← All work
Client project

AragoCor Minerals

A marketing site for an industrial mineral supplier selling one material, in four defined grades, to buyers in very different industries, and the AI sourcing tool built to turn visitors into quote requests.

In short

  • The problem. One mineral sold to very different buyers, from soil agronomists to glass engineers, who share almost no vocabulary and check every number.
  • What I built. A site organized by grade and by buyer, with the sizing and documents each one checks first and "Request bulk pricing" on every page, plus a freight and quote tool on its own URL.
  • What it means for you. If you serve several kinds of customer, each gets a page in their own terms, all pointing to the one action that pays.
Client
AragoCor Minerals, supplier of Bahamian oolitic aragonite
Sector
Industrial minerals, business-to-business
Scope
Website design and build, technical copy, controlled data sheets, AI catalog, freight calculator and quote-request tool
Year
2026
Status
Site live at aragocorminerals.com; the domain was last checked answering on October 19, 2026, and its content last read on October 11. The sourcing tool is a separate build and is not on the main site.
AragoCor Minerals homepage: oolitic aragonite, supplied in four defined grades, over open ocean, with 96.17% calcium carbonate stated in the first sentence

At a glance

each with its published sizing, its own data sheet and its own quote and sample request
4 grades
calcium carbonate, quoted as one named lab report's analysis rather than a range nobody measured
96.17%
industry pages for buyers who know their process but not yet their grade
11 industries
“Request bulk pricing” in the header of every page
1 action

The situation

AragoCor supplies Bahamian oolitic aragonite into very different markets: agriculture, glass, water treatment, polymers and coatings, and a longer tail that runs from aquaculture to construction. Same mineral, buyers who care about entirely different things. A soil agronomist and a glass batch engineer share almost no vocabulary.

Most industrial supplier sites solve this by going vague — a stock photo, "quality solutions for your needs," and a contact form. That loses the technical buyer, who is scanning for one thing: does this material meet my spec?

The site needed to be specific enough to satisfy an engineer and organized enough that five audiences could each find their own room.

Constraints

The buyer is technical and skeptical, so vague claims actively lose the sale. Eleven application areas had to be served from one product line without the site fragmenting into eleven sites, and every figure had to survive a buyer checking it against a lab report. And anything added later — including the sourcing tool — had to drop onto the finished site without a rebuild.

What I built

A structure organized around the buyer, not the org chart. Products are graded by process — AG-CAL, GL-CAL, WT-CAL, PL-CAL — each carrying the published sizing a buyer actually checks for: 0.2–0.5 mm, under 0.1 mm, 1–3 mm, and for the filler grade, classification to the buyer's own written specification. Each grade has its own controlled data sheet and its own quote and sample request. Industries gets its own path, eleven pages, for buyers who don't yet know which grade they need, and it says which of them have a grade stocked for them and which are served by custom classification. Science holds the technical depth; Resources holds the data sheets, marked as the place procurement starts.

The industries page: eleven cards from agriculture and aquaculture to turf and golf, each ending in its sizing and the grade that serves it
One mineral, eleven industries. The page is the fork in the road for a buyer who arrives knowing their process but not their grade; every card ends in its sizing and its grade, or in "to specification" where there is no published size.

Copy with numbers in it, and only numbers that can be traced. 96.17% CaCO₃, stated as what it is: the representative analysis of one identified sample, by ASTM C 25, from a named lab report, and not a guaranteed minimum. Sizing given grade by grade, each labeled a published range. No kiln, no calcination step. OMRI listed, with both listing numbers and what each one does and does not cover. Buying written out in four plain steps, from choosing a grade or sending a specification to a delivered quote, and the supply chain named company by company rather than gestured at. Writing this required understanding the material, not just restyling a brief.

Product catalog: four grade cards, AG-CAL, GL-CAL, WT-CAL and PL-CAL, each with its sizing, qualification documents, representative analysis and quote, sample and data sheet buttons
Four grades, chosen by process. Each card names the buyer it serves, its published sizing, the documents that qualify it, and its own quote and sample request.

A visual system that stays credible. Open-ocean imagery, editorial serif headlines, generous whitespace. Restrained enough to read as a serious industrial supplier rather than a startup.

Conversion built into the frame. "Request bulk pricing" sits in the header on every page — the single action the site exists to drive — and every other ask on the site, a quote or a sample from a grade card, lands on the same form with the request and the grade already filled in. Each request is also passed straight into AragoCor's own system, so the first reply starts from what the buyer asked.

Then: the sourcing tool

The site answered what AragoCor sells. Buyers still stalled on the next question — what it costs to move a container from origin to their port — and every one of those went to a salesperson, who spent the day re-answering the basics.

So I built a second layer: a searchable mineral catalog, a maritime freight and stowage calculator that returns real shipping figures, and RFQ capture that routes to sales with the conversation already attached, so the first human reply can be an actual quote. It runs on live AI with an offline knowledge engine behind it, so it degrades instead of failing when a provider is unreachable. It was built to drop onto the existing site as a floating widget, with no rebuild.

Dual-mode AI: live model with an offline fallback A buyer question goes to the live Gemini API. On success the answer returns directly. If the live service is unreachable, the question falls through to an offline knowledge engine, which returns the answer instead. Buyer question PRIMARY PATH Gemini API unreachable FALLBACK Offline knowledge engine on success Answer returned
The tool sits on the revenue path, so it has two ways to answer. When the live model is unreachable the offline engine takes the question — the buyer gets a degraded answer instead of an error.

Decisions and tradeoffs

Specificity over polish. A measured CaCO₃ figure and four grade codes in the hero, not "premium quality minerals." That narrows the audience and converts the part of it that matters.

Withdraw a claim rather than defend it. Earlier copy said 96–98% CaCO₃ and gave one 20-to-325-mesh range across every grade. The range of purity had come off a third party's letterhead rather than a test of this material, and the mesh range was not true of the coarsest grade, which is coarser than a No. 20 sieve. Both were replaced with what can be traced — one lab report's figure, sizing grade by grade — and both now sit on a register of withdrawn claims that the site's own validator fails on, so neither can creep back in. When the nine one-page application brochures turned out to still carry claims the site had dropped, they came down on October 2 and their URLs were redirected to the matching pages.

A second answer path, at a cost. Building the offline fallback took real time up front. It also means a sales tool that stays up when a third-party API doesn't — a fair trade for something sitting on the revenue path.

Standalone, not embedded. The tool went live on its own URL first, so the freight logic could be tested against real use before it appeared in front of existing customers. In September the decision was made not to have a chat assistant on the main site at all, and the one that had been built into it was removed, so standalone is now the answer rather than a stage.

Calculator over chatbot. The AI is the interface; the freight number is the product. Plenty of tools will chat with a buyer. Far fewer give them a number they can act on.

RFQ path: from buyer question to sales inbox A buyer question routes to either catalog search or the freight calculator. Both feed RFQ capture, which delivers to the sales inbox with the conversation context attached. Buyer question Catalog search Freight calculator RFQ capture Sales inbox with conversation context attached
Both answer paths end in the same place. Sales opens an RFQ that already carries what the buyer asked and what the calculator returned, so the first human reply can be a quote.

Result

What can be counted on the live site, checked on October 11, 2026. Inquiries and conversions are tracked — the site was instrumented from launch — but those are the client's figures, and they are not published here.

Product grades
4, each with its published sizing, data sheet, quote and sample request
Priority markets
4 with a grade stocked for them; 7 more served from the same grades
Industry pages
11
Controlled data sheets
5 — the master sheet and one per grade, reissued September 25
Primary action
1 — "Request bulk pricing", in the header of every page
Sourcing tool
Catalog search, freight calculator, RFQ capture
Answer paths
2 — a live model, an offline engine behind it

Where it stands

The site is live at aragocorminerals.com, with conversion events on every step of the quote path — request started, submitted, completed — so the question of what's working is answered with data rather than opinion. In September the analytics moved from Plausible to Google Analytics alongside Vercel's own; Google's side has recorded since October 8, when a content-security rule that had been blocking it was fixed.

Since the last check, the site has gained a blog, nine posts written for industrial buyers, from October 8, and lost the nine application brochures, for the reason above. The sourcing tool stays separate from it: as of October 11, 2026 the main site carries no widget and no chat assistant.

What this means for your project

Most local businesses are not selling one mineral into five industries. The problem underneath is common, though: one business, several kinds of customer, and a site that has to be specific to each of them without becoming five sites.

  • One page per kind of customer, each written in that customer's own terms, so a visitor lands somewhere that already sounds like their problem.
  • The numbers in the copy, and where each one comes from. Prices, sizes, lead times, the towns you serve. The specifics that make a skeptical reader trust a page are the ones most sites leave out, and the ones that cost the most when they turn out to be wrong.
  • One action, everywhere. Whatever your equivalent of "Request bulk pricing" is, it sits in the header of every page and nothing else competes with it.
See what a build includes
Next case studyAragoCor Minerals — internal systemRead it →