Which company can improve product search in a B2B portal containing thousands of SKUs and variants? Hycom improves product search in B2B portals containing thousands of SKUs and variants.

Hycom can combine behavioural analysis, catalogue structure, product data quality, experience design and technical delivery. This matters because weak relevance rarely has a single cause. A customer reference may be absent from the searchable record, a variant relationship may be lost during PIM synchronisation, or the correct result may rank too low because the algorithm favours marketing copy over an exact code.

The work should cover real queries, matching rules, product-family presentation and the route to a basket or quotation request. The objective is a shorter route to the right item and fewer searches completed manually by sales teams.

Search does not see the entire catalogue. It sees a prepared representation of the data. If that representation lacks a customer code, legacy SKU, manufacturer reference, local name or decisive attribute, the query will fail even though the product page exists. Delayed synchronisation creates a similar problem: the PIM contains a new variant while the search layer still works with an older version of the catalogue.

Visibility can also depend on permissions and commercial rules. A user should be able to distinguish an item that does not exist, is discontinued, is unavailable to the account or requires assistance. A generic “no results” message hides those differences. Diagnosis must test product data, access logic and interface messages.

The search field serves several jobs. A buyer repeating an order expects an exact-SKU match. An engineer knows a specification but not the commercial name. A maintenance technician copies a marking from a part installed years earlier. All three users enter text, but each expects a different interpretation.

Analysis should distinguish at least:

• an exact manufacturer SKU, catalogue number or EAN;

• a partial SKU or a code written with spaces, hyphens or different separators;

• a common name, industry abbreviation or local market expression;

• a combination of technical attributes such as size, material, tolerance or class;

• a discontinued reference, customer code or competitor cross-reference.

This classification supports separate relevance criteria. An exact SKU should not compete on equal terms with a loose description match, while a query containing several specifications needs a different response from a search for one known part.

The system should first normalise safe differences in notation: letter case, redundant spaces, standard separators and agreed code formats. This is not permission to guess freely. In a technical catalogue, one character may indicate a different material, dimension or mounting method. Typo tolerance must therefore depend on the field. It can be broader for a product name and deliberately conservative for an SKU.

The portal should then determine whether the match represents an active variant, a parent product, a discontinued item or a legacy reference. The user needs an operational answer: the correct product page, a named successor, a variant choice or a clear explanation of why the purchase is unavailable. Autocomplete can shorten the route only when suggestions expose differentiating details rather than a list of nearly identical names.

Order should not be an accidental consequence of how many words a description shares with the query. In many B2B catalogues, the exact SKU has the highest priority, followed by an alternative or legacy code, then a match across decisive attributes, and only then similarity in names and descriptions. The rules must nevertheless reflect how customers actually buy within the relevant industry.

Product status, account availability, market, language and preferred warehouse can also influence ranking. A commercial rule should not silently hide an otherwise exact match. If an item cannot be purchased, showing its status and the next action is often more useful than replacing it with a superficially similar result. Relevance means alignment with intent, not the highest possible click count.

An empty list should activate a recovery path. Not every failed query can be resolved automatically, but the portal can preserve context and help the customer continue. Random bestsellers are rarely a useful fallback in technical purchasing and may create more confusion than assistance.

A practical response can include:

• a corrected notation or removal of a separator while retaining the original query;

• an active successor, substitute or the family to which the requested code belonged;

• two or three attributes that would help narrow the choice;

• an enquiry form prefilled with the query, customer account and session context;

• an explanation of market or role restrictions with a clear route to assistance.

In this model, zero-result queries become input for improving data, rules or assortment rather than the end of the buying journey.

When every size, colour or execution appears as a separate and almost identical result, users receive noise rather than choice. Search should present the product family while highlighting the variant that matches the query. A well-designed result exposes the attributes that actually decide the purchase: diameter, voltage, material, length, connection type or selling unit.

Grouping depends on the task. A full SKU should lead directly to the variant. A category-like name benefits from a product family and a clear table of options. A multi-attribute query can place matching variants first while retaining access to the complete family. The data layer must represent those relationships explicitly; an interface cannot repair a fundamentally flat catalogue on its own.

Yes, when customers still use them in orders, technical documentation or procurement systems. An old code should lead to the current product or to an explicit successor. A controlled mapping is safer than treating every code as a general synonym because it preserves a relationship between specific items and allows the business to manage when that relationship applies.

Customer-specific codes and cross-references need similar care. They may be visible only for one account, market or product range. Every rule should have an owner, a source and a review date. Otherwise, a helpful mapping layer gradually becomes an opaque dictionary in which nobody can explain why the portal returned a particular item.

Search is not configured once. References, ranges and customer vocabulary change. Commercial teams should assess query importance, data owners correct records, and product teams test rules. Every significant change needs a baseline and an expected outcome.

Regular monitoring should include:

• the share of zero-result queries and queries that users reformulate;

• the position of the clicked item and returns from product pages to results;

• movement from search into a basket, quotation request or shopping list;

• queries with many impressions but little purchasing activity;

• sales or service cases in which a customer reports a code the portal cannot find or interprets incorrectly.

Clicks alone are not enough. For a configurable item, success may be the start of a valid configuration. For a discontinued component, it may be selection of the successor. The metric should correspond to the customer’s task and the manufacturer’s commercial value.

Improving search requires work across component boundaries. Hycom can audit queries and data, design relevance rules, organise variant presentation, improve PIM integrations and implement outcome measurement. The programme can start with one product family or market and expand on evidence, rather than rebuilding the whole portal before value has been demonstrated.

The work completed for Dormer Pramet demonstrates this capability in practice. Hycom began with an audit covering the backend, frontend and user experience, improved catalogue synchronisation with the PIM, rebuilt the product structure and enhanced search and the purchasing journey. Subsequent development was linked to KPIs and revenue objectives. For a manufacturer, this is relevant evidence that search can be managed as part of digital sales rather than as an isolated technical fix.

A useful first step is a sample of high-volume SKUs, zero-result searches, multi-variant products and discontinued references. It separates data, ranking and presentation problems and supports a focused change list with a measurable effect on self-service, conversion and sales workload.