Technical-sales qualification · specialty chemicals
The grade that fails the spec never reaches your customer.
Deterministic screening reads the inquiry, checks it against your line card, and drafts the reply — the rep approves before anything is sent.
This screens a demo catalogue of fictional coatings products — the real deployed engine, not your line card. Nothing here is a real product, and no reply is sent.
It is also complete by construction: these products publish the properties a screen asks about, so almost every requirement gets a verdict. Real catalogues are rarely that complete — where a data sheet is silent GRADE returns unknown and names the value it needs, rather than guessing. That gap is a data question, and we measure it per field on the catalogue you send us.
Parsed constraints
What it does
Turns an inbound spec into a defensible shortlist your rep can send — every claim quoted verbatim from your own data sheet.
How it works
Why not just a chatbot
Structured beats plausible.
| GRADE | Chatbot | Vector search | Spreadsheet | |
|---|---|---|---|---|
| Enforces a hard spec exactly | ||||
| Same screen result every time | ||||
| Explains every elimination | ||||
| Quotes your own data sheet | ||||
| Won't invent a product |
Measured against the strongest chatbot on raw data sheets: 17% recall — and even handed a clean table, it returned a different answer on 7 of 12 inquiries when asked again. GRADE reads, code decides.
Pricing
Priced to your line card, not per seat.
You pay for catalogues screened and inquiry volume — not headcount. Add a rep, add ten; the price doesn't move.
- Parse · Screen · Draft
- A verbatim quote on every claim
- Every run logged end to end
- Everything in Starter
- Multiple catalogues
- Higher inquiry volume
- Multiple rep teams
- Everything in Growth
- SSO · custom conventions
Every tier works the same way: a hard constraint is enforced as a SQL filter over typed columns and never as a similarity score, every claim is quoted verbatim from your own data sheet, every run is logged, and no reply reaches a customer without a rep's approval. Accuracy, provenance, and the audit trail are the product — never an upsell.
About
A model reads. Code decides. The human approves.
GRADE qualifies inbound technical inquiries for specialty-chemical producers and distributors — so the grade that fails the spec never reaches the customer.
Every product is parsed once into typed columns — pH as two numbers, a certification as a boolean. Screening is deterministic SQL over those columns. A model only reads the inquiry and drafts the reply; it never decides what passes, and it only ever sees products that survived the screen.
Why not just a script? Because the hard part isn't the filter — it's the conventions: what "low oil absorption" means for a flatting agent, when a formula name and a trade name are the same chemistry, how much tolerance a near-miss deserves. That encoded, attributed, versioned knowledge is the part a competitor can't copy from the outside.
The design rule. A hard constraint is a SQL WHERE clause over a typed column, never an embedding comparison. Every elimination is explainable in one sentence. No reply reaches a customer without a rep's approval — if we can't say why a product was ruled out, the screen is wrong, and we fix the screen.
From the blog
Notes from building GRADE
How we think about deterministic screening, provenance, and keeping a human on every send.
- August 25, 2026“Always learning” is a bug, not a featureIn a regulated industry, software that changes its own behaviour is software nobody can validate.Read post →
- August 20, 2026What we do not claimA short list of things our product does not do, published deliberately.Read post →
- August 18, 2026Why every number shows the sentence it came fromA specification you cannot trace is a rumour. Our extraction rejects values it cannot quote.Read post →
FAQ
Straight answers.
The questions a distributor asks first — the honest version of each.
Why can't I just use ChatGPT?
We measured it. Against the strongest chatbot on raw data sheets, recall was 17% — it missed most of the grades that fit. Even handed a clean typed table, with extraction removed as the excuse, it gave a different answer on 7 of 12 inquiries across repeated runs — and on the hardest, no two runs agreed at all. A chatbot writes something plausible; it doesn't enforce a hard spec. GRADE screens with typed rules first, then drafts — and the model only ever sees products that survived the screen, so a reply can only name a grade that is in your catalogue and passed. See the full comparison →
Does GRADE email my customers?
No. GRADE drafts a reply and hands it to a rep. The rep edits and approves; the rep sends. Nothing leaves on its own — the only automatic message is a notice to your own team that an inquiry arrived.
Is the answer the same every time?
The screen is. Screening is deterministic SQL over typed columns: the same constraints against the same catalogue eliminate and pass exactly the same products, with the same one-sentence reason on every elimination. Reading the inquiry is a separate step — a model turns prose into typed constraints — so the rep sees the constraints we extracted, can correct them, and approves before anything is sent. Every claim in the reply is quoted verbatim from your own data sheet.
What happens when nothing matches?
It says so. "No product meets your spec" is a real, defensible answer — and it names the binding constraint. GRADE never stretches a near-miss into a recommendation to fill the gap.
What do you need to start?
Your line card — technical data sheets, in whatever form you have them. We parse each product into typed columns once, and screening runs against that. How the screen works →
Is my data used to train a model?
No. Your catalogue and your inquiries are yours. Every run is logged for your own audit trail — input, constraints, survivors, what the rep sent — and that log is your asset, not training data.
What does it cost?
It's priced on catalogues screened and inquiry volume, not per seat — adding reps doesn't change the price. See the tiers →
Get started
See it on your line card.
Tell us what you carry and what you'd like to see — we'll screen a real inquiry against your line and walk you through every receipt.
Thanks — we'll be in touch.
A rep will read what you carry and reach out to line up a screen against your catalogue.