Fix searches that miss the right product.
Shoppers find the right product however they ask, and exact codes, stock and facts stay true.
- “rain jacket”Returned socks212Fixed
- “washable jacket”The wash ranked above the jacket96Fixed
- “waterproof jacket”Results never say waterproof unless the facts do64Held
- “gift for a hiker”Too open for a list of results58To the assistant
The failed-search ledger. The merchandiser starts each week here.
Sound familiar?
- “Rain jacket” returns socks.
- A merchandiser boosts one product at a time, every week.
- Searching the style code doesn’t always find the product.
A shopper who types a search has told you what they want. If the result misses, you paid for the visit and lost the sale.
Fernway’s launch week, failed searches
1,240 searches failed in launch week.
They returned nothing, or the shopper left without a click. Each one is a shopper using a word you don’t.
1,100 were understood.
AI reads intent, spelling and subjective phrases like “gift for a hiker”. The other 140 weren’t about this launch, so they waited.
940 got the right product first.
Rules decide the order: style codes always win, stock is checked, and “not waterproof” is shown rather than hidden.
210 bought.
One of them searched “washable jacket” and found the jacket, not the wash. Your merchandisers approved every rule behind it.
The best query got a before and after.
For “washable jacket” the wash used to rank first. Now the jacket does, and the ranking is on file.
| Rank | Before | After |
|---|---|---|
| 1 | Canvas wash | Canvas field jacket |
| 2 | Wax bar | Care Kit |
| 3 | Care Kit | Canvas wash |
| 4 | Canvas field jacket | Wax bar |
Accessories still show, below the product
What changes.
- Fixing a miss
- A merchandiser patches one query at a timeAI reads intent and spelling, and the fix covers the group
- Exact codes
- Mostly work, until a similar item outranks themStyle codes always win, by rule
- Stock and facts
- A sold-out size is the first resultStock is checked, and unclaimed facts aren’t implied
- Judging
- A few searches tried by handA judged set of real queries, run on every change
How it works. 5 steps, 4 stages.
- AI1
- Rules and automation3
- Your team1
- Approval
1Collect
Rules and automation
Collect the failed searches
Zero-result searches, rewrites and exits, grouped by query and counted every week.
- Shopify
2Understand
AI
Read what the shopper meant
Synonyms, misspellings and open phrases like “gift for a hiker”, mapped to products or to the assistant.
3Decide
Rules and automation
Keep exact codes and facts true
Style codes, stock and claims are fixed rules, never ranking guesses. “Not waterproof” is shown.
Your teamApproval
Approve the rules
Merchandisers approve the commercial priorities and the facts search may state.
4Check
Rules and automation
Run the judged set
Real queries with known right answers run on every change, including codes, misspellings and deleted products.
Next week’s failed searches start the cycle
What you get. Real work, every week.
| Rank | Before | After |
|---|---|---|
| 1 | Canvas wash | Canvas field jacket |
| 2 | Wax bar | Care Kit |
| 3 | Care Kit | Canvas wash |
| 4 | Canvas field jacket | Wax bar |
Accessories still show, below the product
The ranking for one failed query, before and after. The jacket is first, and the care products follow it.
19 of 20judged queries return the right product first
- CAT-042 finds the Canvas field jacket
- Sold-out Olive M offers Navy M
- “jakcet” still finds jackets
- A deleted product never appearsOne cached result
The test sheet that runs on every change, so a fix for one query can’t break another.
How you’ll know it’s working.
Searches that find the right product
Judged on real queries, including exact codes and misspellings.
Contribution per search user
Held out against shoppers on the old search. Comparing search users with all visitors proves nothing.
Merchant hours spent patching
The weekly boosts and synonyms that shouldn’t be needed any more.
Baselines agreed before we build.
Works with what you have.
Commerce
- Shopify
- Adobe Commerce
Search & AI
- Algolia
- Constructor
Measurement
- Google Analytics
- We configure first
- As of October 2026, Shopify’s Search & Discovery is free and includes semantic search, and Adobe Commerce ships its own. We judge what you have on your real queries before suggesting anything else.
- We build when
- Exact codes, stock rules and a judged query set need logic your platform’s settings don’t hold.
What a first build includes.
Built in your accounts, around your team. Scope, fee and baseline agreed before we start.
- 1Design
- 2Build
- 3Run it, or hand it over
We build
- A weekly failed-search review
- Intent and spelling rules around your platform’s search
- Exact-code, stock and claim rules
- A judged query set that runs on every change
We start from
- Search logs and click data
- Product catalog and stock
- Past boosts, synonyms and redirects
From your team
- Owner
- Head of ecommerce, with merchandising
- Involves
- Ecommerce
- Merchandising
- Catalog
- Engineering
Questions teams ask.
Isn’t semantic search built into our platform now?
Often, yes. Shopify and Adobe Commerce both ship it. We test it on your real queries first. What’s usually missing is the rules around it: exact codes, stock and what a page may claim.
Can AI show things we don’t sell, or claim things that aren’t true?
Not through search. Rules decide what’s eligible and which facts show, and a semantic match never outranks an exact code. A claim with no source stays out.
Do merchandisers lose control?
No. They approve the rules and the priorities, and every change is on file. What goes away is boosting one product at a time.
Who owns it once it’s running?
You do. It’s built in your accounts, documented, with training for the people who run it. Ownership usually sits with your head of ecommerce, with merchandising. We can keep running it with you, or step back.
How do we start?
Usually with a conversion review: two to four weeks, a fixed fee credited to the build, and a plan that’s yours. If the job is already clear, we can scope the build directly.
Part of your conversion engine. One of 14 use cases, in 5 parts.
See the whole conversion engineIn search & discovery
- Site searchYou’re here
- Search platform choiceDecide whether to keep, tune or replace your site search
Works with
Help shoppers find what they meant. Start with a conversion review.
We reply within one business day.