Design process
A case study in design thinking: the brief behind Makers Intelligence, the directions we walked away from, and the decisions — with their costs — that shaped what shipped.
The client & the brief
Makers Intelligence™ is a real, pre-launch venture: an AI operating system for fashion, beauty, artisan and craft businesses — customers, orders, bookings, follow-up, content and growth held in one system. Born in Ghana, priced in cedis first, built to travel.
The brief for this site: open a founding-makers waitlist of two hundred businesses, and earn the trust of people who have been burned before by software that wasn’t made for them. The site had to feel premium enough for an atelier exporting to London, and warm enough for a seamstress in Osu who runs her order book in an exercise book.
What’s real and what isn’t. The product is pre-launch: the dashboard shown on the home page is a design illustration, the three maker stories are composite personas from research conversations — not testimonials — and founding pricing is stated as pre-launch pricing, confirmed at launch. Every one of those caveats appears on the page where the claim is made, not buried here.
The problem
Talk to makers in Accra or Kumasi and the pattern repeats. The craft is under control — it’s the remembering that overwhelms: measurements, deposits, fitting dates, the forty December orders, the customer who only replies on WhatsApp voice notes. Existing tools fail them three ways:
So the design problem was never “make an AI product look impressive.” It was: make intelligent software feel like it was made by people who respect the work of hands.
Creative direction
Each candidate direction was tested against one question: does it serve the maker looking at it, or the technology behind it?
Rejected
The default costume of AI products: dark gradients, orbiting particles, terminal type. It photographs well and says absolutely nothing to a seamstress. It sells the technology, not the care — and technology-first is precisely the posture Makers exists to reject.
Rejected
Beautiful hands at sewing machines, golden-hour workshops. Two fatal flaws: this collection permits no photography — every visual must be drawn in code — and borrowed warmth is a lie anyway. Stock photos of “African artisans” are exactly how software that wasn’t built here decorates itself.
Rejected
Ample white space, tidy grid, feature cards — competent, safe and placeless. Nothing of Ghana survives except the copy. When “Born in Ghana” is the first line of the brand, the design system itself has to prove it, not just the words.
Chosen
Kente is Ghana’s most sophisticated information system — pattern as language, woven by hand. So the site became a loom: pages compose as woven bands, edges interlock, icons are single threads, and the hero literally weaves itself. The craft metaphor is structural, not decorative — which is the same promise the product makes.
Key decisions
The woven hero could have been a Three.js scene or a looping video. A hand-written fragment shader costs ~8 KB and one draw call; Three.js would add ~150 KB for capabilities we don’t use, and video would betray the code-drawn thesis. Trade-off: more engineering care — context loss, DPR capping, a designed SVG fallback — all of which is documented in the build guide.
The dramatic move would be a full dark site. But makers will read this on mid-range phones in daylight, and long ivory passages read better and feel more like paper — more like a pattern book than a pitch deck. Emerald is saved for chapters that carry weight: origin, the waitlist. Trade-off: the site leads quieter than a full-dark flagship; the loom hero carries the drama instead.
Pricing leads with GH₵ — not converted from USD, and not hidden behind “contact us.” The dollar approximation sits second, for makers who sell abroad. It is a small typographic decision that makes the entire positioning credible. Trade-off: international visitors do a little math; that’s the right person doing the work.
The product moment shows follow-up drafts and a Makola shopping list — not analytics dashboards. Copy names real conditions: tro-tro data, cash deposits, Sunday booking messages. Trade-off: it reads intensely local. That is the moat, not a bug: “built for the world” only lands after “born in Ghana” is proven.
Pre-launch products are usually dressed in fake social proof. Here the personas are labelled as composites on the page itself, the dashboard is captioned as an illustration, and pricing carries its pre-launch note. Trade-off: the page loses fake credibility signals and instead earns the durable kind.
Intended impact
The product hasn’t launched, so there are no conversion numbers to report — and none will be invented. The design goals it will be measured against:
The build itself — the shader, the identity system, the accessibility and performance work — is documented as a teaching artifact in the build guide.
Process pages are arguments; the site is the evidence. See whether the decisions above hold up in the cloth itself.