Thirty questions you can ask your Linnworks order data once an AI can read it

These are not demo questions. They are the ones that come up in a working week and currently cost somebody twenty minutes each.

Open orders and despatch

What is in open orders right now, oldest first? Which open orders are more than three days old? Which open orders are waiting on stock we do not have? How many open orders are for trade accounts rather than marketplaces? Is anything in open orders flagged as paid but unpicked? What is the total value sitting unshipped?

Sales and trends

How many orders did we process last week compared with the week before? What were our top twenty SKUs by units last month? Which channel grew fastest this quarter? What is our average order value this month by channel? Did that line sell better after we changed the price? Which days of the week are busiest for us, really?

Stock

Which SKUs are below their reorder point right now? What is our total stock value? Which items have not sold at all in ninety days? Which SKUs have stock in one location and none in another? Are there items with stock on hand but no sales history — the quiet dead money? What did we archive this year and should any of it come back?

Buying and suppliers

What purchase orders are open and when are they due? Which supplier is latest against their quoted lead time? What is on order for the SKUs currently below reorder point? How much have we committed in open purchase orders?

Customers

Who spent the most with us this quarter? Which trade accounts have not ordered in sixty days? Which customers order most frequently but at the lowest value? Find the customer whose name I only half remember and show me their last three orders.

Returns and refunds

What did we refund last month, and for which reasons? Which SKUs are returned most often as a proportion of units sold? Are there refunds still awaiting action? Did returns on that line drop after we changed the supplier?

Why the follow-up matters more than the question

Every one of the above is answerable with a report, given enough patience. What is not answerable with a report is the second question — the “of those, which…” that occurs to you the moment you read the first answer. That is the whole argument for conversation over exports: the interesting question is almost never the one you set out to ask.

The honest caveats

Answers are only as good as the data. If cost prices are stale, margin answers are fiction. If the same product exists under three SKUs, counts will be wrong in a way that looks right. And for anything that will be quoted to a customer, an accountant or a bank, check the figure in Linnworks before you send it — not because the connector guesses, but because you should be able to point at the source.

Also: ask one thing at a time. A question with four clauses in it gets a confident answer to roughly two of them. Short questions, then follow-ups.

All thirty of those are read operations, which is why a Linnworks MCP connector you host yourself is worth deploying without its four write tools to begin with. Related reading: ten questions an ops team can ask Claude.

Questions, or want a tool we don't have yet? Email hello@grafto.co.uk — a real person replies.