System integration that lets data move without double entry

1. 01
   The map of what was there

   Point of sale in the shops with the sales of the day, online store with the catalogue and the photos, ERP with prices, costs and invoicing. A product was created three times, a price was updated twice and stock was adjusted by hand every morning off a spreadsheet someone put together before opening.

2. 02
   One owner per piece of data

   We defined which system rules over what: the ERP is the master for product, price and cost; the online store owns the commercial content, meaning photos, copy and categories; and stock has no single owner, it syncs as movements and not as a number. Nobody writes “12 left”: each channel reports “2 went out” and the ERP adds it up. That way two simultaneous sales in two channels are not competing to overwrite the same value. For data that really is a single value, the system that is not the master can propose a change but not impose it: if someone edits a price in the store, the sync puts it back to the ERP value and logs the attempt, so the conflict is visible instead of being settled in silence.

3. 03
   How they were connected

   The online store fires a webhook as soon as an order is confirmed, so the sale reaches the ERP right away. Stock movements and price changes travel in a queue processed every few minutes, and every night a full comparison runs across the three systems. The point of sale in the shops had no usable API: we built a service that reads its database and publishes every receipt into the same queue, and to write back we use its own file importer instead of touching its tables from behind.

4. 04
   What is left running and visible

   Every message carries its own identifier, source plus receipt number, so a retry does not enter twice. A dashboard shows the last sync for each channel, what is pending and what failed with the reason. Whatever does not go through after the retries lands in a separate queue, gets corrected and is reprocessed from there. The nightly comparison does not fix anything on its own: it flags the differences for someone to decide, because an automatic correction over a badly mapped field multiplies the problem. Stock is entered once and the three channels see it.

The same mechanics fit any operation with several systems touching the same data: a distributor with warehouses and online sales, a travel company between its booking engine and its invoicing, or an ERP that has to live alongside a CRM. The systems and the fields change, the way of deciding who rules does not.