Parts Search With 60% Faster Time-to-Find Across 500K SKUs Using Typesense
We replaced a distributor's SQL search with Typesense, so buyers find parts as they type, even with partial part numbers, typos, and trade shorthand.
faster average time-to-find
SKUs searchable on every keystroke
empty pages from a mistyped character
A parts catalogue where one wrong digit meant zero results
Our client is a B2B distributor of industrial parts, supplying maintenance, repair and operations buyers, contractors, and manufacturers. Its catalogue spans roughly 500,000 SKUs, many identified by manufacturer part numbers, abbreviations, and cross-references rather than descriptive names. The business wanted its online channel to carry more revenue without adding sales headcount, but a part number with one wrong digit, or a trade shorthand the catalogue did not recognize, returned nothing. Each failed search became a phone call or a lost order.
The client partnered with Ksolves to replace the database-bound search with one that matches how buyers actually type, while keeping PostgreSQL as the system of record.
C H A L L E N G E
01
Slow Exact-Match Search The legacy search ran pattern matching on PostgreSQL across 500K SKUs and returned nothing until the buyer finished typing and submitted.02
Part Numbers Typed Loosely Buyers entered partial part numbers, dropped dashes and prefixes, or typed only the first few characters, and the search could not match them.03
Zero Results for Small Typos One transposed character in a part number or product name returned an empty page, even when the exact part was in stock.04
Trade Shorthand Ignored Common abbreviations and alternative names for the same item produced different results, or none at all.05
No Way to Narrow Results Broad queries returned long, unfiltered lists with no facets for category, brand, or specification, so buyers scrolled or gave up.06
Failed Searches Became Phone Calls Customers who could not find parts online called the sales desk, raising cost-to-serve and limiting how far the online channel could scale.S O L U T I O N
Search that matches how buyers type, not how the database stores data
Ksolves, an AI-first DevOps services company, replaced the database-bound search with a dedicated Typesense layer. PostgreSQL stays the system of record, and catalogue data is synced into a purpose-built search index. Every relevance decision, from typo handling to part-number matching, was tuned against real buyer query patterns. Each fix below answers the challenge with the same number.
- 01 Instant, as-you-type results All 500K SKUs are indexed in Typesense and return results on every keystroke through InstantSearch.js, ending the submit-and-wait pattern.
- 02 Prefix matching and normalized part numbers Prefix search plus part-number fields stripped of dashes, spaces, and prefixes let partial, loosely formatted entries resolve to the right SKU.
- 03 Typo tolerance tuned per field Misspelled product names and mistyped part numbers still return the intended item, with tolerance set separately for codes and names.
- 04 Synonym sets for trade terms Industry abbreviations, alternative names, and cross-references map to the same products, so equivalent terms return consistent results.
- 05 Faceted filters for fast narrowing Facets for category, brand, and key specifications cut a broad query down to the right part in a few clicks.
- 06 React storefront search experience One React interface combines an autocomplete dropdown, faceted result pages, and part-number-aware result cards, built to resolve searches online instead of at the sales desk.
Architecture
Technology Stack
| Category | Technology |
|---|---|
| Search Engine | Typesense |
| Frontend Search UI | InstantSearch.js |
| Frontend Framework | React |
| System of Record | PostgreSQL |
D E L I V E R Y
How a keystroke becomes the right part
Ksolves delivered a dedicated search layer that runs alongside PostgreSQL. Catalogue data is synced from the system of record into Typesense, and the React storefront queries it on every keystroke through InstantSearch.js. Because search sits on its own tunable layer, the business can adjust relevance without touching its transactional database. The same foundation supports merchandising rules, personalized ranking by customer account, and search analytics to guide catalogue improvements.
From submit-and-wait to finding the right part as buyers type
Ksolves replaced slow, exact-match SQL search with a Typesense layer that matches how buyers actually type. Average time-to-find dropped by 60%, typos no longer end in an empty page, online order share rose, and phone enquiries fell. Search now runs on a tunable layer beside PostgreSQL, so the business can adjust relevance without touching its system of record.
- 60% Faster Average Time-to-Find Buyers used to type a full query, submit it, and refine it several times. Instant as-you-type results cut average time-to-find by 60%.
- 500K SKUs Searchable on Every Keystroke Slow pattern-matching queries returned results only after submission. All 500,000 SKUs are now searchable as the buyer types.
- 0 Empty Pages from a Mistyped Character A single wrong character used to return nothing. Typo tolerance now returns the intended part across the full catalogue.
- ↑↓ Online Orders Up, Phone Enquiries Down Buyers who could not find parts online used to call the sales desk. Online order share has risen while phone enquiries have fallen.
T E A M
Meet the Engineers on Your Project
Every engagement is staffed by senior engineers from our DevOps team, who will design, build, and support your platform.
READY TO BUILD YOURS?
Stop Losing Orders to Searches That Find Nothing
How your catalogue handles part numbers, typos, and trade terms decides whether customers buy or leave. The engineers who built this will show you what better search looks like.