Skip to main content
Nova Analytics
Try for free

How AI can tell you why your Amazon sales dropped this week

AI can tell you why your Amazon sales dropped only with live access to your sales, traffic, ads and stock, and Nova's MCP connector gives Claude, ChatGPT and other assistants that access to Seller Central and Amazon Ads. Through Nova, the assistant ranks the products that lost the most sales, then compares their traffic, conversion, price, Featured Offer, promotions, ads and rank with last week, and tells you the likeliest cause to check.

Updated

Connect your AI Copy the playbook
Example with made-up figures
Claude

Why did my sales drop last week?

NUsed Nova

Net sales fell $8,700 (18%). The bamboo cutting board explains 68% of it. Conversion, price and Featured Offer held, but sessions fell by 1,100 and ad clicks by 680, after Tuesday's 30% Sponsored Products budget cut. That is the likeliest cause, not a proven one. Check the campaign budget.

Ask anything

The playbook

Every step calls a Nova tool that reads your live Amazon data, so paste this into Claude, ChatGPT or any assistant connected to Nova's MCP server. Change the period if you need to.

This needs an assistant connected to Nova. Start the 14-day free trial, then connect your assistant.

Use Nova to find out why my Amazon sales dropped.

Period: the last 7 complete days. Baseline: the 7 days before. If I name other dates, use those. Never compare a partial day or week with a complete one.

1. Size the drop. Call Nova's query_metrics with netSales and units for both periods, once in total and once with groupBy "marketplace". Say whether the drop is real and where it sits.

2. Find the products. Call Nova's find_movers on netSales, direction "down", dimension "childAsin", limit 10. Ignore rows with an empty id or __UNALLOCATED__. Say what share of the drop the top products explain. Stop at the products that explain most of it.

3. Split traffic from conversion. For those products, call Nova's query_metrics with groupBy "childAsin" and the metrics sessions, conversionRate, units and netSalesPerUnit. Sessions and conversion do not exist per SKU, so read them per child ASIN. Say whether fewer people came or fewer people bought.

4. Read the levers. Add segmentBy "sku" to get the SKUs. For each SKU that matters, call Nova's explain_movement. It returns units, net sales, net sales per unit, Featured Offer share, customer promotions, ad spend, ACOS, impressions and clicks. It does not return sessions, conversion or rank.

5. Check availability. Call Nova's query_inventory and query_listing_health for those products. Both show the state today, not last week, so a product back in stock can still have sold out during the period.

6. Check rank and demand. Call Nova's query_bsr with the same baseline. Then call Nova's query_sqp with grain "week" for the latest complete week. It runs about 10 days behind, so use it for the market trend, not for this week's dip. Use period.covers as the dates for any comparison with it.

7. Check what changed. Call Nova's list_events for the period and the week before it. If ads moved, call Nova's query_advertising and query_search_terms for those products.

8. Report, one block per product:
   - Sales lost, and its share of the total drop
   - The broken link: traffic, conversion, price, Featured Offer, promotions, ads, stock, listing or demand
   - The figures behind it, this period and the baseline
   - The likeliest cause, your confidence, and what else could explain it
   - The one check in Seller Central that would confirm it
   - The one action to take, or "wait" if the data does not support one

Rules: these are correlations, so never state a cause as proven. Say which figures are missing and why. Relay every notice Nova returns (P&L rules, marketplaces added up, COGS gaps, currency). Do not add Seller Central and Vendor Central figures together.

Raw markdown

How Nova customers use this

Real requests from Nova customers, names removed.

  • “Daily check: the biggest sales decliners this week against last week, and why each one fell: traffic, conversion, Buy Box or ads.”

    What the assistant did

    A recurring daily run. It reads stock, listings and inbound shipments, ranks the decliners and gainers, explains the products that fell, then reads spend and sales per campaign.

    Nova tools used (6)
    1. query_inventory
    2. query_listing_health
    3. query_inbound_shipments
    4. find_movers
    5. explain_movement
    6. query_advertising

  • “Daily briefing: yesterday's sales per client account against the day before, month to date against last month, and which SKU drove each drop.”

    What the assistant did

    A recurring morning briefing across client accounts, with a named SKU next to each drop.

    Nova tools used (3)
    1. list_accounts
    2. query_metrics
    3. find_movers

  • “16-week trend of sales, traffic, conversion, price and ads in one marketplace, to find the main cause of the drop.”

    What the assistant did

    It split the weeks by child ASIN to see which sizes drove the drop, then checked rank and whether a campaign had been paused.

    Nova tools used (3)
    1. query_metrics
    2. query_bsr
    3. query_advertising

  • “Top products by week-over-week net sales decline for the weekly business review, then stock health on those products to see whether out-of-stock is a plausible cause.”

    What the assistant did

    It pulled the week against the week before and the same week last year, named the biggest decliners and checked their stock.

    Nova tools used (4)
    1. list_accounts
    2. list_metrics
    3. query_metrics
    4. query_inventory

  • “Market-wide or our side? Weekly market searches and purchases against our share, after a sales drop.”

    What the assistant did

    It read the daily funnel, then the market against our share, then ruled reviews, listings and stock in or out.

    Nova tools used (5)
    1. query_metrics
    2. query_sqp
    3. query_reviews
    4. query_listing_health
    5. query_inventory

Which products actually lost the sales?

Start with the products, not the account. find_movers ranks your SKUs, child ASINs or parent ASINs by how much they lost against the week before. With direction set to down, the biggest losers come first, including products that fell to zero.

Each row shows this week, last week and the change, and the account total comes back for both weeks. So the assistant can tell you straight away that two or three products explain most of the drop, and leave the rest alone.

Did fewer people come, or did fewer people buy?

Units are sessions times conversion. One of the two moved, and that tells you where to look. The assistant calls query_metrics grouped by child ASIN, which returns sessions and conversion for this week and last.

Amazon reports traffic per ASIN, so asked per SKU these figures come back empty. The playbook compares complete periods only, so a partial week is not read as a drop.

What changed on this product?

For each SKU that matters, explain_movement lines up units, net sales, net sales per unit, Featured Offer share, customer promotions, ad spend, ACOS, impressions and clicks, each against last week. It does not return sessions, conversion or rank, which is why the step before reads those.

Then list_events reads the dated notes on your charts: a price change, a new image, a PPC change, a launch. A drop that starts the day after a change is the first thing to check.

How to read the result

Sessions down, conversion flat

Where it points

Fewer people reached the page: ads, rank or demand

Check next

impressions and clicks, query_bsr, query_sqp

Sessions flat, conversion down

Where it points

People came and did not buy: price, Featured Offer, listing, reviews

Check next

netSalesPerUnit, buyBoxPercentage, query_listing_health, query_reviews

Units flat, net sales down

Where it points

You sold as many for less: price or promotions

Check next

netSalesPerUnit, productPromotions

Units at zero

Where it points

Nothing to buy: stock or a listing problem

Check next

query_inventory, query_listing_health

One marketplace only

Where it points

A local cause, not the product

Check next

query_metrics with groupBy set to marketplace

Most products down by a similar share

Where it points

Demand or the whole account

Check next

query_sqp, list_events

Is it my listing, or the whole market?

query_inventory shows current stock, days of inventory and stockout risk. query_listing_health shows suppressions, listing issues, and who holds the Featured Offer at what price. Both are the state today, with no history.

query_bsr compares Best Seller Rank with last week. query_sqp reads Search Query Performance: the whole market's searches on a query and your share of them. The weekly data runs about 10 days behind and needs Brand Analytics, so it tells you about demand, not about this week's dip.

Example with made-up figures

A drop, worked through

Net sales were $48,200 last week and $39,500 this week, down $8,700 (18%). Three products lost most of it: Bamboo cutting board -$5,922 (68% of the drop), Knife block -$1,700, Serving trays -$600.

Bamboo cutting board

Sessions

Last week

4,100

This week

3,000

Conversion

Last week

12.0%

This week

12.2%

Units

Last week

492

This week

366

Net sales per unit

Last week

$47

This week

$47

Net sales

Last week

$23,124

This week

$17,202

Featured Offer

Last week

98%

This week

97%

Customer promotions

Last week

$310

This week

$300

Ad spend

Last week

$2,100

This week

$1,480

ACOS

Last week

19%

This week

20%

Ad impressions

Last week

210,000

This week

138,000

Ad clicks

Last week

2,100

This week

1,420

Best Seller Rank

Last week

1,850

This week

3,400

One event was logged on the Tuesday: "Cut Sponsored Products budgets by 30%".

Conversion, price and Featured Offer held. Sessions fell by 1,100, and 680 fewer ad clicks line up with most of them. Rank slipped in the same week. The likeliest cause is the budget cut. Lower demand is also possible, and the weekly Search Query Performance data would show it about 10 days later.

Check the campaign's budget in Seller Central. Then restore part of the budget on this product and compare next week.

What this cannot tell you

  • Nova shows what moved at the same time. It does not prove why.
  • Stock and listing health are the state today, not a history.
  • Sessions and conversion exist per child ASIN, not per SKU.
  • Sessions and conversion refresh once a day, so the latest day can be missing while orders are already in.
  • Search Query Performance runs about 10 days behind per week and about 7 weeks behind per month.
  • Nova sees the current Featured Offer owner and price on your listings. It does not see a rival's ad spend or sales.
  • On Vendor Central, Amazon reports no sessions, Featured Offer or promotions. Read vendorShippedRevenue and vendorShippedUnits, and pass channels: ["VENDOR"].

Sessions and conversion refresh daily, orders in about 20 seconds: see refresh times. To log what you changed and see it on your charts, see brand history.

Questions sellers ask

Can ChatGPT or Claude tell me why my Amazon sales dropped without connecting my account?

Only from reports you export and paste. Without a connection the assistant cannot see your numbers, so it can only work from the files you give it. With Nova it reads your Seller Central and Amazon Ads data directly.

What data does the assistant need?

Your Seller Central account for sales, sessions, conversion, Featured Offer and stock, and your Amazon Ads account for spend, impressions and clicks. Search Query Performance needs Brand Analytics data. You connect them once in Nova, then add Nova to your assistant: see the guides for Claude and ChatGPT.

Why are sessions and conversion not shown per SKU?

Amazon reports traffic per ASIN, not per SKU. Ask per SKU and those two figures come back empty, so the playbook reads them per child ASIN.

How recent is the data?

Orders reach Nova in about 20 seconds. Fees refresh hourly, and sessions and conversion once a day, so the latest day can be missing for traffic while orders are already in. Weekly Search Query Performance runs about 10 days behind. Details on refresh times.

Does this change anything on my Amazon account?

No. Every tool on this page only reads. The connection uses OAuth, per account, and you can revoke it at any time. The assistant can log a note on your charts with create_events, and only after showing you a preview and getting your approval.

Does this work for Vendor Central?

Yes, with channels set to VENDOR on find_movers, explain_movement and query_metrics. Amazon does not report sessions, Featured Offer or promotions for Vendor Central, so those come back empty. Read vendorShippedRevenue and vendorShippedUnits instead.

Can I run this every day?

Yes. You can paste the prompt daily, or run it on a schedule if your assistant supports scheduled tasks. Keep the comparison to complete days so a partial day is not read as a drop.

More on the server itself on the Nova MCP server page.

Ask your assistant why sales dropped

Connect your AI