The Named Extension Officer AgriVoice Needs, and What an Unanswered Question Could Cost

A Ghanaian cocoa institution should recruit 20 to 50 farmers for a two-week AgriVoice pilot only if it names an extension officer to receive unresolved questions and commits to a working response path. The pilot measures whether voice advice is useful and safe in practice, including whether uncertain questions reach a person quickly enough.

A safe system needs a human owner

In April 1970, Apollo 13’s crew faced a damaged spacecraft after an oxygen tank explosion. Jim Lovell, Jack Swigert and Fred Haise were far from Mission Control in Houston, and the carbon dioxide problem became urgent because the lunar module’s filters did not fit the command module’s system.

The documented account in Jim Lovell and Jeffrey Kluger’s Lost Moon describes the ground team working with the materials available to the crew and devising an adapter. The crew used it, and Apollo 13 returned safely to Earth.

The lesson is practical: a technical system can identify a problem, but someone still has to own the response when the normal path cannot continue. AgriVoice has the same boundary at a much smaller scale. It can select from reviewed cocoa guidance and speak it in Asante Twi. It must not guess when a farmer asks about an uncertain disease, a pesticide dosage, re-entry after spraying, or a harvest interval.

That question needs an accountable person.

The decision the partner is making

The briefing should begin with the real request. Neuralis is asking the institution to recruit a defined cohort of 20 to 50 Asante-Twi-speaking cocoa farmers for two weeks, then help judge whether the service deserves one focused iteration, continuation, or a stop decision.

The institution also decides whether it can name the extension officer who receives escalations. This cannot be a shared inbox with no owner. A farmer who sends a voice note about a suspected cocoa problem needs a clear route when AgriVoice cannot safely select an approved answer.

The partner should leave the briefing knowing what AgriVoice will do:

• Receive farmer questions through WhatsApp voice notes.
• Recognize the speech, select reviewed content, and return a spoken response where the answer is supported.
• Escalate missing, unclear, out-of-domain, or safety-sensitive questions to the named officer.
• Record pilot measures that show where the service helped, where it failed, and where human support carried the decision.

It should also know what the pilot will not do. AgriVoice does not author agronomy advice. It does not release pesticide instructions that lack verified dosage, re-entry, or pre-harvest information. It does not turn a two-week test into a public launch.

The measures that make the briefing concrete

A pilot can sound promising while producing no evidence. The briefing needs a scorecard before recruitment starts.

For AgriVoice, the central measures are straightforward: successful answers or correct escalations, zero unsafe pesticide answers, farmer comprehension, repeat use in the second week, escalation response time, end-to-end response time, and cost per completed question.

A useful target is at least 70% of questions answered or correctly escalated. The safety threshold is stricter: no unsafe pesticide answer may reach a farmer. If the voice routinely prevents understanding, or confident wrong answers appear, the institution and Neuralis should stop or redesign the workflow rather than recruit another group.

The named extension officer is part of the measurement, not an extra support detail. The pilot aims for a median escalation response in less than one working day. If questions sit unresolved because nobody knows who owns them, the pilot has found an operational failure that better speech recognition cannot repair.

This is why the conversation belongs beside existing work on what happens when a cocoa question needs a human answer. The escalation path is the service promise when reviewed content reaches its limit.

What the institution should confirm before recruitment

Before inviting farmers, the partner should be able to confirm four things: it owns recruitment, it has named the escalation officer, it can support a two-week measurement period, and it understands the consent and deletion flow for voice questions.

Neuralis must bring equally concrete evidence. The reviewed content needs spoken Asante Twi from a farming-aware translator. Chemical blocks remain withheld until a Ghanaian agronomist verifies the missing safety details. WhatsApp must pass a real end-to-end message test. Staff must rehearse the complete flow without an unsafe answer.

Apollo 13 did not depend on a filter alone. It depended on a crew, a ground team, a known constraint, and a response that could be carried out under pressure. The AgriVoice briefing should produce the same kind of clarity: which questions the system can answer, which ones it must keep open, and whose name is attached to the next action.