Asking questions your dashboards can't answer
For long-tail questions the standard reports don't cover, the agent can query your own Amazon data directly, read-only. What it can reach, how to ask, and how to export the result.
TL;DR: When a question needs a custom date window, search-term detail or a cut no report offers, the agent writes and runs a read-only query over your own Amazon data and answers from the rows. You just ask in plain language. It cannot change anything this way, and it can hand you the full result as a CSV download.
Autron's ready-made numbers answer the common questions fast. For the ones they do not cover — long-tail search terms, an unusual date window, a comparison nobody built a report for — the agent can go to the underlying data and work it out.
What this is for
The standard numbers are prepared in advance for fixed periods: the last 7, 30 or 90 days, last month and last quarter. That makes them quick, and it is also their limit. They cannot answer an arbitrary window, week-over-week movement, or a slice no report offers.
For those, the agent writes a read-only query over your own data and runs it. You describe what you want to know; it writes the query. It can only read, so nothing in your account or on Amazon changes.
What it can reach
- Your ad reports, down to each customer search term, and by placement.
- Sales and traffic from Seller Central, day by day and product by product, when Selling Partner is connected.
- Search Query Performance from Brand Analytics, week by week.
- Your Sponsored Products campaigns and which products each one advertises.
- Review requests: which orders Autron asked Amazon to request a review for.
- The ad engine's work: each change it sent to Amazon, and the bid it priced for each target.
It looks at the marketplace you are working in, unless you ask it to compare across marketplaces.
When to reach for it
- Long-tail search-term questions that a summary report rolls up.
- Custom comparison windows that do not match a standard period.
- One-off questions nobody built a report for.
For everyday performance questions the standard numbers are faster, and the agent uses them first. You never need to ask for a query by name.
Asking a good question
Describe the answer you want, not the query. "Which search terms spent more than £50 last month and never converted?" works better than naming tables or columns. If the first answer is not quite the cut you meant, say so and ask again; refining in conversation is faster than getting it right first time.
Export the result
The agent reads up to 1,000 rows in a chat. Ask for a CSV and it saves the full result, up to 50,000 rows, as a download link that works for 30 days.
Where it works
In the Autron app, and through the connector in Claude or ChatGPT. See Use Autron with Claude AI and Use Autron with ChatGPT.