VendoolyWhat actually took the time
Pulling reports, reconciling them in a spreadsheet, cross-referencing keyword spend with stock, working out which ASINs lost money after fees and ads. Hours of evidence-gathering before any decision was possible.
// Our position
Automation should propose. People should decide.
"Fully autonomous" was the industry's favourite promise until sellers started counting what it cost them. We took the other route on purpose: the assistant investigates without limits, and asks before it touches anything. As of 2026, that is not just the safer design — it is the one Amazon's agent rules expect.
FreeReading: orders, stock, Buy Box, campaigns, account health — as much as you want.
Proposed"Lower this keyword's bid from £0.62 to £0.35 — 1 row affected."
YoursConfirm, adjust or reject. Nothing reaches Amazon until you do.
The boundary stays visible at all times
The honest objection
"Doesn't asking me every time defeat the purpose?"
It would, if the slow part of running an Amazon account were the clicking. It never was.
What automation removes
All of that. The assistant queries your authorized data directly and hands you the finding — with the numbers behind it — in the time it takes to ask the question.
What confirmation costs
A second. And it is the second that stops a misread instruction from repricing forty SKUs below margin while you sleep.
How it works in practice
Three states, never blurred.
Investigate
Read operations run without friction: catalogue, listings, inventory, orders, pricing, Buy Box context, account health, Sponsored Products performance. Ask follow-up questions freely.
Propose
Any write is shown first: the SKU or campaign, the value before and after, and how many rows the instruction would touch. Nothing has happened yet.
Decide
You confirm, change the numbers, or reject. Only confirmation sends the request to Amazon — and bulk operations are additionally capped server-side.
Why the industry is turning
Autopilot became a liability.
The rules changed
Amazon's agreement now describes a tiered model in which high-impact actions — large price moves, bulk listing creation, account configuration — expect a documented human authorization. Tools built to act unsupervised have a design problem, not a marketing problem.
The failure mode was always mundane
Not dramatic breaches: ordinary mistakes at scale. A competitor's pricing glitch chased to the bottom, campaigns paused that were carrying the month, the wrong SKU set edited in bulk.
Accountability needs a record
When something goes wrong you have to reconstruct what changed, when, and on whose authority. An approval step is where that record comes from.
Summary of public reporting, not legal advice. The authoritative source is the current agreement in your Seller Central account.
Related
Read on.
Ask for a change and watch it wait.
Connect one seller account and request a price or bid update. The proposal appears; nothing moves until you say so.