Sacks of maize stacked in a warehouse
  1. Home
  2. Our work

Our work from field to dispatch

Photo: Cut angles, CC BY-SA 4.0

The tool replaces paper intake, storage and dispatch notebooks with short smartphone and tablet forms that work offline and sync when there is signal. Every intake gets a batch code, so each bag can be followed from the farmer who supplied it to the buyer who received it.

This is a proof-of-concept implementation study, not a large impact evaluation. It tests whether the tool works in a real warehouse and what it would take to scale it, through five work packages over twelve months.

How a batch is traced

  1. Step 1

    Register the source

    Farmer or supplier, location, contact and the crops they bring.

  2. Step 2

    Record the intake

    Date, crop, quantity, moisture reading and a visual quality check.

  3. Step 3

    Code the batch

    A unique batch code or QR label links the bags to their source, date and quality records.

  4. Step 4

    Monitor storage

    Storage location, time in store, periodic checks, pests and condition notes.

  5. Step 5

    Dispatch to buyers

    Buyer, date, quantity and batch codes, plus any rejection feedback.

Three layers of support

Layer 1

Digital records

Captures the essentials at each step, with as few required fields as possible and validation prompts to keep records complete.

Creates traceability at the point where records currently break down.

Layer 2

Rule-based risk flags

A dashboard flags high moisture, long storage, missing source information, visible spoilage and poor storage conditions.

Gives operators practical warnings without needing complex AI.

Layer 3

Machine-learning proof of concept

An interpretable model, such as logistic regression, a decision tree or a random forest, trained on pilot data to score batch-level spoilage or quality risk.

Tests how structured records can support AI-enabled food-safety decisions, with its limitations reported openly.

  • Works offline
  • Smartphone and tablet friendly
  • Short forms with visual cues
  • Open source, no licence fees
  • Paper backup during the switch-over

Five work packages over twelve months

Warehouse staff, farmers, suppliers and buyers take part from the start as co-designers, testers and evaluators.

Work package
M1M2M3M4M5M6M7M8M9M10M11M12

WP1Baseline and co-design

Map current intake, storage and dispatch practice, interview 15–20 farmers and suppliers, estimate the cost of avoidable losses and co-design the workflow.

Months 1–3

WP2Tool development

Build the offline-capable mobile and web tool, with a user guide and data dictionary.

Months 3–5

WP3Pilot at Bvumbwe

Run the tool with real produce and suppliers: register farmers, record batches, link them to dispatches and log what works.

Months 6–9

WP4Risk model

Clean the pilot data, test an interpretable spoilage and quality risk model, and add a risk indicator to the dashboard.

Months 8–10

WP5Training, evaluation and dissemination

Hands-on training, evaluation, a business case and learning note, a stakeholder event and the open-source release.

Months 6–12

Pilot targets

What the pilot sets out to show at the Renai Farms warehouse in Bvumbwe.

Target20–40

farmers and suppliers registered and linked to their produce

Project plan
Target100+

produce batches recorded with source, intake, quality and storage information

Project plan
Target70%

of batches with complete records from intake to dispatch

Project plan
Target5 min

or less, on average, to record a batch after training

Project plan
Target15+

buyer dispatches linked to traceable batch records

Project plan
Target80%

of qualifying risk conditions flagged by the system

Project plan

These are targets from the project plan, not results.

Open software, protected data

  • The software and its documentation will be released under an open-source licence, so other warehouses, cooperatives and SMEs can adopt it without vendor lock-in.
  • Confidential supplier and transaction data will not be published.
  • Taking part in interviews and feedback is voluntary and based on informed consent.
  • Personal identifiers are kept to a minimum, and access to data is role-based.
Hands entering data into an ODK Collect form on an Android phone

Data collection portal

Field surveys and pilot data are collected with ODK, open-source data collection that works offline on Android phones and tablets, and managed on the project’s own ODK Central server.

Sign in to ODK Central

Access is limited to project staff and enumerators with an account.

Photo: ODK Collect in use in another programme. President’s Malaria Initiative, public domain