Frequently asked questions

It is a weighted score across eight components: specification match, semantic similarity, keyword relevance, price fit, application fit, availability, demand, and supplier reliability. Specification match carries the most weight, and the weighting shifts by category — ingress protection matters more for an enclosure than for a valve.

The figure is computed by deterministic code, not generated. The same query against the same catalogue always produces the same number, and the breakdown beside each result shows the component scores that produced it.

Scores are banded between 42% and 97%. A perfect match is a claim nobody can honestly make — there is always a requirement that was not stated, an installation constraint, a preferred brand. Presenting 100% would imply the system knows everything about your application, and it does not.

Only two things filter absolutely: a stated maximum price and an explicit in-stock requirement, plus anything you have ruled out by name. Everything else is scored rather than excluded, so if you ask for stainless steel you still see the bronze alternative — clearly labelled as a substitute rather than silently dropped. The criteria panel marks every miss.

Zero-result searches are recorded and reviewed. They are the most useful signal the platform generates: each one is either a sourcing gap — the catalogue genuinely lacks the product — or a vocabulary gap, where the parser did not recognise a term. The two need different fixes, and the parsed intent recorded alongside the query is what separates them.

Two things. It converts your description into a structured requirement, and it rephrases results that have already been computed.

It has no write authority over anything a buyer relies on. It does not choose which products are shown, does not order them, does not compute the match score, and never touches a price. Those are all deterministic server-side code.

The assistant keeps working. A built-in parser handles the same vocabulary — specifications, units, budgets, negations like “not brass” — and the ranking engine is unchanged, because the model was never part of it. Explanations become templated rather than conversational, and the interface says so rather than pretending nothing changed.

Every product shown is a record from the catalogue, retrieved by the search engine and rendered from the database. The model is never asked to produce a product, a specification or a price, so there is nothing for it to fabricate.

Only when one question would meaningfully narrow the results. The assistant picks the specification whose answer eliminates the most candidates — measured by the spread of values across the current result set, weighted by how much that specification matters in the category. If no question would help, it does not ask one.

Industrial supply is quoted, not transacted at list price. What you pay depends on quantity, delivery location, GST treatment and often the relationship. A checkout button would have to invent a number. Instead a shortlist becomes a quotation request that carries the specification with it, and suppliers respond with real pricing and lead times.

It depends on the supplier and the line count. You are notified in the app and by email when pricing is ready, and the request itself shows its status throughout — submitted, under review, quoted. Nothing requires you to keep checking.

No. Catalogue prices are indicative list prices, shown so you can compare like with like and filter by budget. GST is stated separately per product. The binding number is the one on a quotation.

Up to 4 side by side. The comparison table shows every specification either product declares, highlights the rows where they differ, and flags where one has data the other does not — an absent specification is not the same as a worse one.

No. Search, the assistant and comparison all work signed out, and a shortlist you build anonymously is carried over if you create an account later. An account is needed to request a quotation, because a quotation has to reach somebody.

Searches are recorded with their result counts to measure where the catalogue and the parser fall short. Signed in, your own history is visible on your dashboard and you can clear it at any time. See the privacy policy for the full picture.

Hashed with bcrypt at cost factor 12 and never stored or logged in any recoverable form. Sign-in takes the same time whether or not the address is registered, so the form cannot be used to discover who has an account. Changing your password signs out every other device.