AgriVoice needs a named cocoa extension officer because uncertain farming questions require accountable human follow-up, not a confident automated reply. In the pilot, that officer is the person who receives escalations, decides what advice can safely continue, and closes the loop with the farmer.
In 1970, Apollo 13’s crew faced rising carbon dioxide inside the lunar module. The available square command-module canisters did not fit the lunar module’s round openings, and Mission Control in Houston had to devise an adapter from materials already aboard. Gene Kranz’s team worked while the outcome remained uncertain. NASA’s history of Apollo 13 and Jim Lovell and Jeffrey Kluger’s Lost Moon document how the crew returned safely after that improvised solution was relayed and used.
The useful part of that story is not the duct tape. It is the handoff. A problem exceeded the equipment’s prepared answer, so a named team took responsibility for the next decision. AgriVoice needs the same structure on a much smaller, everyday scale: reviewed answers for questions it can handle, and a real extension officer for the questions it cannot.
The first morning has one clear operational test
A two-week AgriVoice pilot with 20 to 50 Asante-Twi-speaking cocoa farmers begins with more than a WhatsApp number and a voice interface. It begins with an extension officer opening WhatsApp and seeing the escalation path they own.
Some farmer questions should receive a reviewed spoken answer immediately. A question about a familiar cocoa disease symptom may match a reviewed content block. A question with unclear audio, mixed language, missing details, or a topic outside the approved content should be marked uncertain and passed on.
The officer needs a practical queue, not a vague promise that “someone will respond.” Each escalation should show the farmer’s question, the system’s uncertainty reason, the relevant reviewed content if any, and a way to return a response through the same channel. The pilot scorecard sets a target of a median response in less than one working day. That target turns accountability into something the partner and Neuralis can measure.
This is where the institutional workflow becomes visible. A farmer does not need to understand the difference between speech recognition, content selection, and escalation logic. They need to know what happens when the answer is unclear. The answer should be simple: their question reaches the extension officer responsible for this pilot.
Chemical advice must stop before it becomes unsafe
The most important escalation cases involve pesticides. Three AgriVoice chemical blocks remain withheld because dosage, re-entry, and pre-harvest intervals have not been verified by a Ghanaian agronomist. The system should refuse to fill those gaps with generated advice.
That restraint may feel slow in the moment. A farmer with a sprayer ready wants a direct instruction. Yet an unverified dosage or timing recommendation can create a much bigger problem than a delayed reply. The officer’s role is to make the pause useful: confirm the formulation, ask for the missing label details, consult approved guidance, or tell the farmer clearly to wait.
This is the operational lesson behind [Yaw’s pesticide question](\/blog\/yaw-s-pesticide-question-cannot-be-guessed-an-extension-officer-must-respond-8b18c23f\/). Safe escalation depends on a person who has been named before the question arrives. Assigning responsibility after a risky message appears leaves the farmer waiting and the institution guessing who owns the decision.
The pilot should record whether the escalation was correct, how long it waited, whether the farmer understood the response, and whether the case was resolved. These records matter as much as successful automated answers. An automated system that recognizes its boundary is doing useful work. The next test is whether the human side honors that boundary.
What the partner is actually agreeing to test
A cocoa-sector partner is not merely helping recruit farmers. The partner is testing a service model with Neuralis.
Before farmer exposure, the pilot needs reviewed Asante Twi content, a working WhatsApp message path, consent and deletion behavior that staff have tested, and real farmer voice questions for measuring speech recognition. The partner also needs to name the extension officer who receives escalations. Without that person, the pilot has no safe route for uncertainty.
During the two weeks, the team can measure whether at least 70% of questions are answered or correctly escalated, whether farmers understand the spoken response, and whether they return to ask again in the second week. The pilot also measures end-to-end response time and the cost per completed question. Those are evidence questions, not launch claims.
The named officer provides a check on the system’s most important claim: that it can help farmers get a useful next step in their own language without pretending every question has an automatic answer.
A queue is a promise with a clock attached
Apollo 13 did not succeed because the spacecraft contained every answer in advance. It succeeded because Mission Control could recognize a new problem, work within known constraints, and send back a usable instruction. The situation was extraordinary; the workflow principle is familiar.
For AgriVoice, the first morning’s WhatsApp queue should make that principle concrete. The officer should be able to see what needs attention, respond within the agreed service window, and record whether the answer resolved the question. A farmer should receive either reviewed guidance or a clear human follow-up path.
That is the institutional workflow worth testing before any broader release: one named person, one visible queue, one accountable response for every uncertain answer.
Comments
No comments yet.