docs guides case studies

Case studies & pilots

Real deployments, evaluation pilots, and “what actually happened” stories when FildraAI was used in fields, not just in demos.

SECTION Case Studies / Overview FOCUS Outcomes, lessons, and constraints
Sally — FieldGuide early adopter and maize farmer, Zambia

From the Field

Real farmers. Honest outcomes.

These case studies are built on field interactions with smallholder farmers across Zambia. Every pilot includes what worked, what didn't, and what it means for how we build. We publish these not to showcase success, but to be accountable about where we are.

Featured pilots

These summaries are designed to be honest — highlighting both impact and limitations — so partners can see where FildraAI fits and where careful guardrails are required.

Zambia · Maize · Smallholders
Image-based maize disease diagnosis in a mixed-variety landscape

FieldVision was used by field staff to capture leaf photos in multiple districts. The pilot focused on Ground Truth collection, model error analysis, and explaining AI focus area outputs to agronomists.

Approx. 500+ image sessions · multi-variety · low- to medium-connectivity settings.

topic: computer vision focus: error analysis
East Africa · Maize · Advisory teams
FieldGuide for structured advisory conversations

Advisory staff used FieldGuide to turn model outputs into farmer-facing plans, with local supervisors reviewing safety language, PHI/REI, and economic feasibility.

200+ advisory sessions · multiple advisors · emphasis on responsible use and verification.

topic: advisory workflows focus: safety language
Multi-country · Program design
Building country packs with national experts

FildraAI country packs were co-created with agronomists and regulators, focusing on PHI/REI, label-aligned recommendations, and explicit “gaps” where local data was missing.

3+ country packs · multiple crops · iterative review cycles with local partners.

topic: knowledge base focus: governance

How to read these case studies

Each case study tries to answer the same core questions: who used the system, what changed, what went wrong, and what we would do differently next time.

  • Context: crops, geography, season, and existing advisory or extension workflows.
  • Setup: which modules (FieldVision, FieldGuide, FieldKB, FieldMap) were deployed and how.
  • Outcomes: what worked well, where farmers or staff saw value, and measurable improvements where available.
  • Risks & limitations: failure modes, boundary conditions, and mitigation steps (including when to turn the system off).

Note: Some case studies are anonymized or partially generalized to protect farmer, partner, or staff privacy, while still preserving the technical and operational lessons.

View related guides Responsible use Discuss a new pilot

Upcoming case studies

Some deployments are still in progress. We list them here so partners know what is coming, and so that we don’t over-claim results before they are ready.

Southern Africa · Maize
Multi-district evaluation of image quality & model robustness

Focus on how real-world phone cameras, lighting conditions, and partial symptoms affect model performance, with structured feedback from agronomists.

status: in design
East Africa · Advisory
Comparing AI-supported advisory vs standard practice

Experiment comparing advisory quality, consistency, and farmer understanding with and without FildraAI support, under strict safety rules.

status: planned