Watch a maintenance engineer use an industrial catalog and you learn something fast: nobody starts with a product code. They start with the problem. "Leak." "Rust." "Worn pump housing." The product code is what they leave with, not what they arrive with.
We have built site search for several industrial catalogs now, including Belzona Baton Rouge, IMS, and Polymer Nation, and this insight drives the whole design. Search is not a text-matching feature. It is the digital version of the counter person who hears "we have a leak on a cooling line" and walks you to the right shelf.
Hand-built synonym maps, not machine guesses
On Belzona Baton Rouge and IMS, the search index carries a hand-built synonym and intent map. On the Baton Rouge build it holds 18 problem-language triggers: typing "leak" expands into leak sealing, pipe repair, and emergency containment; "rust" lands on the corrosion products. The buyer's vocabulary is mapped to the catalog's vocabulary before ranking even starts.
These maps are written by hand, from the client's own material and the way their customers actually talk on the phone. No query-log harvesting, no black box. A person read the catalog and wrote down what a plant engineer says when a coupling starts weeping on a Friday night. It is unglamorous, and it is why the first result is so often the right one.
Prebuilt indexes that cost non-searchers nothing
There is no search server behind any of this. The index is a static file generated at build time: 283 records in about 118 KB on Belzona Baton Rouge, about 137 KB on IMS, 164 records on AAS. It covers products, applications, industries, case studies, and certifications in one place.
Neither the index nor the search library loads with the page. Both are fetched the first time a visitor actually opens search. Until that moment the feature costs zero bytes, which keeps product pages fast for the majority who never search at all. And when search does open, it opens instantly and runs entirely in the browser: no round trip, no spinner, no timeout.
That architecture also means search cannot go down. There is nothing to scale, no service to crash under load, and no third party in the loop when someone needs an emergency repair product on a holiday weekend.
Static indexes have one honest cost: they update at build time, not in real time. For an industrial catalog that changes a few times a month, that trade is free money. The index regenerates with every deploy, so it can never drift out of sync with the pages it points to.
When codes do matter, respect them completely
Problem language is how buyers arrive, but distributors and repeat customers really do know their SKUs, and a search that fumbles an exact code is worse than none. On Polymer Nation we built a five-tier deterministic ranking engine that runs ahead of the fuzzy matcher.
- Tier one: exact product code. Typing "u51" goes straight to U-51.
- Tier two: code and series prefix. Typing the single letter "u" lists the entire U series, sorted naturally so U-20 comes before U-51.
- Then name and keyword word-start matches, for people who remember the first half of a name.
- Then separator-agnostic substring matching, so "poly guard" and "polyguard" both find POLY-GUARD.
- Last, typo-tolerant fuzzy matching, so "polyspartic" still resolves to the polyaspartic products.
Deterministic ranking means the same query always returns the same order, and the logic is legible: a code hit beats a prefix, a prefix beats a substring, a substring beats a guess. When results feel random, buyers stop trusting the search, and then they stop using it.
The details that make it usable
The ranking engine is half the job. The rest is interface craft: results grouped by type so the visitor sees the shape of the answer at a glance, real product thumbnails where the catalog supports them, full keyboard support behind Cmd+K, and a genuinely accessible combobox pattern underneath rather than a styled div. None of it requires exotic tooling. It requires deciding that search is a first-class feature and giving it the same design attention as the homepage hero.
The empty state matters most. When nothing matches, our search does not shrug. It hands off to a human with a direct "talk to a specialist" path, because a failed search on an industrial catalog is usually a qualified lead in disguise.
Search is a sales conversation
The through-line in all of this: treat search like your best counter person. Understand the problem the customer walked in with, know the shelf layout cold, never make them wait, and when you genuinely do not have the answer, introduce them to someone who does.
Do that, and the search box stops being a utility and starts being the shortest path between a plant-floor problem and a purchase order.
