ERP and shop systems diverge: Inventory errors during the Christmas season

In the entire shop, there is only one number that makes a claim about the physical world. The price is a fixed value, the description is text, the image is an image—they may be unwise, but not literally wrong. The inventory, on the other hand, claims that there are currently seven boxes on a shelf in your warehouse. Whether this is true, the shop doesn't know. It believes it because another system told it so—fifteen minutes ago, an hour ago, or at three o'clock tonight.

For eleven months of the year, this naiveté is cheap. If there are forty minutes between two orders, the data becomes outdated faster than it's needed. In December, the situation reverses: the same interface, the same cycle, the same minor booking errors—only now three orders per minute are processed through a data set that's past its prime. The error was always there. The peak season just makes it more expensive.

In our article on shipping costs and delivery times, we concluded that the only delivery time estimate that's still accurate in December is the calculated date—and that this is contingent on reliable inventory. Today, we'll discuss precisely this foundation: why ERP systems and online shops are diverging, what the costs are, and how you can safeguard your inventory before it becomes overcrowded.

To clarify: We've already covered which system is suitable for which shop in our comprehensive comparison , "Connecting ERP, CRM, and Inventory Management Systems to a Shop ," and how such a connection is generally set up is explained in "Connecting an Inventory Management System to a Shop System ." Here, we're talking about something different: operation. About two systems that are correctly connected, running smoothly—and yet still showing different figures in December.

The cost of incorrect inventory — in both directions

The most egregious case is overselling. The shop sells a product that is no longer available, and now begins the most expensive form of customer communication: the apology email. You issue a refund, you offer alternatives, you explain—and the customer who ordered in December had a deadline. A gift that doesn't arrive isn't rebooked; it's purchased elsewhere, along with all of that customer's future orders.

The silent case is more expensive because nobody notices it: underselling. The shop displays "sold out" while forty items remain on the shelf—because a return wasn't processed, because an incoming shipment is stuck in the daily routine, because a safety buffer from last year was forgotten. There's no complaint, no email, no ticket. There are only customers who buy the goods from a competitor, and statistics that record this as "weak demand."

The EHI Retail Institute measures the gap between inventory figures and reality across the industry every year. For 2025, the survey quantifies inventory discrepancies in German retail at €5,11 billion —and the portion that interests us here is not theft: €780 million is attributed to organizational shortcomings, or, in the study's words, "primarily errors in recording, booking, and valuation." The figures are based on data from 103 companies with over 21.000 retail outlets and are predominantly brick-and-mortar stores, as detailed in the EHI press release . However, the underlying mechanism applies wherever records are kept: A significant portion of inventory figures is simply incorrect—not because someone is stealing, but because someone is typing incorrectly, forgetting, or entering data twice.

For online retailers, this means: your shop doesn't sell from your actual warehouse. It sells from a database table that more or less closely resembles your warehouse. How closely it resembles your warehouse determines your inventory for December.

The six ways in which inventory and shop diverge

When we take over a shop with inventory problems, we don't look for "the one problem." It's almost always several of these six issues piling up.

1. The pace: sold on old number

Most connections transfer inventory at intervals—every 15, 30, or 60 minutes, some only at night. Between two runs, the shop sells based on the inventory from the last run. With three orders a day, this window is negligible. With three orders per minute, it's a gaping hole: Anything that flows out through other channels in the meantime still exists for the shop.

2. The race for the last piece

Two customers simultaneously add the last item to their cart—or your own shop and the marketplace sell it twice in the same sync window. The crucial factor is when the reservation occurs: upon order placement or only upon payment or booking. Every minute in between is a period in which the same item can be promised twice. Technicians call this a race condition; in customer service, it's known as a cancellation email.

3. Returns: the goods are there, but the booking is not.

The returned item is physically in the warehouse, but the refund is still pending – waiting for quality control, for the colleague who scanned the receipt, or it ends up in a different storage location unknown to the shop. In January, this is reflected in the accounting; in December, it's lost revenue: precisely the returned items are the ones in demand.

4. Manual correction bypassing the leading system

Someone sees an incorrect figure in the shop and corrects it—in the shop itself. On the next run, the ERP system overwrites the correction, the error reappears, and now nobody trusts any figure. Incidentally, this isn't just a process problem: The German Federal Ministry of Finance's GoBD (Principles for the Proper Keeping and Use of Books, Records and Documents in Electronic Form and for Data Access) explicitly classifies the merchandise management system as an upstream and downstream system of accounting, including its interfaces—and as soon as entries there have a document function, an undocumented change is inadmissible ( GoBD, version of March 11, 2024 , paragraphs 20 and 52). Therefore, silent correction via a database field is also the wrong approach from a tax perspective.

5. Multiple warehouses: not available for sale

Main warehouse, retail space, external storage, consignment stock, the package the photographer still has—as soon as inventory is stored in more than one location, "available" is no longer the same as "sellable." Which locations are included in online availability and which are not is a decision, not a given; the basics are explained in What is Multi-Source Inventory ? It becomes dangerous when no one has made the decision and every system advises differently.

6. Sets, variants, duplicates: two sets, one truth

The gift set maintains its own inventory, as do its individual components—does the set sell what its components allow, or what its own inventory field indicates? The same question arises with variants that have their own item numbers and with duplicate master data, where the same physical item is listed under two numbers. Each of these scenarios works—as long as it's precisely defined who derives their data from whom. As soon as both sides maintain their own figures, one of them is misleading.

Why it always explodes in December

None of these six paths are new, and none are wider in December than in June. Three things do change, however, and all three contribute to the same outcome.

The cycle time eats up the buffer. The risk of a sync window is order rate times window length. If the rate increases tenfold, the risk increases tenfold—with the cycle time remaining unchanged. Goods become scarce. In June, an inventory of eighty units cushions every booking error; in December, when replenishment is slow and sales are fast, you're constantly working with near-zero stock for the most sought-after items—and only there does every incorrect unit hurt. Demand concentrates. Promotional items concentrate a large portion of the volume under just a few item numbers. From our projects: Where things go wrong in December, it almost always involves a handful of promotional items, not across the entire product range—this is experience, not statistics, but it aligns with the simple calculation that errors occur where the activity is.

Added to this is the human factor: seasonal workers in the warehouse record data with the diligence of people in their second week of work. A significant portion of the data entry errors measured by the EHI occur precisely when the most data is being processed.

The marketplace forgives nothing.

In a seller's own shop, overselling costs them a customer. On the marketplace, it also costs them a metric—and that has consequences for all future customers. eBay considers it a transaction defect if a seller cancels a purchase, explicitly also "because the item was no longer in stock." The minimum standard is a maximum of 2 percent of transactions; those who want to maintain "Top Rated" status must not exceed 0,5 percent . According to the seller standards policy , those who fall below the minimum standard must expect lower search engine ranking, reduced sales limits, withheld payments, and a different sales commission. Amazon uses its own cancellation rate metric for seller fulfillment with the same objective.

Consider this in relation to your December figures: With a thousand marketplace orders, two percent equates to twenty cancellations. A single promotional item listed on the marketplace with incorrect stock levels for a weekend can achieve this single-handedly—and thus damage the visibility of all other offers during the most important week of the year. Therefore, maintaining a stock buffer for marketplaces isn't just a precaution, but rather risk management with a clear cost-benefit analysis.

Secure yourself before it gets full: leadership, buffer, pace, report

E-news for online retailers - read your way to wealth Integrating inventory management systems for Magento, WooCommerce and Shopware: SAP Business One, Dynamics 365, NetSuite and Actindo compared

Identify the leading system – for each date, in writing. For inventory management, this is almost always the inventory management system or ERP, not the online shop. "Leading" means that corrections are made only there, logged, and then adopted by all other systems. The rule is simple, yet it's violated daily because the online shop is where the error is first noticed.

Give each channel a safety buffer – differentiated accordingly. The marketplace gets inventory minus X, because every cancellation there costs twice as much. However, a flat X across the entire assortment misses out on sales from slow-moving items and doesn't adequately protect fast-moving ones; a buffer that grows with sales velocity is more sensible. And: Today's buffer will be a liability in January – note down the date it needs to be reduced now, otherwise it will be next spring's undersold stock.

Increase the frequency where the movement is most active. If the system supports event-based data transfer, now is the time to activate it. If not, the entire product range doesn't need to be faster: often it's enough to put promotional and fast-moving items on a shorter frequency and leave the rest running at hourly rates. And check the reservation timing: reservations should be made upon order receipt—not only after payment has been processed.

Measure the difference instead of just believing in it. A daily reconciliation report—shop inventory versus ERP inventory, sorted by variance—costs one developer day and answers the question every morning that the customer would otherwise have to answer. Negative stock levels, items with persistent discrepancies, and transactions without corresponding entries are the three lines that should raise your alarm. In the toolkit of many projects, this report is the cheapest piece of software with the highest return on investment.

And what about inventory? Under commercial law, you're required to take stock of your inventory anyway—but the law explicitly rewards accurate inventory management: According to Section 241 of the German Commercial Code (HGB) , you can forgo a physical inventory count on the balance sheet date if a procedure that complies with the principles of proper accounting ensures that the inventory can be determined in any other way—the perpetual inventory system. This system relies on reliable inventory records: Those who cultivate discrepancies throughout the year will count on December 31st. Those who maintain accurate records count when it's convenient—for example, in the quiet month of October rather than between Christmas and New Year's. Whether and how this applies to your business should be clarified with your tax advisor; we're not the right people to give you that information, but we are the right people to provide the report that makes it possible.

What to do in October — and what's taboo from November onwards

This week: Do the delta measurement. Export your shop inventory, export your ERP inventory, create a spreadsheet of the discrepancies—this takes an hour and replaces any discussion with a list. In addition, physically count your twenty most important seasonal items. If the list and the shelf match, you have a solid foundation. If not, you'll know now—not on December 8th.

In October: Define the leading system and technically or organizationally disable manual corrections in the shop. Set up channel buffers with a decommissioning date. Increase the cycle time for fast-moving items and check the reservation time. Completely review the return process from package receipt to refund and eliminate any bottlenecks. And create the reconciliation report—it will be your early warning system starting in November. Checking that availability is also correct in your feeds should be part of the same process; we discussed potential sources of error there in our article on product data feeds before the season.

From November onwards: no new interfaces, no major system changes. From now on, only values ​​will be adjusted—buffers, cycle times, thresholds. Check the delta list daily, and if an item is out of sync: take it offline, resolve the issue, and reactivate it. An item that's unsellable for two hours costs two hours of revenue. An item that's incorrectly sold twenty times costs you the entire month of December. If the integration itself is the problem, that's a project for the fall—we handle that kind of thing under ERP integration for Magento, WooCommerce, and Shopware , and like all system changes, it should be completed before November.

Conclusion: The inventory is an assertion — check it before December does.

Inventory errors aren't operational accidents, but rather inherent system characteristics: sync windows, races to the last item, pending refunds, manual corrections, unclear storage locations, and duplicated figures are present in almost every interface. For eleven months of the year, these issues remain invisible because buffers and a lack of activity mask the gaps. December eliminates both—and then your reputation isn't determined by the interface, but by whether someone measured the discrepancy before the customer discovered it.

The good news: None of this requires a new system. Defining a leading system, setting up channel buffers with a decommissioning date, short cycle times for fast-moving items, a daily reconciliation report – all of this adds up to less effort than dealing with the wave of cancellations that it prevents.

What is your delta?

Run the measurement and let me know the results: Shop export versus ERP export, one hour of work. My guess: There are more items with discrepancies than you expected—and at least one of them is on your action list for December.

And the question I'm most interested in hearing about your experience with: Do you synchronize at intervals or based on events—and what were the results of the switch? Anyone who's already made the change knows the costs and the effects. Nobody else seems to be writing about this; I'd love to read a report on your experience.

If you would like to have your inventory management reviewed by an outside expert before the season, you can contact us via the contact page.

This post is part of our Q4 series. Every Tuesday, a new post will be published here to help you prepare for the peak season. Next week: Christmas landing pages that rank — SEO timing for seasonal pages.