Merchandising controls for search and discovery

Merchandising control works best when each commercial change has a defined scope, an owner and a way to test its effect. Use boosts, exclusions and ranking objectives for the jobs they suit, preserve mandatory shopper requirements, and review the whole configuration before increasing automation across search, listings or recommendations.

Give every change a clear purpose

Start with the trading decision: introduce a relevant new range, fulfil a defined promotion or change the ordering of suitable products. Name the intended audience, surface and duration. State which products must remain excluded and which shopper requirements must not be relaxed.

Avoid turning an uncertain goal into a permanent global boost. A request to support a seasonal range needs an explicit scope and review date, not an unexplained rule that stays active after the campaign ends.

Separate requirements from preferences

Some controls determine whether a product can appear; others influence the order of already suitable candidates. Make that distinction visible in the operating plan. Required compatibility, availability or selected attributes should not be silently traded away for a commercial preference.

Ask the implementation team to demonstrate the precedence of relevant controls, including what happens when rules conflict or the requested product is unavailable. Confirm which review, scheduling and approval capabilities exist in your configured product; a desired governance workflow is not automatically a shipped feature.

Preview the full configured experience

Use a representative query and category panel before and after a change. Inspect exact product requests, broad discovery queries, new products and low-stock items. Keep a saved configuration so the team can identify which rule or objective caused a difference.

If several changes are released together, record that bundle. Otherwise an observed gain or regression may be attributed to the wrong intervention. Include the front-end result and destination page in review, not just the rank number in a dashboard.

Measure the outcome you intended

Choose a commercial outcome and relevance guardrails before release. A higher promoted-product click rate can coexist with a poorer overall journey, so inspect effects beyond the promoted item. PA’s Merchandise Analytics describes analysis of boost and recommendation tactics, including product and basket measures.

Use controlled experiments where appropriate and label observational comparisons honestly. Record why a rule remains active, when it should be reconsidered and who owns the decision.

Build automation around an agreed decision

Review recurring manual work to identify which decisions can be expressed as a stable objective and which need explicit exceptions. PA’s Margin Management provides the product context for value objectives within eligibility and relevance boundaries.

Bring a representative rule inventory and a difficult promotion conflict to a demonstration. Evaluate the controls your team would actually use, then expand automation when their behaviour and measurement are understood.

Choose the control that fits the decision

A rule can express a narrow requirement, such as excluding an unavailable item or applying a time-limited promotion. An objective can express a continuing preference among eligible products, such as improving an agreed value measure. Use explicit exceptions when the objective cannot represent a genuine business requirement.

Consider a fictional promotion for a new shoe range. Keep required size and availability checks intact. Define the campaign scope and end date, then compare a targeted boost with the baseline configuration. Do not assume that promoting the range globally is the only way to meet the goal.

Audit inherited rules when retrieval or ranking changes. PA’s grocery case describes a configuration where a legacy rule layer degraded the evaluated model’s ranking quality. That is a reason to test interactions, not evidence that all rules are harmful.

For each rule, decide whether to retain it, narrow its scope, replace it with a tested objective or retire it. Keep the decision and its evidence so the next trading team can understand why the configuration exists.

Frequently asked questions

Can boosts override every shopper requirement?

They should not be used to hide a mismatch with a mandatory requirement. Define eligibility and relevance boundaries, then demonstrate rule precedence and behaviour in the actual configuration.

Should every rule be removed when AI ranking is introduced?

No. Some rules represent real policy or commercial requirements. Review each rule’s purpose and effect, then retain, narrow, replace or retire it based on evidence.

What governance controls should we ask to see?

Ask for the scope, preview, conflict handling, change history, scheduling and reversal workflow relevant to your team. Confirm which controls are available and which need an agreed operating process or additional implementation.

Are objectives better than rules in every situation?

No. A hard requirement or specific exception may need a rule, while a broad preference may be better expressed as an objective. Judge each against its intended outcome and constraint behaviour.

How can we tell whether a rule is harmful?

Compare the configured experience with and without the rule on representative queries and relevant live outcomes. Inspect the affected products and query groups; an overall average may hide important failures.

What should happen when a rule and an objective conflict?

Define precedence before release and test a concrete conflict. Preserve mandatory requirements, make the applied decision inspectable and give an owner responsibility for resolving ambiguous cases.