The Extension Officer AgriVoice Needs, and What an Escalation Could Otherwise Cost

A two-week AgriVoice pilot can show whether Ghanaian-language voice guidance helps cocoa farmers safely, provided one cocoa-sector institution recruits 20 to 50 farmers, names an extension officer, and reviews the results with us. We are inviting one institution to co-run that pilot and make a measured decision from the evidence.

In 1970, Apollo 13’s crew faced rising carbon dioxide inside a damaged spacecraft. At Mission Control in Houston, NASA engineers had to make a square command-module filter fit a round lunar-module opening using materials already on board. The outcome was uncertain while the crew depended on a fix that had to work under real constraints, not in a lab. James Lovell and Jeffrey Kluger document the episode in Lost Moon.

A useful answer needs a safe path when it is uncertain

AgriVoice is built for a similarly practical constraint: a farmer may ask a question in Asante Twi by voice, often from a field, and the answer must be useful without pretending to know more than it does.

The system does not generate agronomy advice from scratch. It selects from reviewed guidance blocks, speaks the selected answer in Twi, and routes missing, unclear, or pesticide-related questions to a person. Three chemical guidance blocks remain withheld because dosage, re-entry, and pre-harvest information has not yet been verified by a Ghanaian agronomist. That refusal is part of the product’s safety boundary.

A named extension officer makes that boundary real. When AgriVoice cannot safely continue, a farmer needs an accountable person who can receive the escalation and respond. The named extension officer AgriVoice needs explains why that role cannot be left vague.

We are seeking one cocoa-sector institution that can own farmer recruitment and appoint that officer for a two-week pilot. The institution might be a cocoa cooperative, an extension program, a buying company with farmer-support operations, or another organization already trusted by a defined farmer group.

The pilot asks a narrow question with visible measures

This is a learning pilot, not a public launch. We will work with 20 to 50 Asante-Twi-speaking cocoa farmers over two weeks and measure the parts that determine whether the service deserves another iteration.

The shared scorecard will track:

• Whether a question receives a reviewed answer or is correctly escalated. The continue signal is at least 70 percent.
• Whether any unsafe pesticide instruction reaches a farmer. The acceptable number is zero.
• Whether farmers report understanding the spoken response. The target is at least 80 percent.
• Whether activated farmers return with another question in week two. The target is at least 30 percent.
• Whether the extension escalation path has an owner and a median response time below one working day.
• Whether median end-to-end response time stays below eight seconds, excluding human escalation.
• What each completed question costs across WhatsApp, speech, language-model, hosting, and human-support components.

These measures make room for an honest outcome. If farmers cannot understand the voice, that is evidence. If latency breaks the conversation, that is evidence. If a question is escalated correctly, that is a successful safety result, not a failed answer.

Digital financial inclusion has been associated with climate-smart practice adoption among Ghanaian cocoa farmers. Voice guidance may help make practical information more reachable, but the pilot must establish whether this particular workflow earns trust and understanding in use.

What the partner brings, and what Neuralis brings

The partner brings the part software cannot supply: a real farmer cohort, local operating context, and a named person who can take responsibility for escalations. We ask the partner to help recruit participants, introduce the pilot clearly, and review the scorecard with us at the end of the two weeks.

Neuralis brings the AgriVoice workflow, consent and deletion testing, reviewed content selection, Twi speech delivery, pilot measurement, and a staff safety rehearsal before farmers receive access. We will also collect real farmer questions, with consent, to measure speech recognition against field conditions rather than relying only on read speech.

The product has limits worth stating plainly. The current Twi voice was trained from scripture recordings and can sound formal. The system is preparing content review and field evaluation before exposure. Pesticide questions with uncertain guidance must reach the extension officer instead of receiving a confident guess. When a farmer needs an answer that cannot safely continue, escalation is the service doing its job.

Make the decision together after two weeks

Apollo 13 did not depend on a polished demonstration. It depended on a constrained solution, clear roles, and people checking whether the solution worked before relying on it. AgriVoice needs the same discipline at a very different scale.

At the end of the pilot, the institution and Neuralis will review the scorecard and choose one of three paths: continue, make one tightly scoped improvement, or stop. A weak result will not be used to justify expanding to another language or another product area.

If your institution works directly with cocoa farmers and can support a small, measured trial, the first practical step is to identify the farmer group and the extension officer who would receive escalations. That gives the pilot its most important operating condition before the first voice note arrives.