Customer Part Numbers Searchable in the B2B Shop

In business-to-business trade an article rarely has a single identity. The supplier keeps its own article number because stock, pricing and picking all depend on it. The manufacturer assigns a type number, usually complemented by a European Article Number. And the buyer keeps a third number, because their ERP knows the item under a key that fits their bills of material, cost centres and purchase requisitions. That third number is the customer part number. It is the only one of the three a buyer can read from memory or from their own screen without looking anything up, which is precisely why it is the one they type into the search box.

The BMEcat catalogue standard has modelled this three-way split for a long time. The supplier's article number is a mandatory element with 32 characters (BME e.V.) and, according to the specification, is the one that governs ordering. The buying company's number has more room than the supplier's own, at 50 characters (BME e.V.), and may be repeated. Even the product group from the customer's ERP has its own element, albeit with only 10 characters (BME e.V.). A shop that does not rebuild this structure squeezes three number worlds into one field and loses the mapping as soon as two customers name the same article differently. How these fields travel from inventory management into the shop decides the later quality of search.

Internal search is a measured weak spot in commerce. More than half of the benchmarked online shops do not adequately support users' search needs: 56% (Baymard Institute) fail in a benchmark covering more than 170 reviewed shops. Even the simplest form, typing an exact product name or model number, causes problems on 12% (Baymard Institute) of shops. And as soon as abbreviations or symbols appear in the query, which is exactly what every article number brings with it, more than every second shop struggles, at 54% (Baymard Institute).

In business-to-business trade a fourth difficulty is added that consumer commerce does not have: the search term is not in the catalogue at all. K-88 4711/2 is neither a typo nor an unusual expression; it is a string that is simply absent from the supplier's data as long as nobody has stored it there. No synonym dictionary and no typo tolerance finds a value that is missing from the index. The task is therefore not a search task but a data task, and it is solved in the product data flow from PIM and ERP before the search engine is involved at all.

A customer part number is a statement about the customer, not about the article. It reveals how a company structures its range, which plants it supplies and occasionally which project a part is ordered for. It therefore belongs behind the login and inside the same permission scope as prices and assortments. Anyone who has already solved customer assortment release cleanly attaches the numbers to that same check instead of placing a second mechanism next to it that can later drift on its own.

Three edge cases are worth deciding up front. First, the group with several purchasing organisations: the same number can stand for two different articles in two plants, in which case the type has to carry the resolution and the result list has to show both candidates. Second, the sales agent who orders on behalf of several customers and switches context per session; here visibility hangs on the selected customer, not on the user account. Third, the guest without a login, who must not see foreign numbers at all because there is no customer to attribute them to. All three are permission questions rather than search questions, and they are answered in the query layer.

The benefit does not end at the result list. The foreign number has to travel through the entire document chain, because the buyer wants to find it again on every document. The European electronic invoice provides two separate, standardised elements for exactly that: BT-155 (OpenPeppol) carries the identifier assigned by the seller, BT-156 (OpenPeppol) the one assigned by the buyer. Both sit in the invoice line and may occur once per line. Keeping the number in the shop but not passing it into outbound invoicing gives away precisely the place where it saves the most manual effort. The same pass-through applies to every document field: how vouchers and item lines are handed over to accounting follows the same mapping logic.

The time pressure comes from the switch to structured invoices. Paper and simple image files are a transitional arrangement in business-to-business trade, and the transition period is staggered: for supplies carried out before 1 January 2027, every company may keep invoicing on paper until 31 December 2026 (Bundesministerium der Justiz) or, with the recipient’s consent, in an electronic format that does not meet the standard; no turnover threshold applies. Only for supplies carried out after 31 December 2026 is that route tied to a total turnover of no more than 800 000 euros (Bundesministerium der Justiz) in the preceding calendar year – and for invoices transmitted by electronic data interchange under Recommendation 94/820/EC, a separate exception without a turnover threshold applies for the same period (Bundesministerium der Justiz). Issuing is time-bound as well: the invoice has to be issued within six months (Bundesministerium der Justiz) of the service being performed. Companies planning the switch anyway should anchor the foreign number in the same step; how that interacts with the e-invoicing mandate is a construction stage of its own.

Numbers arriving by machine are the norm in business-to-business trade rather than the exception. Almost half of industrial companies with electronic sales receive orders via EDI: 46.04% (Eurostat) in reference year 2024, surveyed among enterprises with ten or more employees. In wholesale and retail, 10.51% (Eurostat) of total turnover comes from EDI sales. Across all sectors more turnover even arrives through EDI than through websites and apps: 11.07% (Eurostat) against 8.39%. The shop therefore stands next to a lane that has known the customer's number for years, and it should reuse the same mapping logic instead of building a second one beside it. How tightly the two routes are coupled depends on the EDI integration via EDIFACT.

A side aspect with consequences: a shop's internal search is mostly invisible to machines. Only 28% (Web Almanac) of the home pages examined describe their own search function in a machine-readable way, measured in the July 2025 crawl across more than 16 million websites. Structured data of any kind is served by 50% (Web Almanac) of home pages. For customer part numbers this is explicitly not an invitation to publish: foreign numbers belong behind the login. Machine-readable description concerns public search and public product identifiers, not the customer-bound ones. Drawing that line cleanly is part of any interface development that keeps both inside one system.

In practice that means the public product page carries manufacturer and EAN numbers, the customer-bound view additionally shows the signed-in customer's own number, and the delivery path for search engines does not see the second layer. This can be verified with a query that runs against the search endpoint without a login and whose hit count has to stay at zero. That check belongs in acceptance testing rather than in hope, because the fault is barely visible from outside: it surfaces only when a customer finds their own number in a competitor's index.