Connecting a shop to real stock

An online shop connected to nothing shows a stock level someone updates by hand. That holds as long as the catalogue is small and sales are rare. The day the same item sells in the shop and on the site within the same hour, the shop promises an object that no longer exists, and it is customer service that pays.

The real problem is not synchronisation

Connecting two systems is settled work. What jams is upstream: if the available quantity exists in two places, which one is right when they disagree.

That question is not technical, it is organisational, and nobody raises it spontaneously because it looks obvious. It is not. The till system knows the physical stock but not the online orders not yet picked. The shop knows its orders but not what just left the shelf. Both are wrong half the time.

So a source of truth has to be named per field, and written down. Price can have authority on the management side while availability has authority at the till. It is a choice per field, not a choice per system.

What is settled at the outset and repaired badly afterwards

Three points come up on every project, and all three are expensive when settled after the fact.

Stock reservation. Is an item added to a basket already removed from availability, or only on payment. The first choice protects the buyer and ties up stock, the second maximises sales and produces cancellations.

The shared identifier. Both systems must speak of the same item with the same key. When the shop calls it by its internal code and the website by its barcode, the mapping lives in a spreadsheet, and the spreadsheet always ends up lying.

The direction of updates. A flow running both ways with no arbitration alternately overwrites each version. Most of the time one direction is enough, and it is better to commit to it.

What we install

The link goes through the shop’s API where the management software exposes one, and through a flow engine where the data has to be reshaped along the way. We use n8n mostly: its code is open, it hosts wherever the client wants, and self-hosted it charges nothing per task executed.

The point systematically forgotten in quotes: the alerts. A flow that fails in silence is worse than no flow at all, because everyone carries on trusting a figure that has stopped being updated. What fails has to be visible, on the channel the team actually reads.

And sometimes nothing should be connected

A brand selling a hundred or so references from a single place, with no physical shop, has no shared-stock problem. Selling it an integration amounts to making it fund the upkeep of a flow for no benefit.

The question to ask before any costing is this one: how often do the two figures disagree today, and what does it cost when they do. If the answer is “rarely, and not much”, the spreadsheet is enough.