AI-counted pink-bollworm traps and threshold-based spray advisories
Farmers can spray against a shared trap-count threshold instead of a calendar or subjective visual assessment.
Agriculture · Karnataka, Maharashtra, and later multi-state India
Executive brief
The operating-model shift, in one view.
This case is valuable because the larger later test was null: AI advice creates value only when the targeted risk is present and the human sensing routine is sustainable.
AI value · Farmer income in initial field evaluations
Verified
Up to 22% higher income in the first-year evaluations; no significant benefit in the 2021-2022 multi-state experiment
The mixed evidence is essential: early evaluations were not all randomized, and unusually high rainfall with low pest pressure eliminated significant benefit in the later multi-state experiment.
Before
Inspect fields and decide pesticide timing from experience, visible damage, or calendar practice. → Apply pesticide, sometimes before or after the economically effective pest threshold.
After
Checks pheromone traps weekly and uploads smartphone images. → Detects and counts pink bollworm moths and compares counts with an action threshold. → Receive a localized spray advisory and decide whether and how to apply treatment.
Human boundary
The model determines whether the observed count crosses an advisory threshold; farmers retain the pesticide application decision.
Why it matters
Farmers can spray against a shared trap-count threshold instead of a calendar or subjective visual assessment.
How the work changed
Before
How the work ran before the change.
Step 1 of 2
Cotton farmer
Inspect fields and decide pesticide timing from experience, visible damage, or calendar practice.
ControlFarmer judgment
Step 2 of 2
Farmer
Apply pesticide, sometimes before or after the economically effective pest threshold.
ControlAvailable pesticide and labor
What changed
Farmers can spray against a shared trap-count threshold instead of a calendar or subjective visual assessment.
Decision rightHuman sets the threshold; AI decides each instance
After
How the same work runs now.
Step 1 of 3
Lead farmer
Checks pheromone traps weekly and uploads smartphone images.
ControlManual trap placement and image capture
Step 2 of 3
CottonAce
Detects and counts pink bollworm moths and compares counts with an action threshold.
ControlIntegrated-pest-management threshold
Step 3 of 3
Farmers
Receive a localized spray advisory and decide whether and how to apply treatment.
ControlFarmer retains application authority
Process model built from the published workflow evidence for Wadhwani AI CottonAce partner programs. Every step, actor, and control appears in full below.Every step, actor, and control
Exception path
Farmers and extension partners can withhold treatment, inspect manually, or use other integrated-pest-management methods; rainfall and low pest pressure can make the advisory produce no measurable benefit.
Decision authority
The model determines whether the observed count crosses an advisory threshold; farmers retain the pesticide application decision.
Before
#
Actor
Action
Control
01
Cotton farmer
Inspect fields and decide pesticide timing from experience, visible damage, or calendar practice.
Farmer judgment
02
Farmer
Apply pesticide, sometimes before or after the economically effective pest threshold.
Available pesticide and labor
After
#
Actor
Action
Control
01
Lead farmer
Checks pheromone traps weekly and uploads smartphone images.
Manual trap placement and image capture
02
CottonAce
Detects and counts pink bollworm moths and compares counts with an action threshold.
Integrated-pest-management threshold
03
Farmers
Receive a localized spray advisory and decide whether and how to apply treatment.
Farmer retains application authority
Work that left the path
Subjective manual moth counting for every participating farmer
Some prophylactic or mistimed pesticide applications
Human role before
Farmers individually diagnosed pest pressure and timed spraying.
Human role after
A lead farmer supplies trap images; the model standardizes counting; farmers act on threshold advisories and still perform field work.
AI role
Computer-vision detection and counting of moths in pheromone-trap images used to trigger localized pest advisories.
Outcomes
Farmer income in initial field evaluations
Verified
Non-adopter or control farmers in 2020-2021 evaluation locations→Up to 22% higher income in the first-year evaluations; no significant benefit in the 2021-2022 multi-state experiment
2020-2022 · Initial evaluations in Ranebennur and Wardha followed by a seven-state experiment
The mixed evidence is essential: early evaluations were not all randomized, and unusually high rainfall with low pest pressure eliminated significant benefit in the later multi-state experiment.
What leaders can reuse
Anti-pattern
Promoting the 22% first-year result without the later null result or ignoring the labor required to inspect traps.
Questions
01What external condition determines whether the intervention can help?
02Is the human data-collection burden sustainable?
Portability conditions
Reliable local image capture
Validated economic action thresholds
Weather- and pest-aware evaluation
Reputation risk
medium
Evidence and authority
What the public record supports.
Watch · updated
Watch status: verify the cited source and deployment condition before reusing this case.
1 independent, 1 peer reviewed; publication outcomes are verified.
Bundle 1.0.0 · reviewed 2026-09-06 · stable ID 968b26cd2d0b5f01
