A named extension officer turns an unclear Twi question into a case with an owner, a response path, and a record of what happened next. Before farmers enter the AgriVoice pilot, the cocoa-sector partner must name that person and agree to receive escalations.
In 1970, Apollo 13’s lunar module had a carbon-dioxide problem. The crew’s available round lithium-hydroxide canisters could not accept the square canisters from the command module, while the spacecraft was far from home. At NASA’s Manned Spacecraft Center in Houston, engineer Ed Smylie led the team assigned to make an adapter from materials already aboard the spacecraft.
The solution is remembered for duct tape and improvised parts. The more useful lesson is accountability. A dangerous problem had a named team, a defined task, and a way to test the answer before it reached the people depending on it. NASA documents the work in its Apollo 13 mission reporting.
An escalation needs somewhere to go
AgriVoice is designed to select from reviewed cocoa guidance in Asante Twi. It does not invent agronomy advice. When a farmer’s question is unclear, outside the reviewed material, or concerns a pesticide instruction that has not been cleared, the safe response is escalation.
That response only protects someone if it creates a real handoff.
An anonymous queue can collect unanswered questions for days. A farmer may assume that silence means the advice was unimportant, retry the question in different words, or act on an earlier answer that did not fit their situation. For cocoa farmers, the gap can matter when a disease symptom changes, a product label is incomplete, or the question concerns spraying and re-entry.
The pilot therefore needs a named extension officer before recruitment begins. The partner owns recruitment. The officer owns the human follow-up. Those are different responsibilities, and both need to be visible.
A failed match can still be a successful outcome
A voice system does not need to answer every question automatically to be useful. In this pilot, a failed match can be the right result when the system says, in effect, “This needs an extension officer.”
That matters most for pesticide questions. AgriVoice currently withholds three chemical blocks because dosage, re-entry, and pre-harvest intervals have not received Ghanaian agronomist approval. The system should refuse to fill those gaps with plausible-sounding words. A farmer deserves a safe answer, or a clear route to a person who can provide one.
The same rule applies when speech recognition hears uncertain audio, when a question contains more than one problem, or when the available reviewed blocks do not fit. The system should preserve the farmer’s question, identify why it escalated, and route it to the officer named by the partner.
That creates a practical distinction between “the app did not know” and “the farmer was abandoned.” The first can be measured and improved. The second is an operational failure.
For a closer look at why uncertain pesticide questions need human follow-up, see Ghanaian-language farmer advice: Why uncertain pesticide questions need human escalation.
The officer’s name makes the pilot measurable
A named owner gives the pilot a testable service standard. Neuralis plans to measure whether escalations have an accountable owner and whether the median response arrives in less than one working day.
That measure should include more than a timestamp. The pilot team needs to know whether the officer received the case, whether the farmer received a follow-up, and whether the answer was resolved, deferred, or referred elsewhere. If the queue repeatedly contains the same unanswered topic, that is evidence for a new reviewed content block or better translator guidance. If the officer cannot respond within the agreed window, the partner and pilot team can see the constraint instead of mistaking silence for demand.
This is where the human workflow improves the product. Escalations reveal the questions that reviewed content does not yet cover. They also show where Twi wording, voice recognition, or the selection process needs work. The farmer receives a safer path while the team gathers evidence for the next tightly scoped change.
Build the handoff before the first farmer message
Before the two-week pilot starts, the partner should identify one extension officer by name, confirm how that person receives cases, and agree on who covers an absence. Staff should then run the full path with a test question: voice message, uncertain match, escalation, officer response, and farmer follow-up.
Apollo 13’s adapter worked because Ed Smylie’s team could work on a defined problem with available materials and a route to the crew. AgriVoice needs the same discipline at a much smaller scale. When a cocoa question cannot safely match reviewed guidance, the farmer should encounter a person’s name beside the escalation, not an empty queue.
Comments
No comments yet.