Can Reviewed Asante Twi Voice Answers Safely Close Cocoa Farmers’ Information Gaps?

A cocoa-sector institution should sponsor one tightly measured AgriVoice pilot before considering wider deployment. The test would show whether reviewed Asante-Twi voice answers, backed by an accountable extension officer, can help farmers get useful information without exposing them to confident, unverified advice.

In April 1970, Apollo 13’s crew faced rising carbon dioxide inside the lunar module. The command module carried square lithium hydroxide canisters, while the lunar module used round openings. The crew had the filters they needed, but the parts did not fit.

NASA engineers in Houston had to devise an adapter from materials already aboard the spacecraft. Astronaut Jack Lousma then read the construction steps to the crew from Mission Control. Jim Lovell, Fred Haise and Jack Swigert followed those instructions in space before anyone could know with certainty that the improvised device would work. The episode is documented in Lovell and Jeffrey Kluger’s book Lost Moon.

The mechanism matters here: expert knowledge alone could not solve the problem. The answer had to reach the people facing it, through a channel they could use, in instructions they could follow, with humans accountable for what happened next.

The information gap is also a delivery gap

A cocoa farmer may need an answer while standing beside a diseased tree, examining a worn pesticide label or deciding whether to spend scarce cash. A technically correct document elsewhere does not close that gap. The guidance must arrive in language the farmer understands, in a form that works at that moment.

AgriVoice is being prepared around that practical requirement. A farmer asks a question by voice. Speech recognition turns it into text. The system selects from reviewed agronomy content, returns the answer in spoken Asante Twi and sends uncertain or unsupported questions to a named extension officer.

The language model does not compose agronomic advice. It selects approved content blocks. Pesticide guidance retains mandatory safety information, and questions without a verified answer must escalate to a person.

That distinction becomes critical when a chemical dose, re-entry interval or pre-harvest interval is unclear. AgriVoice currently withholds three chemical blocks because those details have not been cleared by a Ghanaian agronomist. Refusal is safer than completing a plausible sentence. The three withheld chemical blocks show exactly why the product needs expert review and human escalation before farmer exposure.

One pilot can answer the questions that matter

The proposed pilot is deliberately narrow: one cocoa-sector partner, one named extension officer, 20 to 50 Asante-Twi-speaking cocoa farmers and two weeks of observed use.

Before exposure, the spoken translations must be reviewed by a farming-aware translator. A Ghanaian agronomist must resolve or continue withholding the three chemical blocks. At least 20 real farmer questions must be recorded and transcribed to measure speech recognition. Consent, deletion behaviour, WhatsApp delivery and the full escalation path must also be tested.

The pilot would then measure outcomes that an institution can act on:

• At least 70 percent of questions should receive a suitable answer or a correct escalation.
• No unsafe pesticide instruction should reach a farmer.
• At least 80 percent of participants should report understanding the response.
• At least 30 percent of activated farmers should ask another question during the second week.
• Human escalations should have a named owner and a median response time below one working day.
• Automated responses should arrive in a median of under eight seconds, excluding human escalation.
• The cost of each completed question should be measured by speech, messaging, model, hosting and human-support component.

These thresholds create a continue, iterate or stop decision. A poor result would identify a specific failure, such as misunderstood speech, formal-sounding audio, weak repeat use or an escalation queue without an accountable owner. It would not justify hiding uncertainty behind a larger launch.

Sponsorship buys evidence, not promises

For a cocoa-sector institution, the immediate commitment is bounded. The partner recruits the cohort, names the extension officer and helps secure the farming-aware language and agronomy review needed for safe exposure. Neuralis supplies the voice workflow, controlled content selection, measurement and internal scorecard.

This structure addresses the concern behind calls from young cocoa farmers seeking clarity ahead of the 2026/2027 purchasing season. When a policy answer remains unverified, the correct workflow should mark it for clarification and route it to someone responsible. The same principle applies to pesticide advice: when the diagnosis is unclear, the system should prevent an unsupported instruction from sounding authoritative.

Apollo 13’s improvised adapter worked because the materials, instructions, communication channel and responsible people formed one operational chain. AgriVoice needs the same kind of proof at smaller stakes: reviewed knowledge must survive speech recognition, selection, Twi audio delivery and escalation to reach a farmer safely.

The concrete next step is to appoint one pilot owner and one extension officer, then clear the translation, agronomy, farmer-speech and WhatsApp gates. Only after two weeks of measured use should anyone decide whether AgriVoice deserves another iteration or a wider deployment.