A reviewed answer should end the moment a farmer needs product-specific dosage, re-entry, or pre-harvest advice that has not been cleared by a Ghanaian agronomist. At that point, AgriVoice should say what it can safely say, explain what is missing, and pass the question to the named extension officer.
The moment uncertainty became the important fact
In 1854, John Snow was investigating a severe cholera outbreak around Broad Street in Soho, London. The evidence he gathered pointed toward the public water pump, yet the cause of cholera remained contested. The pump handle was removed, stopping people from drawing water while the danger was still being understood.
The John Snow Archive and Research Companion documents the investigation, including Snow’s mapping of deaths near the Broad Street pump. His response did not require him to pretend he knew every detail of the outbreak. It required him to act on the part supported by evidence and prevent people from taking a risk that could not be undone.
A pesticide question can carry the same shape of decision. The stakes differ. The principle does not: when evidence runs out, the system must stop giving instructions that could cause harm.
Where a reviewed cocoa answer can safely help
Imagine a cocoa farmer sends a Twi voice message: a pod has changed colour, leaves look damaged, and a neighbour recommends a named pesticide.
AgriVoice can begin with the reviewed guidance it has. It can help the farmer describe the visible problem, identify safe immediate actions already approved for that situation, and explain that a product recommendation from another farmer is not enough to establish the right treatment. It can also surface any reviewed safety content that applies before spraying, such as following the product label and avoiding improvised mixing.
That response has a clear purpose. It helps the farmer avoid a rushed decision while preserving the facts an extension officer needs to assess the case.
The boundary arrives when the farmer asks, “How much should I mix?” It also arrives when they ask when they may return to the field, whether spraying is safe before an upcoming harvest, or whether a named product fits their particular crop condition. Those details depend on the actual product, formulation, label, crop stage, local conditions, and approved safety intervals.
Three chemical blocks in the current AgriVoice content remain withheld for precisely this reason: they lack verified dosage, re-entry, and pre-harvest intervals. Silence at that point would leave a farmer stranded. A confident answer would be worse.
The handoff needs an accountable person
A safe handoff is more than “ask an expert.” The pilot requires a cocoa-sector partner to name the extension officer who receives escalations. The system can pass along the farmer’s question, relevant context, and the reason it could not give a final answer. The farmer should hear that their question has been escalated and understand what information may help the officer respond.
This is where a voice workflow earns trust through restraint. It does not turn uncertainty into polished Twi. It names the uncertainty plainly.
For a farmer, the practical response may sound simple: do not mix or spray this product until the extension officer confirms the dose and timing. Keep the product label available. Share the crop symptoms, the product name, and the planned harvest date if known. That is useful guidance because it prepares the next decision without supplying unverified chemical instructions.
The handoff also creates an operational obligation. A named owner must receive the escalation, and the pilot measures whether those questions are answered within one working day. A queue without an accountable person is only a polite refusal.
For a related example of why missing ownership changes the safety outcome, see What Happens When an Unverified Chemical Question Has No Named Owner?.
Build the boundary before asking farmers to trust it
The reviewed-answer boundary should be tested before farmer exposure. Staff need to run a complete question through the system and confirm that no unsafe answer reaches the user, the escalation reaches the correct person, and the response remains understandable when spoken aloud in Asante Twi.
The pilot’s success measure reflects that responsibility: zero unsafe pesticide answers. A question answered safely through escalation counts as a successful outcome. A fluent but unreviewed dose does not.
Snow’s 1854 investigation is remembered for the pump handle, a concrete intervention that stopped a dangerous action while the evidence was still being worked through. AgriVoice needs its own equivalent: a firm, understandable stop at the exact point where reviewed guidance ends, followed by a real person who can take responsibility for what comes next.
Comments
No comments yet.