bounded; it is not the only contribution to it.

The qualitative signal mattered as much as the number. Writers stopped briefing in opinion and started briefing in framework moves. PMs cited the rubric in reviews without my prompt. The same matrix that scored a New member welcome card was scoring a Tenured anniversary email three quarters later, which is the kind of durability that turns a framework into a working system.

Another signal worth naming. The framework was most useful at the moments it disagreed with the team. When a PM wanted to add another savings callout to a New member journey, the framework said no. The segment was at Activation, the messaging objective was Guide and Empower. Adding a savings callout would have pulled the screen back toward Acquisition messaging. The framework gave the team a vocabulary to disagree productively.

Chapter 05

Here's what the implementation taught me.

Reflection
A content system is only as good as the smallest team that adopts it.

The framework worked when the smallest team in membership growth could read it, apply it, and ship from it. Writing for adoption is a different skill than writing for clarity. Clarity convinces an executive. Adoption convinces a team mid-sprint. I had to learn to write the framework twice: once to defend it, again to make it stick.

The other lesson was about restraint. The framework's job was to remove decisions, not add them. If a writer had to consult three matrices and a rubric for a single string, the system was failing them. The implementation pushed me to keep simplifying the framework even after it shipped. Every quarter, fewer decisions to make, more clarity per decision.

Part 03 picks up where this left off: when the same rubric, the same matrices, and the same five objectives were handed to AI personalization to score, generate, and adapt content at a scale no human team could match.

Up next
Part 03 of 03
Milestones, the first LLM-led celebration in Prime.
Part 03 / The scale

Milestones, the first LLM-led celebration in Prime.

The framework was built so a writer could brief from it, a PM could review against it, and eventually a model could generate from it. Milestones was where that last part went live, with cross-functional governance holding the line on quality at machine scale.

LIVE
The experience

A celebration moment, generated by an LLM, governed by the framework.

The video shows the live experience of a Prime member opening their account on a milestone day, with the savings dashboard, anniversary card, and surfaced benefits all driven by framework-aligned content. Every line was generated, scored, and shipped through the pipeline I designed with Data Science.

Chapter 01

The hypothesis: if a rubric can score copy, a model can generate it.

Most loyalty programs at this scale celebrate the same way: a generic anniversary email, a discount code, a thank-you that feels machine-written even when it isn't. The framework gave us a chance to do something different, because every piece of it (the pillars, the principles, the matrices, the rubric) had been designed from the start to be machine-parseable. The question I brought to leadership was whether we could prove that by handing the framework to a language model and letting it generate celebration copy at the moment a member hit a milestone, with quality controlled by the same rubric a human writer used.

This had never been done at Prime. Generative copy in production was happening in pockets, but nothing had attempted to direct an LLM through a multi-input content framework with cross-functional approval baked into the pipeline. The reason I pushed for this work was that scale was the only honest test of the framework. If it held under machine generation, it was real. If it didn't, the team would learn faster from that failure than from another quarter of human iteration.

THE LLM PIPELINE From framework inputs to live celebration copy. INPUTS Member segment Pillars + tones Milestone trigger LLM Drafts copy across surfaces at scale RUBRIC Auto-score against the 20 framework points SHIP Brand, Legal, Prime Video approved ↑ THE RUBRIC TURNED A SUBJECTIVE EVALUATION INTO A REPEATABLE QUALITY GATE FOR MACHINE OUTPUT.
Fig. 27 The pipeline I designed with Data Science. Member segment, pillar weights, and milestone trigger feed into the LLM as structured inputs. The model drafts copy across surfaces. The rubric auto-scores the output against the framework's twenty-point standard. Brand, Legal, and Prime Video review and approve before launch.
Chapter 02

The governance work was the real unlock.

Generative content in a regulated environment doesn't ship because the model is good. It ships because the right people have agreed in advance on what the model is allowed to do, and on how to verify the model did that. For Milestones, that meant aligning four distinct partners before a single celebration card went live, with my role being to hold the center and translate between them.

CROSS-FUNCTIONAL GOVERNANCE Four orgs aligned before a single line of LLM copy went live. Content Design + UX writing FRAMEWORK OWNER Prime Video CREDITS & CATALOG Brand VOICE & CHARM Legal CLAIMS & COMPLIANCE Data Science LLM ENGINEERING ↑ I HELD THE CENTER. THE FRAMEWORK WAS THE SHARED LANGUAGE.
Fig. 28 The cross-functional governance I built around the LLM pipeline. Prime Video owned the catalog and credit logic. Brand owned voice and Charm & Disarm. Legal owned claims, compliance, and the language of incentives. Data Science owned the model and the rubric instrumentation. I owned the framework that translated between all four.

Each partner contributed a non-negotiable. Prime Video would not allow credit language that overpromised on catalog access. Brand would not approve copy that drifted from Charm & Disarm voice. Legal would not approve celebration claims that risked being read as transactional inducements in regulated markets. Data Science would not run a pipeline whose outputs couldn't be deterministically scored. The framework gave us a shared language to honor every one of those constraints inside a single artifact, which is what made the project shippable.

Chapter 03

The system shipped three flavors, two variants, and five surfaces.

The first launch covered three milestone flavors that mapped to different stages of the member relationship. The 1-month anniversary celebrated early belonging, when a new member is still deciding whether Prime is for them. The 1-year anniversary celebrated proven loyalty, where the work shifted from persuasion to recognition. The "just because" surprise was the most interesting test, because it ran on no anniversary at all and was triggered by accumulated value (savings, deliveries, streaming hours) rather than time.

MILESTONE VARIANTS Three flavors. Two variants. Five surfaces. FLAVOR 01 1-month anniversary Cheers to your first month of Prime EARLY BELONGING FLAVOR 02 1-year anniversary Happy 1-year with Prime PROVEN LOYALTY FLAVOR 03 Surprise "just because" Your benefits just paid for itself UNPROMPTED DELIGHT TWO VARIANTS Incentivized Prime Video credit attached Non-incentivized Pure recognition, no offer FIVE SURFACES PMP YOUR ORDERS POP-UP PUSH SAVINGS TAB
Fig. 29 The variant matrix. Three flavors, two incentive variants (Prime Video credit attached or pure recognition with no offer), and five surfaces (PMP, Your Orders, Pop-Up, Push, Savings Tab). The combination created sixteen distinct celebration moments the LLM had to generate for, each one scored against the framework rubric.

The 1-month anniversary, in two variants.

The 1-month flavor showed up first as a personalized member page (PMP) hero, a Your Orders banner, and a pop-up triggered on entry. The incentivized variant carried a Prime Video credit; the non-incentivized variant was pure recognition. Both were generated against the same Empathetic principle but emphasized different tones, with the incentivized version leaning Celebratory and the non-incentivized leaning Considerate.

Anniversary 1 month variants across PMP, Your Orders banner, and Pop-Up
Fig. 30 1-month anniversary, incentivized variant. "Cheers to your first month of Prime!" carried across PMP, Your Orders, and Pop-Up, with a $6 Prime Video credit attached. Each surface adapted the headline length and CTA placement, but the framework's tone was constant.

The 1-year anniversary and the "just because" surprise.

At 1-year, the copy shifted from "Cheers to your first month" to "Happy 1-year with Prime," and the visual language gained a celebratory weight that recognized the relationship had earned more emphasis. The "just because" flavor was the riskiest test, since it had to feel earned without any anniversary to anchor on. The framework solved that by tying the celebration to accumulated value: "Your membership just paid for itself" when savings crossed a meaningful threshold during a Prime Big Deals Day.

Push and reminder copy.

Push notifications carried the most stylistic risk because they reach members outside the app, on a lock screen where every word counts. The framework's Empathetic principle constrained the LLM toward gratitude rather than urgency, and the reminder logic was tuned to surface only when a credit was about to expire, so the messaging never read as nagging.

Chapter 04 · What landed

First-of-its-kind LLM celebration, with the engagement signal to back it.

1st
LLM-led celebration system at Prime, scoped to membership growth
200M+
Members reachable through the pipeline
+38%
Push open rate vs prior celebration baseline
5.5 mo
From hypothesis to first live milestone

The result that mattered most to me was the time-to-launch. From the first hypothesis pitch to the first live celebration moment took approximately five and a half months across four partner orgs, which is fast for an initiative that had to clear Brand, Legal, Prime Video, and Data Science review on a shared content artifact. The reason it moved that fast is that the framework arrived at every meeting as the shared language, and reviewers stopped debating taste and started debating rubric scores, which compressed every approval cycle.

The push open rate was the strongest engagement signal. Push notifications sent through the framework-aligned LLM pipeline ran at a +38% open rate against the prior celebration push baseline during the first quarter of the launch. That comparison was scoped to a like-for-like audience of members who had received a celebration push in the prior two quarters, which controlled for novelty and seasonality as best the data could support. Push is the surface where copy quality shows up most directly: a generic notification gets ignored, and a personally meaningful one gets opened. The LLM, working through the framework, was producing copy that members were choosing to engage with at a measurably higher rate than what the same teams had shipped manually for years.

I want to be precise about the "first" claim. This was the first LLM-led celebration system shipped inside Prime, scoped to membership growth surfaces. Generative content existed elsewhere in Amazon and across the industry; what was novel here was the multi-input governance pipeline, where Brand, Legal, Prime Video, and Data Science approved the LLM output through a shared rubric before it reached members. The framework was the governance, the rubric was the quality gate, and the LLM was the execution layer.

The qualitative signal followed the engagement number. Brand reviewers stopped flagging tone drift, because the framework constrained the LLM toward Charm & Disarm by design. Legal reviewers approved faster, because claim language was structurally bounded by the matrix rather than a writer's individual choice. Prime Video stopped worrying about catalog overreach, because the credit logic was anchored in framework-defined boundaries. Approval cycles that had historically taken weeks were closing in days.

Chapter 05

Here's what scaling through AI taught me.

Reflection
A framework's hardest test is whether it holds when the writer isn't human.

Milestones changed how I think about content design at scale. The framework was built for human writers, but the moment I started thinking about it as a structured input for a model, every choice I'd made about it got tested differently. The principles had to be parseable. The matrices had to be deterministic. The rubric had to be quantitative enough to auto-score. Designing for human clarity and machine readability turned out to be the same discipline, applied with more rigor.

The hardest lesson was about cross-functional pace. Generative content in a regulated environment moves at the speed of the slowest reviewer, and getting four orgs to a shared standard required me to spend more time translating than writing. Brand needed to see the framework in voice terms. Legal needed to see it in claims terms. Data Science needed to see it in instrumentation terms. Prime Video needed to see it in catalog terms. The framework was the only artifact that could speak all four languages, which is why holding the center mattered as much as building the system.

What I'd do differently. I would build the rubric instrumentation hooks into the framework from day one, instead of bolting them on when the LLM pipeline was being designed. The retrofit cost us a quarter, and a portion of that cost was avoidable if I'd treated the framework as a measurement system from the start, alongside being a writing system.

Up next
Issue №02
Making tax law feel human.
razacontent · Mohammed Raza · Seattle · 2026
Designing content systems since 2017