A10. COSMO. Rufus. Three quiet shifts that broke every SOP we wrote before 2025. This manual is our ground-up rebuild — domain by domain, account by account. Read it, run it, push back on it. This replaces nothing yet, but every SOP it conflicts with is on the rewrite queue.
If your accounts have been bleeding since March, it's not because you stopped doing your job. The job changed under us. Every team, every L-level, every niche — the same curve. February peak. March cliff. April keeps falling. That's the market, not you.
What is on us, starting now: 50% standard SOP, 50% reasoning from the new rules in this manual. If you run only the SOPs, you'll keep losing. If you run only your gut, you'll lose differently. Half and half. Test, log, escalate. Self-learning is now a KPI — see the Crew Learning Log.
I'm not going to write you a script for every account. I'm giving you the new rules. You apply them.
— Zain
A clean look at what changed across our book of accounts, drawn from the STS sheet through April 2026.
| Account | Feb avg/day | Mar avg/day | Apr avg/day | ACOS Feb→Apr |
|---|---|---|---|---|
| Amelia | $1,036 | $279 | $187 | 52% → 35% → 69% |
| PWY | $410 | $133 | $121 | 32% → 35% → 52% |
| FLE | $266 | $120 | $140 | 52% → 54% → 77% |
| Gaurav | $270 | $117 | $111 | 58% → 67% → 80% |
| Dan | $237 | $119 | $102 | 37% → 39% → 51% |
Three layers got rewritten. Most of our SOPs are still A9-era. That's the gap.
A9 was simple: bid × relevancy × velocity. A10 layered three new pillars on top:
COSMO is the knowledge graph under search and ad delivery. It infers why someone is searching, not what they typed. "Book for anxiety in teens" and "anxiety book teens" are now different audiences, eligible for different ads. Match types still exist — but they sit on top of intent matching. Broad match in 2026 is a semantic engine, not a fuzzy-string engine.
Rufus mediates 15–20% of mobile queries and ~38% of Black Friday 2025 sessions. Rufus reads your listing — title, description, A+ text, reviews — to answer the buyer's question. Rufus-mediated clicks convert ~3.5× higher than search clicks because the AI pre-qualifies intent. If your listing reads like a keyword soup, Rufus skips you. If it reads like a clear answer, Rufus lifts your sentences.
Publishing flow needs a rebuild for an AI that summarizes (Rufus) and an algorithm that infers (COSMO). Walk every step.
The "low-comp keyword + generic book" era is dead. COSMO clusters readers by intent, not keyword. Specificity is now a moat.
| Niche signal | Verdict | Why it matters for us |
|---|---|---|
| Bold & Easy coloring | Go | Biggest growth trend since late 2024. Still scaling. Test on accounts with kids/leisure portfolios. |
| Wellness journals with prompts | Go | FLE territory. Guided prompt journals outsell blank consistently. Push into burnout/anxiety sub-segments. |
| Cottagecore / cozy micro-niches | Go | <200 books per sub-theme, $500–3K/mo realistic. Good early-stage account fit. |
| Personal finance — system-based | Go | Readers reject "get rich" framing. Specific systems (debt, budgeting, sinking funds) win. |
| Educational layered workbooks | Go | Repetitive low-content is dying. Layered complexity holds dwell time and review depth. |
| Generic lined journals / blank notebooks | Avoid | If we're still publishing these on any account, stop. No semantic anchor for COSMO. |
| Mass AI-generated fiction | Avoid | Rufus reads reviews. Bad AI fiction gets shredded fast and the signal kills rank for the whole pen name. |
| Generic "self-help" without a problem hook | Risky | Without a specific reader problem, COSMO can't cluster you. You bid against everyone — see RHEA. |
Rufus reads inside your book — Look Inside, sample reads, review-mined themes. The current SOP "Write a Book Using AI" is solid for outline-driven generation but needs updates for Rufus-readability.
Current SOP says: research → competitor analysis → outline → chapter-by-chapter generation. Add these steps:
Cover hits the human before the algorithm. CTR feeds dwell-time and conversion signals downstream. Weak cover punishes everything.
Our current Listing Details SOP covers 9 elements: Title, Subtitle, Author, Keywords (PB/HC + Kindle), Categories (PB/HC + Kindle), Description, HTML Description (Book Beam), Book Blurb, A+ Content. The structure is right. The logic inside it is A9-era. Specifically: the keyword approach uses variant stuffing, which now wastes slots.
ramadan bedtime stories for kids / ramadan stories for children / ramadan books for kids / ramadan tales for kids / islamic stories for childrenEach slot must target a different angle. Same Ramadan book, redone:
| Slot | Intent angle | Example phrase |
|---|---|---|
| 1 | Reader problem / outcome | "teach my child about ramadan" |
| 2 | Audience age / format | "picture books for muslim toddlers" |
| 3 | Trope / theme | "family-centered islamic traditions" |
| 4 | Occasion / use-case | "ramadan gift for nieces and nephews" |
| 5 | Comparable / category context | "goodnight moon style for muslim kids" |
| 6 | Specific value / benefit | "teach gratitude through stories" |
| 7 | Long-tail problem search | "how to explain fasting to children" |
Bad: "Anxiety Workbook"
Better: "The Anxiety Workbook for Teens: 12 Weeks of Guided Exercises to Quiet a Racing Mind"
Total title + subtitle ≤ 200 characters. NLP layer flags unreadable, stuffed titles. If a human wouldn't say it, the algorithm down-weights it.
Every sentence in the description should be liftable by Rufus to answer a shopper question. Specific claims, clean sentences, zero stuffing.
| Slot | Strategy | Goal |
|---|---|---|
| 1 | Broad parent | Credibility, foot traffic |
| 2 | Mid-competition sub-category | Reachable BSR rank |
| 3 | Tight niche where we can rank top-20 | Bestseller badge eligibility |
Watch ghost categories — appear in browse menu but get zero browse traffic. Validate with Rocket category traffic data before committing.
A+ was a "nice to have" under A9. Under A10/Rufus it's a primary ranking and AI-feeding asset. Books with A+ see ~12% higher conversion. Rufus actively scrapes A+ text modules. Image-only A+ is now leaving money on the table.
SB Video is one of the highest-leverage formats we under-use. Custom-image SB ads get 48% higher mobile CTR than product-only. Video pulls dwell time and pre-sells intent before the click.
The auto + broad + phrase + exact mining structure that worked under A9 leaks money in 2026. COSMO already does the matching. Our job: organize intent, control placement, feed external pull.
Stop setting ACOS targets by feel. Set by math.
$5.14 royalty on a $12.99 paperback = 39.6% breakeven. Anything under is profit. Anything over is paying Amazon to take readers. Run every account at its breakeven, not "industry average."
| Stage | Target | Strategy |
|---|---|---|
| Launch (0–60 days) | +10–20% over breakeven | Buy rank, feed velocity. Loss leader by design. |
| Growing (60–180 days) | ≈ breakeven | Hold rank, build review base. |
| Mature (180+ days) | −10–20% under breakeven | Profit phase. Cut underperformers, double on winners. |
| Series book #2+ | Higher than breakeven OK | Read-through pays back the loss. Track series LTV separately. |
2026 industry average is 32–35% ACOS. Top performers hold 23–26%. RHEA at 164% and GAURAV-2 at 213% are far past the threshold where bid tuning matters — those are listing/category problems, not ad problems.
Old: one campaign per match type. New: one campaign per intent layer, match type as a control inside.
| Layer | Targets | Bid posture | ACOS expectation |
|---|---|---|---|
| Problem | Reader pain / outcome KWs ("how to sleep with anxiety") | Aggressive — highest intent | Below breakeven |
| Competitor | Competing author/title KWs + ASIN targeting | Moderate — defensive | At/below breakeven |
| Branded | Author name + previous titles | Defensive minimum | Lowest ACOS, profit core |
| Generic / Head | Broad category terms ("self help," "romance") | Cap-and-control | Highest, often unprofitable alone |
Match types still matter — but as control layers, not the organizing principle. Exact for proven phrases, phrase for testing, broad sparingly with negative-keyword rigor (because COSMO is now doing the matching for you).
Most of our SOP work is on bids. The math has changed and most operators get this wrong.
$1.00 base × Dynamic Up & Down (max 100% TOS) × 50% TOS multiplier = $3.00 effective CPC. Operators forget this. One keyword burns the daily budget and they wonder where the spend went.
| Stage | Bidding | Placement | Why |
|---|---|---|---|
| Launch | Dynamic Up & Down | TOS +25–50% | Buy velocity. Algo rewards conversion-likelihood with up-bids. |
| Scale | Dynamic Up & Down | TOS +20%, PP +20% | Stack what works. Watch compounding burn. |
| Stabilize | Dynamic Down Only | 0% on losers, +20% on winners | Cut waste. Down-only stops Amazon over-bidding shaky placements. |
| Defend (branded) | Fixed Bids, low | Low TOS | Don't need Amazon to optimize the author's name. |
A10 weighs an external sale ~3× an internal click in rank. Almost no account in our book sends external. This is the single biggest unfair-advantage gap we have.
Brand Registry-enrolled accounts get ~10% of sale price credited back on sales driven from external traffic via Amazon Attribution tags. Stack with the 3× rank lift and external traffic becomes our single highest-leverage move.
Auditing our actual SOP behaviors against 2026 reality. The audit:
| Tactic | Status | Why retire |
|---|---|---|
| The 5-minute campaign pause | Myth | No evidence Amazon "resets" anything in 5 min. Post-pause "boost" is regression to mean. Stop building SOPs on it. |
| Auto-only or auto+manual default structure | Outdated | COSMO does the matching. Intent-layered campaigns outperform mining-from-auto. |
| Bid-only optimization (raise/lower 10% weekly) | Outdated | Ignores placement compounding. Tuning the wrong knob. |
| Image-only A+ Content | Outdated | Rufus can't read image text. Invisible to AI summarization. |
| Variant-stuffed backend keywords | Wasteful | 5 slots collapsing to 1 intent. Lose 4 slots of reach. |
| Same campaigns for ebook + PB + HC | Outdated | Different price points, different reader behavior, different breakeven. |
| "Industry average" ACOS targets | Lazy | Use breakeven math. Industry average is irrelevant to our P&L. |
| Aggressive broad without rigorous negatives | Bleeding | COSMO broad-matches on intent. Without negatives, we spend on synonyms we never wanted. |
| Move | Difficulty | Expected lift |
|---|---|---|
| Rewrite description for Rufus-readability | Low | 5–15% conversion lift |
| Convert A+ to text-rich 6-module structure | Medium | 10–20% conversion lift |
| Re-bucket campaigns into 4 intent layers | Medium | 10–25% ACOS reduction |
| Audit placement modifiers, set winners only | Low | 5–15% ACOS reduction in 2 weeks |
| Set up Amazon Attribution + 1 external channel | Medium | 3× rank lift on external + 10% BRB |
| SB video for top 3 books per account | Medium | 2–4× CTR vs product-only |
| Dayparting on top-spending campaigns | Low | 5–10% ACOS reduction |
| Search Term Report → 100+ new KWs/quarter | Medium | Up to 4× sales scale (case-dependent) |
| Trope-clustered KW expansion (fiction) | Medium | 10–25% reach lift on series |
The OPT Log shows L1 optimization activity. Hassaan's L1s have been logging 60–85% non-optimized days (0️⃣) on 7 accounts. That's the biggest single avoidable bleed in the book right now.
Concrete list of what's getting rewritten and the priority. Anyone can pull a P0 — assignment in the next sync.
| SOP | Priority | Why | Owner |
|---|---|---|---|
| Bidding SOP (Asana 1209509081406353) | P0 | Uses A9-era Ad Rank formula. L1s execute it daily. Every day costs us money on RHEA / GAURAV-2. | Zain + Inam |
| Listing Details SOP (Asana 1210283955451775) | P0 | Variant-stuffing keyword logic wastes 4 of 7 slots. Highest-volume SOP. Touches every new title. | Inam |
| Level 1/2/3/4 SOP (Asana 1208779988398298) | P0 | Defines what each L-level does daily. Currently anchored to A9 doctrine. Every team member runs against this. | Zain |
| Amazon Features Update April 2024 | P1 | 2-year-old. Replace with rolling 2026 update doc. | Zain |
| Write a Book Using AI | P1 | Add Rufus-readability + humanize + AI-disclosure steps. | Publishing crew |
| A+ Content + Video Ads Example | P1 | Replace with text-rich 6-module structure + 5-beat video script. | Publishing crew |
| OPT Log doctrine | P1 | New rules above need to land in actual OPT Log doc. | Hassaan + Inam |
| NEW — Placement-Modifier Audit SOP | P0 | Doesn't exist yet. Highest-leverage missing SOP. | Zain to draft |
| NEW — External Traffic + BRB Setup SOP | P1 | Doesn't exist. Biggest unlock for accounts >$2K/mo. | Zain to draft |
| NEW — Crew Learning Log doctrine | Done | Template already drafted. Roll out next sync. | Crew |
Self-learning is now a measured KPI, not a soft expectation. The Crew Learning Log template is the mechanism — every Crew member fills it out weekly.
A9 was Amazon's product ranking algorithm from ~2014–2024, weighted on bid × relevancy × sales velocity. A10 (2024–present) added semantic relevance, external traffic quality, and engagement depth as primary pillars.
Amazon's commonsense knowledge graph that infers user intent from co-buy and search-buy behaviors. Powers Rufus and the broad-match layer in advertising.
Amazon's conversational AI assistant. Mediating 15–20% of mobile search queries. Reads listings to answer shopper questions. Rufus-mediated clicks convert ~3.5× higher than search clicks.
Amazon program crediting ~10% of sale price back to brand-registered sellers for sales driven from external traffic via Amazon Attribution. KDP-eligible if Brand Registry enrolled.
The ACOS at which ad profit is zero. Royalty per sale ÷ list price. Above = paying Amazon. Below = profit.
2026 replacement for match-type-first campaign structuring. Organize by reader intent, control match type within.
Final CPC = Base × Dynamic × Placement. $1 × 2× Dynamic × 1.5× TOS = $3 effective. Common cause of unexplained budget burn.
This manual is a living document. If something here doesn't match what you're seeing in the data on your accounts, that's not a problem — that's the entire point. Tell me. The fastest way to improve this is for the Crew to push back on it with real account evidence.
If a section costs you a sale, I want to know in the next 1:1.
We figure this out together or we lose together. I'd rather the first one.
— Zain