Pesticide label guidance: When Kojo’s formulation required extension review

A pesticide question can change after the product label is read aloud because the formulation may fall outside the reviewed guidance. When that happens, AgriVoice should stop short of instructions and send the question to a named extension officer.

At 6:40 in the morning, Kojo is standing beside a cocoa plot with a phone in one hand and a partly mud-streaked container in the other. The leaves nearest the path show fresh damage. He has already asked, in Twi, what he should spray and when.

A general answer might sound possible at first. The question names a familiar problem. But then Kojo reads the small print on the label into a voice note: the product name, the formulation, and the concentration. The container in his hand does not match the reviewed block that covers the broader question.

That difference matters. If he uses the wrong mixing guidance, sprays at the wrong interval, or goes back into the field too soon, the harm is no longer theoretical. His crop, his health, and the people working near the plot are all on the line. The safer answer is not a confident guess in clear Twi. It is a pause.

A label can turn a broad question into a specific safety decision

Farmers often start with the question that feels most urgent: “What should I use?” or “Can I spray this now?” That is a reasonable place to begin. It is also incomplete when the answer depends on the exact product in front of them.

A pesticide label can contain details that change the advice: the active ingredient, the formulation, the dilution instructions, the crop stage, and warnings about when people can re-enter a treated area or harvest. A product name alone may not carry enough information. Similar-looking containers can require different handling.

For AgriVoice, the important boundary is simple. Its language model can select from reviewed advice blocks. It does not create agronomy instructions from scratch. If the farmer’s voice note reveals a formulation that the reviewed library does not cover, the system must say so plainly and escalate the case.

That protects the farmer from an answer that sounds useful while resting on missing facts. It also protects the extension officer from inheriting a mistake after it has already been spoken aloud as advice.

The safe turn happens when the system knows what it does not know

Kojo’s first voice note may be answered with safe, reviewed guidance about observing symptoms, avoiding a rushed purchase, or preparing the information an extension officer will need. Then he reads the label.

The answer changes. AgriVoice should not try to force the label into the nearest matching product category. It should record the relevant details, explain in Twi that the reviewed guidance does not cover that formulation, and route the question to the named extension officer.

This is where a voice workflow earns trust. A farmer should not need to know the technical difference between two formulations before asking for help. The system needs to recognize when that difference changes the safety of its response.

The escalation also needs an owner. “Ask an expert” leaves Kojo holding the same container with no clear next step. A named extension officer and a working queue turn uncertainty into an operational handoff. That distinction matters most when the farmer is already at the field, ready to act.

For a closer look at why a pesticide answer sometimes has to stop, see What Happens When a Farmer Needs an Answer That Cannot Safely Continue?.

Reviewed advice needs room for uncertainty

It can feel frustrating to withhold an answer when a farmer asks a direct question. Yet pesticide safety depends on details that a general library cannot responsibly fill in. Dosage, re-entry guidance, and pre-harvest intervals should come from verified content for that exact situation, or from an accountable human reviewer.

That is why AgriVoice’s pilot design treats escalation as a successful outcome when it is correct. The goal is not to answer every voice note automatically. The goal is to give a useful, safe response or get the question to the person who can safely resolve it.

This approach also makes the pilot measurable. The team can examine whether the label was captured clearly, whether the system recognized the missing coverage, whether the officer received enough context, and how long the response took. A confident wrong answer would hide those failures until they reach a farm.

Ghana’s current agricultural investment focus on cocoa depends on collaboration among farmer organizations, institutions, agribusinesses, and government. A local-language voice service fits that work only if its escalation path connects to people with real responsibility for the advice.

The next voice note should carry the details that change the answer

Later that day, Kojo does not receive a made-up mixing instruction. He receives a clear message that his product needs extension review, along with a request to keep the container available and share the label details again if needed.

The immediate problem has not vanished. The spray decision is still waiting. But the uncertainty is visible now, and it has somewhere to go.

That is a better outcome than a smooth-sounding answer built on the wrong product. When the label changes the question, the response should change with it.