Highlighting a pack or bundle without burying search

The symptom, before the cause

On a professional hardware catalog, a search for "18V drill" consistently surfaced 3-to-8-tool packs ahead of any standalone drill — sometimes with no single product appearing at all in the first fifteen results. Nothing technically broken: every pack genuinely contained an 18V drill, the engine was right to find them.

The real question was elsewhere: does someone typing "18V drill" actually want, as a first choice, a several-hundred-euro investment in a full kit?

Why the text wins, not the product

A listing describing four tools mechanically accumulates more signal than a listing describing one. Every mention of "drill driver," every "18V-XXX" reference repeated across a long marketing description weighs into the score — the right behavior elsewhere (a word repeated several times in a listing is a legitimate relevance signal), but one that mechanically favors the more verbose content, not the content most relevant to a precise query intent.

A pack listing isn't a product gaming the system. It's simply longer, by nature.

The wrong answer, and why we didn't take it

The first temptation is to fix this inside the score itself — penalize long listings, or categories containing "pack." Two problems. First, a pack ISN'T less relevant: it genuinely contains what's being searched for, often more. Demoting it would hide a legitimate option, sometimes the better deal for the buyer. Second, adjusting a relevance score always has effects that reach beyond the one case you're trying to fix — a tweak designed for one product type ripples into every other one, rarely in a way you can fully predict in a single pass.

The answer we shipped: a separate field, not a trade-off

Rather than choosing "pack or single product first" inside the ranking itself, the search response now carries a separate field, highlighted_bundle: the best-matching pack for the query, distinct from hits.

{"query": "18V drill", "total": 47, "hits": [...],
 "highlighted_bundle": {
   "product": {"name": "4-Tool Pack 18V", "category": "Pack 4 tools", "price": 1200},
   "score": 35.66,
   "in_stock": true
 }}

hits doesn't change — it's still the natural ranking, standalone products included. It's up to your interface to decide how to present the two zones: a visually distinct highlight above the list, an endcap rather than a spot forced into the regular ranking.

One limit, to be honest

Detection runs on the category field — a product counts as a pack or bundle if its category contains "pack," "kit," or "bundle" (case-insensitive). No extra data to supply: it works with a catalog you've already indexed, as is. The trade-off, knowingly accepted: it depends on the vocabulary already present in your categories, not an explicit declaration. A brand might just as well write "Lot" or "Combo" without today's detection recognizing it — the three words chosen cover what we've observed on real catalogs, not an exhaustive list of every possible term in the industry.

The question was never "do packs deserve to be seen" — they do, a higher average order value serves the distributor as much as the buyer saving money by purchasing together. The question was how to show them without their natural length accidentally burying what a visitor in a hurry was precisely looking for.

Available now

The highlighted_bundle field is active on every search via POST /v1/index/{catalog}/search, with no configuration or migration — absent from the response if no product matching the query has "pack," "kit," or "bundle" in its category, present otherwise. It also respects stock: an out-of-stock pack is never highlighted, even if it's the best match — recommending something that can't be bought helps no one. Full details in the documentation.