Remains on the site that match the composition

By reversibility, and not by the principle "the more often, the better." The slow assortment lives on several runs per day. Peak season requires an hour or half an hour: the difference here is measured by canceled orders. The benchmark is simple - how many sales do you have between two runs. And the fair limit of the mechanism itself: the window between runs always remains, so the last unit can theoretically be bought twice. Exchange narrows this window, but does not close it - it closes its reservation at the time of registration.

It will work if you do a full run every time. Therefore, on large catalogs, the exchange takes only changed items: accounting gives a list of what has moved since the last time, or a time stamp of the change. For reference: in the garden equipment store we measured, there were 55,321 product pages with a first server response of 840.5 ms — at such a volume, a full update every half hour simply will not have time to pass. The second fuse is the schedule: we do not set heavy runs during the hours of peak sales.

This is the most common reason why the exchange "seems to work" and half the directory is unchanged. There were two exits. Compiling codes once is content work, it counts separately, and all other integrations benefit from it, from feed to platform to billing. Or keep an intermediate table of correspondences, which is replenished as discrepancies are found. Our recommendation: start with a table to make the exchange work, but put code consolidation in the plan - code divergence hurts more than just balances.

It's your decision and it's different for different groups. Hiding a position is easy, but you lose an address that was already bringing conversions. Show as absent is better for search, but the buyer sees something that cannot be bought. Allowing pre-orders is best if delivery times are predictable. We establish a rule separately by category, and not one for the entire catalog: the answers here are different for running trifles and for custom-made equipment.

The exchange should survive this without consequences for the store. When the source is unavailable, the last known balances are left in place, not reset: otherwise, an hour of idle time in accounting turns the entire catalog into "out of stock", and you'll see it already when orders fail. A notification is sent to the messenger about the failure itself - you know that the data is out of date, and you know from what time. Therefore, in the log, we record not only errors, but also the time of the last successful run.

No, and it would be dishonest to promise such a thing. The exchange does not correct the account, it quickly and accurately transfers its status to the site. Reassortment, non-executed write-off, goods that physically left without a document — all this will come to the catalog along with the rest. Therefore, at the audit, we ask not only about the format of the data, but also about the extent to which the balances in the accounting coincide with the composition. When the discrepancy is large, it's fairer to deal with it first, otherwise you're automating the spread of inaccuracy.

Yes, and this is one of the main arguments for the exchange. Maidanchyk and Merchant Center do not read your page, but the feed: name, price, availability, picture. If the feed is collected from the same source as the directory, the stale residue from there disappears with the directory. Otherwise, you're paying for a click that leads to a product that doesn't exist. The limit is immediate: the feed does not make your price more profitable - in the commodity issue, the buyer sees you next to everyone, and he will also see the difference in price faster.

Yes, this is the second direction, and it is usually made a separate stage after the residuals. The order from the site is registered with a document, the manager works in the same system instead of rewriting by hand. The difficulty is not in the transfer, but in the coordination of directories: counterparties, warehouses, types of payment must match on both sides. Our recommendation is not to enable both threads at the same time, otherwise mapping errors and transfer errors are resolved in one week and not separated.

12–35 working days. It starts with an audit of the exchange: 3-5 days, free of charge, at the output is a diagram and a list of items that do not match by code. Then there are six stages, including a week of observation mode - the exchange already works on combat, but only counts discrepancies. The lower term limit is a one-way exchange of balances with an account that can provide data. The top is filled with prices at several levels, warehouses at points and transfer of orders back: the latter goes in a separate flow on top of these stages.