Neuralis

Free farm inputs can reduce a farmer’s cost, but they do not tell her which product fits the problem, how to use it safely, or when to stop and ask an extension officer. Safe guidance has to travel with the sachet, in language the farmer understands and with clear limits on what the advice can confirm.

At sunrise, Abena stands at the edge of her cocoa plot with a free fungicide sachet folded in her palm. The leaves are wet, her youngest child is still asleep at home, and dark marks on a few pods have made her worry that rain will spread the damage through the plot.

A neighbour told her the sachet would help. Another said to mix it strongly so it “works quickly.” Neither had seen the pods up close. Abena has one chance to decide what goes into the sprayer that morning. A wrong product, wrong concentration, or unsafe timing could damage the crop, expose the person spraying, or leave her with a harvest she cannot safely sell.

The free input is real help. The missing instruction is where the risk begins.

A sachet cannot answer the question in front of the farmer

For the 2025/2026 season, Ghana announced free fertilizers, insecticides, spraying machines, fungicides, and flower inducers, alongside rollout of a plot-to-port Cocoa Traceability System. Those inputs can matter deeply when the cost of buying them would otherwise delay action.

But a fungicide packet does not identify a disease. It does not explain whether a pod mark needs treatment, removal, a field visit, or a different response entirely. It does not tell a farmer what dosage applies to that exact product or how long to wait before re-entering a sprayed field or harvesting.

That gap often gets filled by memory, a neighbour’s advice, or a voice note forwarded through a group chat. The advice may sound confident. Confidence is not evidence.

Abena turns the sachet over in her hand and asks the question that changes the situation: “What exactly is this for?” If nobody can answer clearly, spraying feels like action while leaving the central risk unresolved.

The most responsible answer may be, “Do not mix it yet.”

That can feel frustrating at sunrise, when the field is waiting and the product is already in hand. It is still better than offering a plausible instruction that has never been checked against the product, the crop condition, and the safety requirements.

Safe guidance needs a route to a real person

Agricultural guidance needs two paths. The first handles familiar, reviewed questions in language a farmer can follow. The second hands uncertain or high-risk questions to a named person who can take responsibility for the next step.

That is the model AgriVoice is preparing to test with Asante-Twi-speaking cocoa farmers. It is designed to receive a voice question, select from reviewed guidance, speak the answer back, and escalate when the answer is missing or uncertain. It does not generate agronomy advice from scratch.

Chemical questions require the strictest boundary. If the reviewed content lacks a verified dosage, re-entry interval, or pre-harvest interval, the system should withhold an instruction and send the question to an extension officer. Silence is a failure when it leaves a farmer stranded. A clear escalation, with an accountable person at the other end, is a safer response.

That is why a named extension officer matters. “Ask someone” places the burden back on the farmer. “Your question has been sent to the extension officer responsible for this pilot” creates a route, an owner, and a reason to expect a reply.

The pesticide detail that stops AgriVoice from answering explains why withheld chemical guidance is a safety measure, not a missing feature.

Language changes whether the instruction can be used

A written label may be technically correct and still fail at the moment a farmer needs it. A farmer standing in a cocoa plot needs to understand the instruction while holding a product, considering weather, and deciding who will spray.

That calls for spoken guidance in the language people use to describe their crop and their problem. It also calls for careful review. Machine translation can produce fluent sentences while swapping a domain word. In cocoa farming, one changed noun can turn safe advice into the wrong advice.

For that reason, AgriVoice’s pilot depends on spoken Asante Twi reviewed by a farming-aware translator. The system also needs an agronomist to clear the chemical guidance that remains on hold. Those are not polish tasks for later. They determine whether an answer should reach a farmer at all.

Abena’s question may begin with a few seconds of speech. The work behind a safe answer is much longer: reviewed content, a language check, a decision about uncertainty, and a human escalation route when the system cannot responsibly continue.

Make distribution carry accountability

Giving out inputs creates a moment of trust. Farmers reasonably assume a product supplied through a programme comes with enough guidance to use it without harm. Distribution should meet that expectation with a practical support plan.

A cocoa-sector partner can make that plan real by pairing input distribution with a short explanation of what each product is for, how farmers can ask questions in Asante Twi, and who receives questions that need expert review. The first pilot should measure whether farmers understand the answers, whether questions are answered or correctly escalated, and whether the escalation owner responds.

By late morning, Abena has put the sachet aside rather than mixing it on a neighbour’s instruction. She has recorded a voice question describing the pod marks and the product in her hand. If the guidance cannot identify a safe, reviewed next step, the question goes onward instead of becoming a guess in the field.

That is the standard free inputs deserve: a product, an understandable instruction, and a person accountable when the instruction runs out.

Neuralis

AgriVoice helps Asante-Twi-speaking cocoa farmers ask farming questions by voice and receive answers assembled only from agronomist-reviewed content, with human escalation when the system is unsure.

Try Neuralis

Comments

No comments yet.