Design-thinking case study

Process

The brief

What might Africa — and each of its 54 UN member states in Africa — become by 2031 and 2036, and what decisions could produce those futures? Build something that lets people explore plausible futures, understand uncertainty, and recognise the signals that move a country toward one outcome or another. Cover all 54 from the first version. Never fabricate.

The problem, stated precisely

The design problem is not "visualise foresight data". It is a contradiction:

The subject demands confidence — nobody engages with a platform that hedges everything. The evidence demands humility — nobody should trust a ten-year national projection presented crisply. Most products in this space resolve the contradiction by choosing: confident and misleading, or hedged and useless.

Everything good about this design comes from refusing to choose, and instead making the uncertainty itself the thing being visualised.

Directions considered and rejected

Rejected: the foresight dashboard

Cards, KPI tiles, a filter sidebar, a chart grid. It is the default, it would have been faster, and it fails the brief in a specific way: a dashboard's visual grammar asserts that everything on screen is equally solid. It has no way to say "this figure is from 2019, from a country with three years of missing data, and you should lean on it less". Rejected because the format cannot carry the honesty the subject requires.

Rejected: the cinematic scroll narrative

A scroll-driven film through Africa's futures. It would have looked spectacular and reviewed well. But a linear narrative makes the author's argument the product. The brief asks for something people explore and question — and a country appearing at scroll position 4,200 is not explorable. Rejected because it would have made a beautiful case for one future rather than an instrument for examining many.

Rejected: the 3D globe

Rejected quickly. A rotating globe hides half the subject at all times, costs a WebGL dependency and a large payload, performs badly on the mid-range Android hardware much of this audience uses, and — most decisively — communicates "technology product" when the requirement is "credible institution".

Chosen: the instrument

Something you look through at a subject, whose readings change as you adjust it. Instruments have an honest relationship with uncertainty: a good one shows you its own precision. That single metaphor generated the lens control, the time spine, the scenario switch, the annotation-style readouts — and the stipple.

The decision the whole design turns on

Confidence is drawn into the map. Every country carries a stipple whose density tracks the strength of its evidence base. Countries the platform knows well render clean; countries it knows poorly render visibly grainy.

This does four things at once, which is how you know it is the right idea rather than a nice one:

• It makes uncertainty impossible to skip. You cannot scroll past it — it is in the shape of the thing you came to look at.
• It replaces disclaimer copy, which nobody reads, with a visual property, which everybody perceives.
• It reveals the continent's real research geography — where the world has bothered to measure carefully and where it has not — which is itself a finding about power, not a technical detail.
• It costs four SVG patterns and one line of assignment.

Its companion decision: countries with no data are drawn as a dashed void, never as a pale shade. Pale reads as "low". Absent should read as absent.

Key decisions and trade-offs

Scenarios calibrated to each country's own history

The credibility problem in foresight is that "optimistic scenario" usually means "what an outsider imagines this country could become". Calibrating Acceleration to mean-plus-0.85-standard-deviations of the country's own recorded growth makes the band reproducible, comparable across 54 countries, and structurally impossible to inflate.

The trade-off, stated plainly: it means the Disruption band cannot represent a war, a default or a climate catastrophe, because those are not in the recent record. Rather than hide that, the method page states it as a limitation and the platform surfaces those risks as named, unquantified warnings. A narrower honest claim beat a wider dishonest one.

Equal-area projection

Web Mercator inflates high-latitude land and shrinks equatorial land. It is the most common way Africa is misrepresented at a glance, and it is usually nobody's decision — just the default. Choosing Lambert Azimuthal Equal-Area cost about ten lines of maths and is the first argument the platform makes, before a word is read.

Evidence and interpretation set in different typefaces

Measurements are set in the interface sans. The platform's own judgements are set in serif and marked interpretation. It is a small typographic decision doing ethical work: a reader can tell what is measured from what is inferred without reading a legend.

Static briefs as the canonical record

Every country has a static HTML page — no scripting, a few kilobytes, prints cleanly. Treating the lightweight version as canonical rather than as a fallback is a position about who this is for. A platform about African futures that only works on a fast connection and a recent laptop has already answered a question it should have asked.

Showing the seams

Research depth is uneven. The honest options were to hide it, to delay until it was uniform, or to publish it. The platform marks each country reviewed or baseline, and baseline countries say plainly that no national plan or budget has been read for them yet. This costs polish and buys the only thing that matters here: a reader can calibrate how much to trust each page.

African-led, as a set of concrete decisions

"African-led perspective" is easy to claim in copy. These are the places it changed the artefact:

• No aggregate "Africa". There is no continental average anywhere. The unit is the country, and the Weave shows the relationships between countries rather than dissolving them into one story.
• The AI reading refuses the standard frame. For a country where most employment is agricultural, "AI will displace white-collar work" is simply not the relevant question. Those readings say the risk is exclusion, not redundancy — and ask who owns the compute, not whether adoption is coming.
• Progress is not defined as convergence. Levers name domestic revenue mobilisation, post-harvest storage, collateral registries and land reserved before settlement arrives — the things that actually bind — rather than foreign investment attraction.
• Remittances are treated as serious economic infrastructure, not a footnote to FDI, because for many of these countries they are larger and more reliable.
• Contested facts stay contested. CEN-SAD's membership roll could not be verified across official sources, so the platform publishes an empty list and says why, rather than repeating a number it cannot stand behind.
• Western Sahara is drawn and named as a UN Non-Self-Governing Territory with disputed sovereignty — neither erased from the map nor absorbed into a neighbour.

What this demonstrates

As a piece of work, the capability on show is not "can build a map". It is the harder thing: taking a brief whose central risk is fabrication, and building a system that is architecturally incapable of it — where the interpretation layer fails closed without data, where gaps propagate to the interface as gaps, and where every projection carries the arithmetic that produced it.

Design decisions and engineering decisions are the same decisions here. The stipple is a visual idea and a data-model idea. The typographic split between evidence and interpretation is a style rule and a schema rule. That is the argument this project makes about designing with AI: the value is not in producing more surface faster, it is in holding a position about what the artefact is allowed to claim — and then building so it cannot claim more.

Attribution and tooling

Africa 2036 Intelligence is an initiative by Orvantis Intelligence, created and directed by Hannah Kwakye.

The implementation — data pipeline, foresight engine, cartography, interface and prose — was built with Claude Opus 5, working under that direction. No part of it was delegated to another model.