Architecture How it works Algorithms Features Data & Trust Pricing 中文 Request Access
Amazon Ads API · Selling Partner API

Advertising autopilot for Amazon sellers

Shepherd connects to your Amazon Advertising and Selling Partner accounts and manages bids, budgets, and keywords automatically — every action statistically grounded, gated for safety, and explained back to you.

OAuth via Login with Amazon TLS in transit · encrypted at rest No buyer PII accessed
shepherd · account overview
SYNCED
ACOS
18.4%
↓ 3.2 pts · 30d
CVR
11.2%
↑ 1.4 pts · 30d
SPEND
$4,820
this month
ACoS trend auto-optimized
Raised bid · "wireless earbuds" +12%
confidence 0.91 · logged with rationale
2m
System Architecture

How data moves through Shepherd

Shepherd reads performance data from Amazon over authorized, encrypted connections, processes it through its decision engine, and writes optimization actions back through the official Amazon Advertising API. Nothing leaves this loop.

AMAZON SHEPHERD PLATFORM ACTIONS read read write Amazon Advertising Ads API · metrics Selling Partner SP-API · catalog Ingestion & Normalization dedupe · align · store (TLS) Statistical Decision Engine scores bids · budgets · keywords Safety & Confidence Gate acts only above threshold Bid & budget updates → Ads API write Keyword / target actions → pause · add · adjust Creative & reporting → to your dashboard
Reads — performance data, over OAuth + TLS Writes — optimization actions via Amazon Ads API No buyer PII ever enters the system
Decision Pipeline

Every change follows the same five steps

No black-box guessing. Shepherd runs each decision through a fixed, auditable pipeline — and refuses to act until it has the statistical confidence to back the change.

STEP 01 / 05

Collect performance data

Shepherd pulls the latest campaign, keyword, and target metrics from your Amazon accounts over authorized, encrypted API connections.

decision.log
$ fetch metrics --account active
  ↳ 1,284 keywords · 30d window
Bidding Algorithms

Finding the optimal bid — then scaling it safely

Shepherd estimates each keyword's response curve from its own conversion data, climbs to the most profitable bid, then ramps exposure in controlled steps — re-checking efficiency before every increase.

Optimal bid search

per keyword · response-curve ascent
AI · ascent
↑ profit bid → optimal bid
// estimate response curve, ascend to peak
max profit(bid) = sales(bid) - spend(bid)
s.t. ACoS ≤ target

Each keyword's curve is learned from its own conversion data; the search converges to the bid with the best marginal return.

Scaling profitable terms

放量 · controlled volume ramp
guardrailed
ACoS ≤ target ↑ exposure steps →
// while profitable & confident → ramp
step+1: raise bid / budget by a capped increment
re-check ACoS ≤ target before next step

放量: once a term clears the confidence gate, Shepherd scales exposure in controlled increments — and holds the moment efficiency would slip.

Multi-armed bandit · live

Thompson sampling

Each keyword is an "arm." Shepherd keeps a Beta(α,β) belief about each arm’s conversion rate, samples from those beliefs, and spends on whichever it draws highest — so traffic flows to the best performer while still exploring the rest. As data accrues, the beliefs sharpen and allocation converges.

PULLS
0
CONV
0
Belief: P(conversion rate) per arm
0% conversion rate 30%
Traffic allocation (last 200 pulls)
"wireless earbuds"
"bluetooth headphones"
"earphones case"
// each round
sample θᵢ ~ Beta(αᵢ,βᵢ) · pick argmax θ
observe · update α or β
Data Intelligence

Beyond ads: supply & demand intelligence

The same authorized SP-API connection powers two more engines — FBA replenishment planning and Amazon Brand Analytics (ABA) opportunity mining — so your inventory and demand decisions stay in sync with your advertising.

FBA replenishment

SP-API · FBA Inventory + Orders
read · SP-API
sales velocitylead timeon-handinbound
safety stock reorder point (ROP) time →
// when on-hand + inbound < ROP → reorder
ROP = D̄·L + z·σ√L
qty = D̄·C − (on-hand + inbound)

D̄ forecast daily demand · L lead time · z service-level factor · σ demand variability · C coverage period.

ABA opportunity mining

Brand Analytics · Search Query Performance
read · ABA
Opportunity high demand · low capture ↑ your conversion share search volume →
// rank ABA search terms by missed demand
score = volume × (1 − your_share)
// high score → add as keyword / fix listing

Surfaces high-demand search terms you under-capture, then feeds them back into bidding and creative.

Stockout forecast · sea vs air

lead-time backsolve
SKU CS-014 · daily sales (forecast) 8.6 · WFS on-hand 210
stockout in 22d · 7/9
Production Export consolidation Sea freight Air freight Customs WFS inbound
Sea plan
Production 12d
Sea freight 26d
Air plan
Production 12d
✓ in stock · 6d buffer
Today +15d 7/9 +30d +45d +60d
Recommendation: the sea-freight window has already closed — it would have needed ordering ~22 days ago. Air freight is the only plan that keeps this SKU in stock, with about 6 days of buffer if you order today.

Red hatch = plan finishes after the projected stockout date (won't arrive in time).

AI Listing Studio

On-policy copy and creative, generated

An agent rewrites titles to Amazon's latest 2026 listing policy and validates them before writing back through SP-API. The same studio generates storefront and A+ imagery sized to Amazon's specs.

Title optimization agent

SP-API · Catalog + Listings
agent
CURRENT · policy flags
Best Wireless Earbuds 2026 🔥 Bluetooth 5.3 Noise Cancelling FREE SHIPPING Sale!! for iPhone Android Sports Gym Waterproof
agent · 2026 policy rewrite
OPTIMIZED · compliant
Acme AirPro Wireless Earbuds — Bluetooth 5.3, Active Noise Cancelling, 36H Battery, IPX5 Waterproof
Brand name leads the title
Within category length limit
No promotional claims (best, sale, free)
No emojis or special characters
Key attributes front-loaded
SP-API read agent rewrite validate SP-API write

AI creative generation

storefront window + A+ content
AI · spec-sized
A+ hero · 970×600 generating
Storefront window · variants

Generates on-brand storefront and A+ imagery sized to Amazon's specs — you review the variants and publish the ones you want.

Under the Hood

The methods behind each decision

No black box. Shepherd applies established statistical techniques — the same families used across quantitative advertising and supply-chain forecasting.

Advertising & bidding
campaign optimization
Thompson sampling (bandit)Beta(α,β)
Balances exploring new keywords against exploiting proven winners.
Bayesian conversion estimationP(cvr | data)
Beta-Binomial posteriors with credible intervals drive the confidence gate.
Sequential significance testingSPRT
Tests significance before pausing, controlling false calls on low-traffic terms.
Marginal-return bid search∂profit/∂bid = 0
Estimates each keyword's response curve and ascends to the profit-maximizing bid.
Constrained budget allocationargmax s.t. ACoS≤t
Distributes daily spend to maximize return under your ACoS constraint.
Inventory & forecasting
FBA restock planning
Holt-Winters forecastinglevel+trend+season
Exponential smoothing captures trend and seasonality in daily demand.
Safety stock (service level)z·σ√L
Buffers demand variability using the target service-level z-score.
Reorder point & EOQROP, EOQ
Decides when to reorder and how much, balancing holding cost against stockout risk.
Lead-time demand distributionD ~ N(μL, σ√L)
Models cumulative demand over variable supplier lead times.
Anomaly detectionrobust z-score
Flags demand spikes and outliers before they distort the forecast.

Method names describe the algorithm families Shepherd draws on; exact parameters are tuned per account and category.

Capabilities

Six engines, one autopilot

Each runs continuously on your campaigns. Every decision is data-driven, statistically grounded, and fully explained.

Automated bidding

Adjusts keyword and target bids from each campaign's own performance history — up where conversions are strong, down where they're not.

Confidence-based pausing

Waits for statistically sufficient data before pausing a keyword or target — no premature calls on low-traffic terms.

Smart budget allocation

Distributes daily spend toward the placements most likely to convert, instead of spreading budget evenly.

Competitive strategy scoring

Evaluates competitive activity around your products and prioritizes targets where you can win placement efficiently.

Creative generation

Helps generate and refine ad creative tailored to your listings — more variations to test, no extra production work.

Performance reporting

Surfaces ACoS, CTR, CVR, and spend trends in one place — with a plain explanation of every automated decision.

Data & Trust

How we use Amazon data

Shepherd connects only through Amazon’s official OAuth flow (Login with Amazon). You stay in control — and you can revoke access at any time from your Amazon account.

What we access

With your explicit authorization through Login with Amazon, Shepherd accesses advertising performance and account data needed to operate your campaigns — campaign, ad group, keyword, and target metrics; spend, sales, and conversion data; and product and listing information.

What we do not access

Shepherd does not request or store buyer personally identifiable information — no buyer names, shipping addresses, or contact details. We use only the data required to optimize advertising.

Why we use it

Solely to provide the automation you sign up for: adjusting bids and budgets, managing keywords, generating creative, and reporting performance back to you.

How we protect it

All data is transmitted over HTTPS/TLS and encrypted at rest. Access is least-privilege and need-to-know. Authorization tokens are stored encrypted and never shared.

Sharing

We do not sell your data, and do not share it with third parties except trusted infrastructure providers that process it on our behalf under equivalent protection obligations.

Retention & deletion

We retain data only as long as needed to provide the service. Revoke access anytime from your Amazon account, or request deletion at support@manualstool.com.

About

Built and operated by 深圳市博纳东方电子商务有限公司

Shepherd is built and operated by 深圳市博纳东方电子商务有限公司, based in 深圳市龙岗区龙岗街道平南社区龙岗路19号东森商业大厦(东嘉国际)221. We build software that helps Amazon sellers run their advertising more efficiently through automation — and we build and maintain this website and our application independently.

Independently operated
This site and the Shepherd application are built and maintained by our team.
深圳市龙岗区龙岗街道平南社区龙岗路19号东森商业大厦(东嘉国际)221
Shenzhen, China — serving Amazon sellers across US and EU marketplaces.
Direct support
Reach a real person at support@manualstool.com.
Pricing

Pricing tailored to your account

Shepherd is priced to fit the size of your Amazon business and managed ad spend. Rather than one-size-fits-all tiers, we scope a quote to your account — request access and we'll share pricing for your setup.

Custom quote
  • Scoped to your account size and managed ad spend
  • Onboarding, configuration, and ongoing support included
  • No long-term lock-in — cancel anytime
Request a quote

Pricing scales with your account — no setup fee.

Contact

Get in touch to request access

For support, questions, or partnership inquiries, reach our team directly. We respond to every message — typically within one business day.

support@manualstool.com