Merchandising and boosting

Relevance answers which products match. Merchandising answers which of them you would rather sell. This article is about the second question, and about the two lines you must not cross while answering it: the buyer's article number, and the buyer's contract.

The rule that comes before every other rule

An exact identifier match is never outranked. Not by margin, not by stock, not by a campaign, not by your own brand. If the buyer typed a code and a code matches, that article is result one.

Everything below applies to the descriptive queries, where several articles answer the buyer and something has to choose the order.

The B2B rule: never boost what this buyer cannot buy

A contract customer sees a catalog shaped by their Rahmenvertrag: a price list scoped to their organization, often a punchout channel with a closed assortment. A boost written for the open web shop ("push the house brand") applies to that buyer too unless something stops it. The buyer clicks the promoted article and finds no contract price, or the list price they negotiated away, or a product their procurement system rejects. Each of these reads as a broken shop, and the buyer who hits it is your most valuable one.

So the rule: a boost may reorder the products a buyer is entitled to; it may never surface one they are not. In the planned model that means every boost is scoped to a channel and evaluated inside the buyer's price list, never on the full catalog. Today it means keeping punchout channels on Hidden until assigned, so nothing outside the negotiated assortment is in the result set to begin with.

What to boost, and by how much

Boosts are conditions with a weight, applied after text relevance. This is the planned model. Four rows are worth having in almost every B2B catalog:

And two that sound sensible and mostly are not:

• Margin. Tempting, and it degrades trust quickly. A buyer who works out that your search sorts by your margin starts checking prices elsewhere. If you use it at all, use it as a tie-breaker between otherwise equal results, never as a strong signal.
• Newest first. Recency is a B2C signal. In B2B the article a buyer wants is usually the one they bought last time, which is by definition not the newest.

Category-specific rules

A single global boost list is as wrong as a single global facet list. What belongs at the top of Fasteners is not what belongs at the top of Pumps.

Plan boosts per category branch where it matters:

• Fasteners and C-parts. Boost the standard sizes and the packaging units people order. Bury single-piece listings if you sell in boxes of 100.
• Spare parts. Boost original parts over equivalents when the buyer arrived from a machine page; the opposite when they searched by description and price decides.
• Consumables. Boost the size a buyer reorders, not the trial size.
• Long-tail technical. Boost nothing. Let relevance work; there are rarely two candidates.

Campaign boosts

A campaign boost is a boost with a validity window. The window belongs on the boost itself, set when you create it, not in someone's calendar. That is the difference between a campaign that ends and a campaign that quietly runs until someone notices in March.

Campaign boosts belong to a channel. A promotion aimed at your web shop should not reorder the results inside a punchout catalog, where the buyer is inside their own procurement system and expects the negotiated assortment in a stable order. See Punchout explained.

How this will work

The order of decisions matters more than the screen. When the Search app ships, work through it in this order:

1. Write down the boosts you want, each as one sentence with a reason. If you cannot write the reason, do not create the boost.
2. Start with in stock alone, for one channel, and measure for two weeks before adding anything else.
3. Give every commercial condition the stock guard and the lifecycle guard.
4. Scope every boost to a channel, and confirm it cannot surface a product outside the buyer's price list.
5. Add category-specific boosts only where the global set is visibly wrong.
6. Give campaign boosts a start and end date at creation.

The planned fields per boost: the condition, a strength, a channel, an optional category branch, and a validity window. Pins (a fixed product at position one for a named query) are a separate, smaller list and should stay small.

What you can enforce today

Two of the rules above are data decisions, and the data exists:

• A discontinued article cannot rank first if it cannot rank at all. Disable it in PIM › Products (enabled off) as part of your lifecycle routine instead of relying on a future boost guard.
• A contract buyer cannot see what is not in the channel. Set the punchout channel to Hidden until assigned under Configuration › Channels (unassigned_visibility = assigned_only) and assign the negotiated assortment explicitly. See Channels.

Neither replaces merchandising, but both shrink the rule set you will need.

What to check

Before and after every change, run the same three tests:

1. The identity test. Ten real article numbers. Each must still be result one. If a boost broke this, remove the boost.
2. The Innendienst test. Five descriptive queries. Show the first three results to someone who answers the phone all day and ask whether they would have named those articles. This finds more problems than any metric.
3. The number. Compare the search-to-order rate for the affected channel over the two weeks after against the two weeks before; see Search analytics for the definition and for where the number comes from today. If it did not move, the boost was noise. Remove it and keep the configuration small.

Keeping it from rotting

Boosts and pins accumulate. Every quarter:

• Delete campaign rules whose window has passed and which nobody renewed.
• Re-check that every pinned product still exists, is enabled, and is in stock.
• Count your rules. More than a few dozen means ranking is being patched query by query, and the fix belongs in the index configuration.

Next

• Working the zero-results report
• Search analytics