A contractor standing at a trade counter does not type circuit breaker. He types circut breakr 20amp, one-handed, on a cracked phone, while a customer waits. A purchasing clerk clearing a backlog of requisitions does not know the manufacturer part number for the conduit fitting on line fourteen; she knows it is three-quarter inch, steel, and the job needs it Thursday.
Every distributor knows this. It is the reason counter staff exist. What is less obvious is how badly the search box on the company website handles it — and how rarely anyone checks.
So in July 2026 we checked. We took twenty North American distributors, mostly electrical and electronics, opened the product search on each of their public catalogs, and typed three things into it. No special access, no API keys, nothing a customer could not do. The results were worse than we expected, and the pattern in them was not the one we went looking for.
What we typed
Three queries, chosen because they represent three different ways a real person arrives at a search box.
- A plain-language description. Something like flexible metal conduit 3/4 inch — the way you would describe a product to a person if you did not have the part number in front of you.
- A generic category term. Wire connectors. Broad, unambiguous, the kind of thing you would type if you were browsing rather than buying.
- A misspelling. One transposed letter in a common part: circut breaker 20 amp. Not gibberish. One letter.
Then we looked at what came back, and whether the first three results were the product the query was obviously asking for.
This is a spot check, not a peer-reviewed study. Twenty sites, one session, three queries each, run through the public search box on each catalog. A handful of sites could not be tested at all — some sit behind bot-detection challenges, some appeared to be geo-restricted from where we were testing — and those are excluded rather than counted as failures.
Search platforms are also configurable. A bad result reflects how a specific deployment was set up and tuned, not the theoretical ceiling of the product underneath it. That distinction matters, and we come back to it.
The category term was fine almost everywhere
Type wire connectors into nearly any distributor's search and you get wire connectors. This was true of the largest sites and the smallest, the ones running expensive commercial search platforms and the ones apparently running nothing in particular.
Hold onto that, because it explains why the problem persists. The category term is the demo query. It is what gets typed in the vendor's sales presentation, in the internal acceptance test, in the five-minute check somebody runs before signing off. It works, everyone nods, and the box gets ticked.
Search that passes the demo query and fails the real one will pass every review it is ever given.
The misspelling is where it fell apart
One letter. That is all it took.
At several of the distributors we tested — including national names, including sites running major commercial search platforms — circut breaker 20 amp returned zero results. Not a degraded result set. Not a "did you mean" prompt. An empty page, at a company that stocks tens of thousands of circuit breakers.
The most striking case was a distributor running one of the search platforms marketed specifically to the distribution industry — the kind that comes pre-integrated into B2B commerce platforms and gets recommended by implementation agencies. It returned nothing for the typo, and nothing for the plain-language conduit query either. Only the bare category term worked.
Others degraded more quietly. One site on a well-known hosted search product returned the right product family for the typo but ranked the wrong sizes first. Another, running a major enterprise search suite, answered a query for circuit breakers with lockout accessories — related products, wrong products. A third, on a self-hosted Elasticsearch cluster, responded to a single conduit query with more than 260,000 results, which is a way of saying it had no idea.
Descriptive queries rank the accessories
The plain-language query produced its own failure mode, and it was consistent enough to be worth naming.
Ask for a conduit by size and material and the top hits come back as fittings, couplings, connectors and lugs. All of these are things that go with the product. None of them is the product. The engine has matched on the word "conduit," found every SKU whose description contains it, and ranked by something — popularity, margin, alphabetical order, who knows — that has no relationship to what the query meant.
A person who knows the catalog spots this instantly. A person shopping does not; they conclude you do not stock it, and they leave.
The finding we did not expect
We went in assuming the split would be obvious: distributors who pay for a commercial search platform would have good search, distributors who do not would have bad search. That is not what we found.
Some of the best results we saw came from sites with no recognizable commercial search product on them at all. And some of the worst came from sites paying for one.
| Search stack | Category term | Descriptive query | One-letter typo |
|---|---|---|---|
| Distribution-specific commercial platform | Works | Zero results | Zero results |
| Major hosted search SaaS | Works | Right family, wrong ranking | Zero results |
| Enterprise search suite | Works | Accessories ranked first | No correction |
| Self-hosted Elasticsearch | Works | 260,000+ results | Zero results |
| No identifiable search vendor | Works | Often fine | Mixed |
| Large electronics distributors | Works | Works | Works |
That last row deserves its own paragraph. The big electronics distributors — the ones with millions of parts, where a search failure is an existential problem rather than an inconvenience — have genuinely excellent search. They have invested in it for years because their entire business model depends on a person finding one specific component among millions. They are the proof that this is solvable.
Buying a search vendor did not predict having good search. Caring about search did.
Why this happens
None of the platforms we saw failing are bad products. Several are genuinely sophisticated. The failure is structural, and it has three parts.
Typo handling is usually a configuration, not a property
In most search platforms, tolerance to broken input is something you switch on and then tune. Fuzziness settings, edit-distance thresholds, synonym dictionaries, "did you mean" rules, redirects for known bad queries. Every one of those is a decision somebody has to make, test, and maintain as the catalog changes.
Which means it decays. The person who tuned it leaves. The catalog grows a new product line nobody wrote synonyms for. The fuzziness setting that worked at 40,000 SKUs starts returning noise at 200,000, so somebody tightens it, and now typos return nothing. That is not a bug. That is the system working exactly as designed, under an assumption that stopped being true.
Distributor catalogs are hostile to the assumptions
Most search technology was built for retail. Retail product names are written for humans: "Ceramic Non-Stick Frying Pan, 28cm." Distributor catalogs are written for systems. Descriptions are truncated, inconsistent across manufacturers, stuffed with abbreviations, and full of part numbers that look like line noise. The same product might appear as CONN CONDUIT 3/4 STL SET SCRW in one feed and something entirely different in another.
A relevance engine tuned on retail language does not know what to do with that. It matches tokens and hopes.
Nobody runs the real test
The category term works, so search "works." Nobody sits down and types the way a contractor types, because everyone involved in evaluating the search box already knows the correct spelling of circuit breaker.
Run this yourself, in thirty seconds
You do not need us, or a consultant, or a report. Open your own catalog and do this:
- Type a common product the way a customer would describe it, without the part number.
- Type the same thing with one letter wrong.
- Look at the first three results, and ask whether a customer in a hurry would have found what they came for.
If the answer is no, you now know something about your own business that your search vendor has never told you. And you can measure what it costs: pull your site search logs, filter for queries that returned zero results, and count how many of them were real products you actually stock. Every one of those is a customer who came to buy, could not find it, and went somewhere else.
A misspelled part number lands on the right product. A plain-language description ranks the product above its accessories. A partial part number narrows rather than empties. Two words glued together get pulled apart. None of this requires a person to have written a rule in advance, because the rules cannot be written in advance — you cannot enumerate every way a human being will mistype something.
What we do about it
We build a catalog search engine, so read the next paragraph with that in mind.
Dartfind treats broken input as a property of the matching math rather than a configuration on top of it. Typos, partial part numbers, glued words and wrong keyboard layouts land on the right product without synonym lists or fuzziness settings, because there is nothing to tune — the tolerance is in how the matching works, not in rules layered over it. It installs as a compiled program on a server you already own, it never sends your catalog anywhere, and it is a one-time license rather than a subscription.
The honest version of the pitch is this: we are not asking you to believe the table above. We are asking you to type a part number wrong on your own site, look at what comes back, and then decide whether it is worth a two-day pilot to see something different.