Featured case study
Lead Product Designer

Designing map based intelligence workflows

Turning crop production and crop performance data into spatial understanding, explorable from national food security policy down to individual field conditions

At a glance

5 monitoring contexts shipped on the same pattern

Crop Production, Fertiliser Subsidy, Water Stress, Crop Health, and Soil Exposure, each a distinct surface, all driven by the same spatial layer, navigation, and drill-down model.

From 5 months to 2 months per new context

First context (Crop Production) took five months from design to ship. By the fourth, the pattern was stable enough to ship a new context in two, the system became its own multiplier.

Used across 3 country governments · ~8 sessions/day

Active in production with food security advisors and ministry analysts across three governments, averaging eight monitoring sessions per day.

Scale of the issue

A country's food system consists of many variables

Food security is the outcome of many conditions, tracked separately, by different teams, across different timescales, with no shared view.

The key variables:

  • Crop production & performance
  • Weather & rainfall
  • Soil quality
  • Input subsidies & seed quality
  • Water stress
  • Crop health
  • Infrastructure
  • Population & food consumption

Countries can access some of this data, through various ministries and regional offices. What didn't exist was a way to see all of it in the same place.

The result was that decisions were made on guesswork. Officials requested data from regional offices, waited weeks for responses, received inconsistent formats, and manually reconciled everything into a picture already out of date. No one could see how drought in one region connected to food security risk in another. The numbers existed, but they couldn't be read together.

Agricultural system, the interconnected data layers, from atmosphere and weather down to groundwater and bedrock
I
User Goal I As a food security advisor
I need a clear, actionable picture of the food system status and intervention options

So that decision-makers can act on a reliable, current picture of food security, knowing which regions are at risk, which indicators are deteriorating, and what the intervention window is. Without this, decision-making is reactive: action only follows after a crisis has already materialised.

From Agriculture Intelligence, Visual Explorer is the surface that responds to this goal.

Design process
I · Frame Spot the fragmentation
II · Define Shape of data + criteria
III · Design V1 ships, then breaks
IV · Iterate Re-render at field scale
V · Scale One pattern, every context
M · 01 The fragmentation Crop data scattered across ministries, formats, timescales. No spatial picture, no shared view.
M · 02 Shape of data Mapped the analytical requirements — behaviours, frequencies, information volumes — before designing anything.
M · 03 Success criteria Benchmarks set against current state: time to gather, time to analyse, confidence in output.
M · 04 V1 ships Provincial block colours. Clean from national view. But averaging hid the failing maize underneath.
M · 05 Epistemically dangerous A province that's 80% performing reads "mostly fine". Averaging is the wrong abstraction.
M · 06 V2 tiles Sentinel-2 native 10m resolution rendered directly. Every field shows its own signal, no smoothing.
M · 07 Five contexts, one pattern Crop Production, Fertiliser, Water Stress, Crop Health, Soil Exposure — same framework, different surfaces.
M · 08 5 months → 2 Pattern became its own multiplier. First context took five months; the fourth took two.
Shape of data

Understanding what the analysis actually demands

The problem we needed to solve was how to bring together the many dimensions that shape food security and show how they connect. Government officials, ministry teams, and national advisors needed to track those dimensions over time, so shifts could be recognised early enough to act on while there's still a window.

Before designing anything, we mapped what that visibility actually requires, the behaviours, frequencies, and information volumes the system had to support for a user like Mayeso.

An analyst spends weeks gathering and reconciling data, then has days — not weeks — to act on it. Six numbers capture the cost of that gap, and what the system had to compress.

6 wks
Time spent looking for data, typically
3 wks min · ∞ max
2 wks
Time analysing the data once gathered
5 days min · 1 month max
Bi-weekly
How fresh the crop data needs to be
weekly min · real-time max
5 yrs
Years of historical data needed to spot trends
3 yrs min · 20 yrs max
5
Key crops to track across the country
1 min · 50 max
All
Regional levels to drill into for decisions
national · provincial · district · ward · field
Design for scale

How do you build a pattern that replicates?

Each monitoring context had its own question, its own visualisation, and its own level of detail depending on who was looking at it. Building for one context first would have locked the system into a shape that didn't extend. The goal was a replicable framework, something that works for crop production and works equally for water stress, crop health, and any context added later.

The data has to move across four dimensions at once. Build for one and the system locks in. Build for all four and the surface becomes legible at every scale.

01
Multiple contexts

Crop production, water stress, fertiliser, health — each with a different question, visualisation, and audience.

02
Administrative depth

The same signal reads differently at national, district, and field level — resource allocation vs deployment vs irrigation.

03
History

Every context has a temporal axis — week-over-week, season-over-season. Static views can't hold it.

04
Within-context depth

Aggregate signal, per-crop breakdown, field-level anomalies — the hierarchy doesn't stop at the administrative boundary.

We explored four ways to structure the two primary axes — contexts and administrative levels. Approach D won because it kept the country itself as the constant: contexts switch as overlays on top, so the user never loses the geographic frame.

Approach A, Contexts as columns, levels as rows
Visual Explorer
Crop Prod.
Water Stress
Crop Health
+more
National
Provincial
District
Field
All combinations visible simultaneously. Any context, any level, immediately comparable across the full surface.
Approach B, Context sidebar, level toggle
Visual Explorer
Crop Prod.
Water Stress
Crop Health
Fertiliser
Pest Det.
Deforestation
Natl.
Prov.
Dist.
Field
One context at a time. Sidebar for context selection; tab strip to switch administrative level within the detail view.
Approach C, Administrative level as primary tab
Visual Explorer
National
Provincial
District
Field
Crop
Water
Health
Fert.
Pest
+more
Enter at your level, national, district, farm. Contexts appear as a secondary selection within that level.
Chosen
Approach D, Map as primary canvas
Visual Explorer
Map
Crop Prod.
Water Stress
Crop Health
The map is the entry point. Context selected via overlay panel; administrative level changes through zoom or a layer control.
Approach

One framework, every context

Each context — production, subsidy, water stress, crop health — answers a different domain question, but the design framework is the same. The same approach, repeated for every context the platform supports.

Crop production

Context 1: Crop production & performance

Crop production is the base layer of food security. How much a country produces, and how that performance shifts over time, is the bedrock every other agricultural decision sits on. Before this, that picture wasn't visible at the regional level — officials had to wait for end-of-season data instead of seeing it as it formed.

What we set out to do was forecast and monitor production in season, so teams could plan ahead for the kinds of decisions that only get made when the picture is visible early enough to act on:

What this data panel answers
  • Production estimate this season, by crop?
  • Production trend YOY, by crop and province?
  • How does this region compare to neighbours?
  • Where is the crop underperforming the baseline?
  • What insights should I act on?
What the map shows

Where each crop is grown across the country and how each is performing against the others. One visual reference for nationwide context.

Fertiliser subsidy

Context 2: Fertiliser subsidy monitoring

Fertiliser subsidies are government interventions designed to make fertiliser more affordable, particularly for smallholder farmers in low-income countries. The aim is to lift agricultural activity, raise crop production, and through that, strengthen food security. Kenya, for example, has spent approximately US$310 million per year on the programme for nearly a decade.

Whether the spending actually delivers, however, is debated. Some programmes show real agricultural lift. Others struggle against:

What we set out to do was surface the data, so the link between fertiliser distribution, farmer participation, and crop outcomes can finally be seen in one place. With that visibility, programme officers can see where the subsidy is working, where it isn't, and make better decisions ahead of next season.

What this data panel answers
  • Is participation rising or falling, by region?
  • Which counties need attention, which are improving?
  • Which fertiliser types are in highest demand?
  • Does subsidy spend correlate with yield and soil health?
  • How is collection trending across the period?
What the map shows

Distribution of warehouses and farmer locations. Shows who applied for the subsidy, who collected it, and who hasn't applied at all, so officers can target action.

Satellite contexts

Context 3: The satellite-derived monitoring signals

Beyond government data, Visual Explorer holds a family of satellite-derived monitoring signals: Water Stress, Crop Health, Nutrient Deficiency, Canopy Structuring, Pest & Disease, Deforestation, Degrading Land, Soil Exposure. Eight contexts, one scaffolding. The design challenge wasn't which spectral indices to use — that's an agronomy decision. It was deciding which signals a ministry official can act on directly, and which need a data specialist to interpret. That distinction shapes what the map shows.

What this data panel answers
  • Where are crops moisture-limited, and how severe?
  • Single-pass anomaly or persistent pattern?
  • Also showing health decline or nutrient deficiency?
  • How does this season's stress compare YOY?
What the map shows

NDWI and MSI indices rendered as a 10m gradient tile map. Stress is classified mild, moderate, or severe per location, with pass-by-pass trends and season-over-season comparison.

Visual evolution

From averaged regions to the data as it really is

The first version of the map averaged everything out by region. Each province was rendered as one colour based on its overall average. The engineering wasn't ready to show finer detail yet, so this was the closest we could get to a workable view.

Feedback from extension workers in the field and ministers in the capital surfaced the same issue: averaging hid the parts of the country that were actually struggling. A region that was 80% doing fine and 20% severely stressed looked, on the map, mostly fine. The places that needed attention disappeared inside the average.

So we rebuilt the rendering so the map shows the data the way it actually exists, location by location, instead of summarised into a single regional colour. Stressed fields show as stressed, healthy fields as healthy, regardless of what's around them. Drag the slider to compare.

After: data rendered location by location across the landscape
Before: each region rendered as a single averaged colour
V1 · Before V2 · After Drag to compare

The shift made the map functional across every level of decision-making. An extension worker on a single farm and a minister looking at the country as a whole now see the same data, rendered at the resolution that matters for their work.

Closing

The real shift

Food security is multifaceted, but one way of approaching it starts with surfacing the data and showing how everything connects. That's what Visual Explorer was built to do.

Everything that came after, subsidies, stress detection, AI-generated insights, scenario planning, only worked because this foundation was in place.

Against the benchmarks we set at the start, the shifts looked like this:

Time to gather data
2 wks → 1 day
Time to analyse + recommend
3 days → 1 day
Confidence in the data
1 / 5 → 5 / 5
Decision-maker confidence in recommendations
1 / 5 → 5 / 5
Ease of use
2 / 5 → 4 / 5
Analyses completed without external validation
2 / 5 → 4 / 5

"I don't have to wait for the next report to know what's happening. If a drought is building in my district, I see it early and advise my farmers in time. I no longer rely on assumptions, I act on what's actually on the ground."

Farm extension worker
Parent project Designing decision intelligence for agriculture Next project Designing trust into government data ingestion