It started with an article that annoyed me
A while back I went down a rabbit hole reading about why online stores lose sales. One piece stuck with me: Baymard Institute's research on e-commerce search. Their large-scale usability testing found that 61% of sites can't handle the way people actually search, meaning the kind of query where a shopper describes what they want in their own words instead of typing the exact product name a merchandiser chose. (Baymard, “The Current State of E-Commerce Search”.)
The detail that got me was the breakdown. Searches like “feature” queries, “use case” queries, and “symptom” queries fail on roughly 4 to 6 out of every 10 sites. Someone types “something warm for hiking in the rain” and the store returns nothing. Not because it lacks the product, but because the words don't line up with the title. The product is right there. The shopper just can't find it, so they leave.
What makes that hurt is who those people are. Shoppers who use the search box are not browsers. They're buyers. Depending on whose data you read, site search is used by something like 15% to 30% of visitors, and those visitors can drive close to 40%+ of revenue while converting at a meaningfully higher rate than people who only click through menus. (See Algolia's roundup of search KPIs, 40+ stats on e-commerce search.) So a broken search box isn't a small UX wrinkle. It's a tax on your highest-intent traffic.
So I built one for myself
I run a small WooCommerce store, and WooCommerce's default search is keyword matching. It looks for your literal words in the product text, which is the exact thing the research said was failing. I'm an engineer, so instead of complaining about it I spent a few evenings building a proof of concept.
The idea is called semantic search. Instead of matching letters, you convert every product and every query into a vector, a string of numbers that captures meaning, and then you compare meanings instead of spelling. “Comfortable hiking shoes” lands near “cushioned trail running sneakers” even though they share no words. It's the same family of technique that powers the search behind the big platforms. I just wanted it on my own store without paying a monthly SaaS bill or shipping my catalog to someone else's servers.
Then I looked at the before and after
This is the part that changed my mind about what I'd made. I ran the same set of real, messy queries against my store twice, once with default WooCommerce search and once with the prototype, and put the results side by side. Queries that used to return the dreaded “no products found” page were now returning the right products. Not approximately right. The actual item the shopper was describing.
I'd read the statistics and nodded along. Seeing it happen on my own products, with my own catalog, was different. The empty results pages were the leak, and I'd just plugged it. At that point it stopped feeling like a weekend experiment and started feeling like something every WooCommerce store should have.
Why I made it public
Here's the thing. The problem I solved for myself isn't special to my store. It's on more than half the storefronts on the internet, including plenty of shops run by people who don't have an engineering background and can't spend their evenings writing vector-search code. The fix shouldn't be locked behind enterprise pricing or a developer's spare time.
So I turned the prototype into a proper plugin. I gave it three engines so any store could use it: a Local engine that runs entirely in the shopper's browser for $0 forever, a free Hugging Face tier, and an OpenAI option for stores that want the highest quality for a couple of dollars a year. No subscription, no sending your data anywhere you don't control, no monthly invoice that makes you regret installing it. You buy it once and you keep it.
How I work on it
It's just me. I write the code, I answer the support emails, and I decide what goes on the roadmap. That has a downside, since I'm one person, so I'm honest about what the plugin does and doesn't do yet (multilingual and multisite are on the list, not in v1). But it has an upside too: when you email support@findastra.com, you're talking to the person who built the thing, not a ticket queue.
If you want to follow along or argue with me about search, I'm on X at @TarikElyass. And if you're still deciding whether semantic search is worth it, skip the sales pitch and try the live demo on the how-it-works page. Same query, your eyes, two very different result sets. That comparison is what convinced me. It might do the same for you.
– Elyass